How Grok Bot designers use AI agents to build personal sites and product prototypes

Brought to you by:

• WorkOS—Make your app enterprise-ready, with SSO, SCIM, RBAC, and more

• Vanta—Automate compliance and simplify security

John Bai and Peng Zheng are designers on the Grok Bot team at SpaceXAI. In this episode, Peng shows how he built a self-updating personal site with Grok Bot handling the entire backend, from looking up a location to generating custom 3D artwork and publishing each new check-in. John demonstrates how he uses voice memos, Grok Bot, and Figma MCP to complete design work from the gym and turn shower thoughts into working prototypes. They offer a fascinating look at how AI is eliminating the tedious parts of design and making it faster and easier to explore new ideas.

Biggest takeaways:

1. The best personal website is one that keeps updating itself. Peng did not want to build another portfolio that launches, gets a few seconds of attention, and immediately becomes outdated. Instead, he created a Grok Bot workflow that turns a photo or place name into a published check-in. The bot finds the coordinates through Google Places, generates light and dark 3D miniature versions of the image, and adds everything to the site. Peng sends a single message, then walks away.

2. People no longer need to know exactly what they are building before they begin. Peng started without a Figma file, a specification, or a roadmap. He described the initial idea to Grok Bot, built something, reacted to it, and followed the most interesting possibilities. The final check-in workflow was not planned up front. It emerged through a series of “What if it could also do this?” moments. Because the bot handled the architecture, Peng could explore the idea without defining the entire system first.

3. The best AI workflows help people leave their computers. John’s Figma Bro workflow is a perfect example. When a colleague requested community assets while John was at the gym, he took two screenshots in Figma, recorded a voice memo explaining what he needed, and went back to his workout. By the time he checked again, the bot had produced three options, adjusted their sizing and colors, and completed the task.

4. Voice memos are becoming a powerful design interface. John gives Figma Bro natural-language instructions by voice, and the bot translates them into actions inside Figma through an MCP connection. He can describe an idea casually, even somewhat messily, and let the system interpret and execute it. The key is configuring the bot’s preferences ahead of time, including artboard structure, spacing, and naming conventions. That foundation allows informal instructions to produce organized files.

5. AI is making previously inaccessible creative skills available to more people. Claire could not illustrate. John wanted to create motion he could only describe with his hands. Peng wanted to produce 3D renders. In the past, each of those gaps would have required a specialist or years of practice. Now they can begin with a prompt. As Peng points out, the cost of making things has fallen so dramatically that unusual combinations and experiments are finally worth trying.

6. The most useful bots have narrow, clearly defined responsibilities. Peng organizes his agents into life and work categories, with separate bots for email triage, calendar management, 3D-printing inventory, green-card status tracking, and other specific jobs. A chief-of-staff bot handles anything that does not have an obvious owner. Each agent knows its lane, which makes delegation feel like handing work to a capable specialist instead of transferring confusion to a general-purpose chatbot.

If you’re enjoying these episodes, reply and let me know what you’d love to learn more about: AI workflows, hiring, growth, product strategy—anything.

Catch you next week,
Lenny

P.S. Want every new episode delivered the moment it drops? Hit “Follow” on your favorite podcast app.

John Bai and Peng Zheng are designers on the Grok Bot team at SpaceXAI, where they’re building one of the most talked-about AI products right now. John writes publicly about his design process (his piece “Designing Grok Bot with Grok Bot” has already made the rounds) and shares bot templates with the design community. Peng brings a product-design sensibility to personal tools, and his website doubles as a live demo of what he builds.

What you’ll learn:

1. How Peng built a self-updating personal website using Grok Bot as the entire backend pipeline, with no CMS and no Figma file

2. The exact check-in bot setup that lets Peng send a photo or a place name and have his portfolio update itself automatically

3. How John’s Figma Bro bot handles production design tasks while he’s at the gym

4. How John uses voice memos to direct Figma work through an MCP connection without opening his laptop

5. The “shower thought to prototype” workflow John uses with DevBot to test interaction ideas without first going through a product manager or engineer

6. The “trash can method” of software development

7. How both designers organize their personal bot ecosystems

8. What John and Peng actually think AI means for the future of design as a craft

Brought to you by:

WorkOS—Make your app enterprise-ready, with SSO, SCIM, RBAC, and more

Vanta—Automate compliance and simplify security

In this episode, we cover:

(00:00) Introducing John and Peng

(02:53) The Grok Bot hype train

(04:35) Peng’s self-updating personal website built with Grok Bot

(15:12) How AI makes design more accessible

(19:35) Website update result

(20:13) John’s Figma Bro bot

(23:35) Creating marketing materials for the bot marketplace

(26:00) DevBot: from shower thoughts to working prototypes

(28:48) The trash can method of software development

(31:19) Other bots John and Peng are using

(39:05) Practical tips for when bots don’t do what you want

Tools referenced:

• Grok Bot (xAI): https://x.ai/bot

• Figma: https://www.figma.com

• Figma MCP server: https://www.figma.com/mcp-catalog/

• Google Places API: https://developers.google.com/maps/documentation/places/web-service

• Notion: https://www.notion.so

• Swarm (Foursquare): https://www.swarmapp.com

Other references:

• Designing Grok Bot with Grok Bot: https://x.ai/bot/guides/designing-grok-bot-with-grok-bot

• Figma Bro bot template (shared by John Bai): https://x.ai/bot/marketplace/bots/figma-bro

• From zero coding background to hardware hacker: How Cursor + a Raspberry Pi makes AI fun: https://www.lennysnewsletter.com/p/from-zero-coding-background-to-hardware?utm_source=publication-search

Where to find John and Peng:

John Bai on X: https://x.com/johnbai

Peng Zheng on X: https://x.com/pengzheng_

Where to find Claire Vo:

ChatPRD: https://www.chatprd.ai/

Website: https://clairevo.com/

LinkedIn: https://www.linkedin.com/in/clairevo/

X: https://x.com/clairevo

Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.

Grok Bot vs. OpenClaw: why I replaced my entire agent stack

Brought to you by:

• WorkOS—Make your app enterprise-ready, with SSO, SCIM, RBAC, and more

• Hyperagent—Deploy fleets of agents that handle real work

In this solo episode, Claire explains why she replaced her entire OpenClaw setup with Grok Bot. She walks through the bots she now uses to manage six inboxes, coordinate her family, review pull requests, monitor SOC 2 compliance, support customers, audit subscriptions, and shop for clothes. She also shares how she safely gives agents permission to take action, what Grok Bot still gets wrong, and the migration process she used to move every agent without losing its personality, context, or routines.

Biggest takeaways:

1. The real moat for agent platforms may be user experience, and Grok Bot wins on simplicity. Claire is highly technical. She runs Tailscale, connects to a Mac Mini over SSH, and even built a rescue bot to manage her other bots. Still, OpenClaw eventually became too difficult to maintain. Grok Bot offers many of the same capabilities in a product that reliably stays online, which was enough to convince her to migrate everything.

2. Thinking of each bot as a new hire makes agent design much easier. Claire gives every Grok Bot a specific name, job, and scope, just as she would when hiring someone for a team. Grok Bot can even infer an agent’s role from its name. Create one called “Prody McProd,” and it already has a sense of what it should do. This framing makes it easier to build focused specialists instead of agents that gradually become unfocused generalists.

3. Multi-account support is a killer feature for entrepreneurs running multiple businesses. Grok Bot allows Claire to connect six Gmail accounts alongside multiple Slack and Linear workspaces, then gives a single agent access across all of them. Other platforms are only beginning to support this kind of setup. For someone operating across several companies or projects, it makes the difference between an agent that is merely interesting and one that is actually useful.

4. Agents built for the home should help people spend less time on their computers. Claire’s guiding principle for TradBot is simple: if she has to spend time interacting with it, something has gone wrong. Its kitchen-table newspaper succeeds because the output is printed, physical, and waiting beside her family at breakfast. Sending the same personalized digest to a Kindle offers another way to bring the information into daily life without adding another screen-bound workflow.

5. Approval gates offer the right balance between autonomy and trust for high-stakes actions. Claire’s customer support agent, Holly Helpdesk, can handle refunds, but it cannot issue one without her approval. When a refund is needed, Claire receives a button that opens the actual Stripe request. She reviews it, confirms it, and the refund goes through. Within a week of launching, Holly began receiving unsolicited five-star reviews. The lesson is not that agents need approval for everything. It is that they should act independently on low-risk work and involve a human when the consequences are meaningful.

6. Grok’s default writing style is too compressed, so it needs explicit training on the user’s voice. While Claude often writes too much, Grok Bot has the opposite problem. It trims sentences so aggressively that its writing can become fragmented and difficult to follow. Claire fixed this by having it study her sent emails and create a detailed voice profile before writing on her behalf. With that context in place, it became capable of handling lower-stakes outbound communication.

7. A good compliance agent should fix problems, not just report them. Claire’s SOC 2 agent, Lockdown, logs into the controls dashboard, scans for issues, investigates new vulnerabilities, and opens a pull request with a proposed fix before notifying her. Claire reviews and approves the change, but the agent handles the underlying work. The broader opportunity is to have AI actively improve a company’s compliance posture, not simply generate another list of tasks for someone else to complete.

8. A structured migration process makes switching agent platforms feel safe and reversible. Before Claire shut down any of her OpenClaw agents, Lifeguard exported a secrets-free archive containing each agent’s identity, scheduled tasks, and gateway configuration. She uploaded those folders into Grok Bot, which effectively performed a brain transplant. The agents returned with their personalities and responsibilities intact. Because the migration preserved everything she needed to go back, switching platforms felt far less risky.

9. Agent proactivity must be intentionally scheduled. OpenClaw’s heartbeat feature made its agents feel naturally proactive, while Grok Bot remains passive unless the user creates routines. The benefit is transparency: every recurring action is visible and easy to audit in the settings. The tradeoff is that users must consciously decide when each agent should act. Without those routines, even a highly capable agent will simply wait for someone to message it.

Build your own company brain: the enterprise AI playbook from Stripe’s engineering team | Sharadh Krishnamurthy

Brought to you by:

• DX—Engineering intelligence for the AI era

• Hyperagent—Deploy fleets of agents that handle real work

Sharadh Krishnamurthy is an engineering manager at Stripe who helped build Kai, the company’s internal AI agent, in just two weeks with a team of 1.5 engineers. In this episode, he explains how Kai grew to serve more than 10,000 employees each week, why scaling AI across a company is ultimately a governance problem, and how Stripe uses projects, skills, tool policies, and human oversight to keep agents useful and safe. He also shares what happened when agents nearly overwhelmed production systems, how Stripe manages a library of more than 2,000 skills, and why strong data and developer infrastructure matter far more than simply choosing the best model.

Biggest takeaways:

1. The hardest part of bringing AI into a large company isn’t the model—it’s governance. When Sharadh and a colleague began building Kai, they quickly realized that giving every employee access to AI was not primarily a technical challenge. The real work was teaching the system how a complex global company operates. That is why Kai’s most distinctive features, including projects, tool policies, and skill routing, are designed to manage access and behavior rather than improve the underlying model.

2. Projects are configuration layers for how teams use AI, not simply folders for organizing work. In Kai, a small group of knowledgeable employees can choose the default model, approve the available tools, and define which actions require human review for an entire team or workflow. Everyone else gets the right setup automatically. Stripe’s HR project, for example, runs on a separate secure backend with tightly controlled tool access, preventing sensitive employee data from leaking into public documents.

3. Infrastructure built before the AI boom becomes even more valuable once agents arrive. Stripe had already invested in a tiered analytics catalog, a resilient Trino query engine, and a trusted metrics layer for its human developers. That same foundation now allows agents to query data safely and reliably at enormous scale. Sharadh’s provocative advice is to double the size of the developer experience and data teams before investing heavily in agents. Without strong infrastructure underneath them, even the smartest agents will struggle.

4. Agents amplify every weakness that already exists in a company’s systems. Sharadh does not believe agents create entirely new categories of problems. Instead, they dramatically increase the volume and severity of existing ones. Stripe saw agents behave unpredictably, overwhelm core systems, and nearly take down production. Better prompts were not enough to solve those problems. Stripe needed agent-specific identities, priority signals, load shedding, and a query layer capable of absorbing the additional demand.

5. Skills are what turn an AI assistant into dependable infrastructure. Kai allows any employee to convert a successful session into a reusable skill. Each skill includes guidance on when it should run, a specification written for the Kai system, and an editor for testing it before release. Stripe now has roughly 2,000 skills, including about 50 used daily across the company. Because Kai automatically selects the right skill, employees can simply describe what they need instead of learning how the system works.

6. With skills, quantity is part of the quality problem. Every new skill adds more information to the system’s retrieval pool, and irrelevant context can reduce the quality of the results. Stripe built telemetry to distinguish between skills used across the company, highly valuable workflows used by small teams, and tools that have not been touched in weeks. That data helps the team promote, combine, or retire skills while preserving useful niche workflows.

7. A small team can build an enterprise AI platform quickly when the underlying foundation is strong. Sharadh and one colleague created the first working version of Kai in just two weeks. A company-wide demonstration triggered a steep rise in adoption, and the platform is now used weekly by more than 10,000 Stripe employees, with an 86% adoption rate. Fewer than 10 people manage it today. That scale is possible because the team is supported by coding agents and years of investment in Stripe’s internal infrastructure.

8. The best enterprise AI products hide organizational complexity from the people using them. Kai appears to employees as a single, straightforward product, but it is built from many configurable layers designed to accommodate different teams, data types, and security requirements. Employees do not have to decide which connector to enable, which model to select, or whether a tool is appropriate for the information they are handling. Removing those decisions improves adoption while reducing the temptation to bypass safeguards because the approved path feels too slow.

GPT-6 Astra is a banger—here’s everything I’ve built

In this solo episode, Claire puts GPT-6 Astra through a series of real-world tests across computer use, coding, design, QA, hardware, and 3D development. She shares how Astra finally solved a ChatPRD feature she had been stuck on for six months, navigated complex production tools without her touching the keyboard, and reverse-engineered a Bluetooth display with no public API. She also explores what the model’s leap in capability means for software interfaces, ambitious personal apps, and the kinds of projects that suddenly feel possible to build.

Biggest takeaways:

1. The biggest leap with Astra is that it expands what someone is willing to attempt. Claire had spent six months trying to build a product intelligence feature for ChatPRD, but earlier models consistently made poor architectural decisions because she could not fully explain the system in a prompt. Astra got the feature 90% of the way there on its first attempt, then finished it with a few follow-ups. When previously impossible projects start working, the real change is not just productivity; it’s ambition.

2. Computer use is finally precise enough to handle complex visual interfaces. Claire now spends roughly 90% of her day with her hands off the keyboard. Astra navigated Adio’s node-based workflow builder, added new nodes, reconnected the logic, and generated custom lead-response emails without requiring her to click anything. She had abandoned similar attempts with earlier models because they lacked the necessary precision. That gap appears to be closing quickly.

3. Browser-based QA may be one of the most valuable uses of computer-use agents. Astra spent an hour and 45 minutes testing a branch for race conditions and streaming edge cases. It inspected the console, deliberately refreshed the application, and continued probing for failures. This is exactly the kind of repetitive but important testing that humans tend to rush through or skip. An agent’s willingness to stay patient with tedious work can be as valuable as its intelligence.

4. SaaS is back, and the case for eliminating interfaces is getting weaker. The “no UI is the next UI” argument assumed that AI would need direct access through APIs or command-line tools. But if an agent can reliably navigate visual software, product teams can choose the interface that best expresses the product rather than optimizing entirely around machine access. In other words, buttons are back.

5. Hardware hacking is becoming a serious use case for frontier models. Claire had spent months trying to work with the Divoom Mini 2, a device with no public API, without making much progress. Astra reverse-engineered its pixel algorithm, built a live-streaming command-line tool, and enabled her to send episode information and images directly to the display. That is more than a clever demo. It suggests AI can unlock physical devices that were previously too difficult or poorly documented to customize.

6. One-shot 3D game development is raising the ceiling for personal software. Astra built a functional Barbie-style dress-up game in Blender from a single prompt. It also created a 3D world for Claire’s family app where her kids can walk to a library or visit a pond to check in with their emotions. These are not projects she could have built with earlier models. As the cost of creating interactive worlds falls, deeply personal software starts to feel much more practical.

7. Astra’s AIM-style Codex wrapper shows how accessible ambitious desktop development has become. In one shot, Astra built a functional 1990s-inspired AIM interface around Claire’s live Codex threads, complete with active-writer indicators and away messages. The nostalgia is fun, but the bigger lesson is that custom desktop tools and unusual personal software ideas are now within reach, even when they would never justify a traditional development team.

8. Astra produces artifacts that fit into an existing brand instead of starting over. When Claire needed thumbnails while her designer was unavailable, Astra worked inside Flora and Figma, studied the existing workflows, and matched the established aesthetic. That ability to understand and extend an existing visual system is what makes the output useful in production. A polished artifact is impressive, but one that actually looks like it belongs is far more valuable.

9. Astra’s pricing makes choosing the right tasks more important than minimizing usage. At $10 per million input tokens and $50 per million output tokens, long-running computer-use and QA sessions can become expensive, particularly on the output side. The key is to deploy Astra where its additional capability is worth the premium: stubborn coding problems, complex visual interfaces, and high-toil QA work. The question is not whether the model is expensive but whether the task is valuable enough to justify using it.

If you’re enjoying these episodes, reply and let me know what you’d love to learn more about: AI workflows, hiring, growth, product strategy—anything.

Catch you next week,
Lenny

P.S. Want every new episode delivered the moment it drops? Hit “Follow” on your favorite podcast app.

Sharadh Krishnamurthy is an engineering manager at Stripe, where he helped build Kai, the company’s internal AI agent used by more than 10,000 employees every week. He’s worked across several of Stripe’s core infrastructure teams, including data and developer experience, which gives him a grounded, systems-level perspective on what it actually takes to make AI work at enterprise scale. He’s currently focused on the governance, skills, and infrastructure layers that let every Stripe employee use AI safely and effectively, regardless of their technical background.

What you’ll learn:

1. Why Stripe built Kai from scratch instead of buying, and what tipped the decision

2. What Kai knows about you by default and what you actually control

3. Why “projects” at Stripe are a governance mechanism, not just a folder

4. How Stripe structured its data layer so agents can query safely at scale

5. Why the infrastructure Stripe built for human developers turned out to be exactly what agents needed

6. How Kai’s skills platform lets any employee package a workflow, and what happens when you have 2,000 of them

7. What Sharadh learned the hard way when agents nearly took down production systems

Brought to you by:

DX—Engineering intelligence for the AI era

Hyperagent—Deploy fleets of agents that handle real work

In this episode, we cover:

(00:00) Introducing Sharadh

(02:46) Why Stripe built an AI agent (Kai) instead of buying tools

(05:18) What Kai knows about you (and what you can turn off)

(06:51) Projects as a governance layer

(10:04) Live demo: Kai builds a dashboard

(12:18) Tools, skills, and the secure sandbox

(17:22) Why Stripe has benefited so much from AI

(19:20) Agentic identity, load shedding, and rogue agents

(20:41) Iterating on the dashboard

(25:01) How they rolled out Kai across the team

(29:07) How projects work

(34:18) Bespoke agents for bespoke use cases

(35:58) The skill builder workflow

(40:40) Skill quality, evals, and telemetry

(43:01) Recap

(45:13) Lightning round

Tools referenced:

• Trino: https://trino.io/

• Anthropic: https://www.anthropic.com/

• Gemini: https://gemini.google.com/

• Cursor: https://www.cursor.com/

Where to find Sharadh Krishnamurthy:

LinkedIn: https://www.linkedin.com/in/sharadhk

Where to find Claire Vo:

ChatPRD: https://www.chatprd.ai/

Website: https://clairevo.com/

LinkedIn: https://www.linkedin.com/in/clairevo/

X: https://x.com/clairevo

Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.

I got early access to GPT-6 Astra: when I say this model broke through tasks I couldn’t crack with 5.6 Sol or Fable, I mean it specifically: the ChatPRD product intelligence feature, building 3D games, the hardware hack, and a handful of one-shot coding projects I’d tried and failed on repeatedly.

What you’ll learn:

1. Why Astra’s computer use feels different, and which production tools I’m trusting it with

2. The one feature I’d thrown every model at for six months, and what finally got it to 90%

3. How I’m using browser use for QA, not building, and what it found that I would’ve missed

4. Why I think UI is genuinely back, and what that means for SaaS and MCPs

5. The hardware hack I’d been chasing since GPT-5.5, and how Astra finally cracked it

6. What Astra built me in Blender in one shot, and why 3D is my new capability benchmark

7. The AIM-style Mac app Astra made in one shot, and what it signals about desktop development now

8. An honest take on speed, cost, and whether Astra is worth making your daily driver

In this episode, we cover:

(00:00) GPT-6 Astra overview

(03:44) Browser/computer use test on my CRM

(09:08) Flora thumbnail generation

(13:00) Browser use for QA

(15:20) Coding: ChatPRD product intelligence feature, finally one-shotted

(18:36) Hardware hack: Divoom MiniToo CLI and live streaming display

(22:23) Building an AIM-style Mac app

(24:24) Blender and 3D assets: Barbie Bench and the kids’ family app

(28:52) Summary: what Astra is great at and what to try first

Tools referenced:

• GPT-6 Astra: https://openai.com/index/gpt-6-astra/

• Codex: https://openai.com/codex

• Flora (node-based AI image/video editing): https://flora.ai/

• Figma: https://www.figma.com

• Blender: https://www.blender.org

• GPT Image 2: https://developers.openai.com/api/docs/models/gpt-image-2

• Divoom MiniToo: https://divoom.com/products/minitoo

• cxo.dev: https://www.cxo.dev/

Where to find Claire Vo:

ChatPRD: https://www.chatprd.ai/

Website: https://clairevo.com/

LinkedIn: https://www.linkedin.com/in/clairevo/

X: https://x.com/clairevo

Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.

I’m running about 30 active agents at any given moment, and in this episode I break down my full Grok Bot setup: what it is, how it compares to OpenClaw, and the nine bots I’ve built for work and my personal life. We go deep on Chief (my chief-of-staff bot sweeping six inboxes and multiple Slack workspaces), TradBot (the family agent that prints a kitchen-table newspaper for my kids), two engineering bots handling my PR queue and SOC 2 compliance monitoring, Holly Helpdesk, and a handful of personal bots I didn’t expect to actually love. I also walk through how I migrated everything from OpenClaw, including the script I used to export and transplant each agent’s identity and schedule.

What you’ll learn:

1. The three primitives Grok Bot is built on, and why one of them changes what agents can actually do

2. How Chief, my general-purpose chief of staff, handles a scope I didn’t think a single bot could manage

3. The writing quirk I noticed immediately with the Grok model, and what I did about it before letting it near my inbox

4. Why I created a family agent, what it produces every morning, and the design principle I used that has nothing to do with a screen

5. The two engineering bots doing work I used to do myself, and how one of them handles compliance in a way that surprised me

6. How Holly Helpdesk started getting five-star reviews from customers who had no idea they were talking to a bot

7. The personal bots I built mostly on a whim, and the one I now look forward to every Monday morning

Brought to you by:

WorkOS—Make your app enterprise-ready, with SSO, SCIM, RBAC, and more

Hyperagent—Deploy fleets of agents that handle real work

In this episode, we cover:

(00:00) Why I migrated from OpenClaw to Grok Bot

(02:10) Grok Bot overview: the three core primitives

(07:41) Chief: my chief-of-staff bot

(11:51) OpenClaw vs. Grok Bot

(12:41) TradBot: my family agent

(19:51) LGTM the PR Closer

(22:03) Lockdown: SOC 2 control monitoring bot

(24:00) Holly Helpdesk: customer support

(26:58) Penny Pincher: subscription audit, insurance negotiation, Rolex shopping

(29:37) ShopZilla and Sylvie Style: personal shopping and wardrobe bots

(32:54) How to migrate your OpenClaws

(34:26) Final take

Tools referenced:

• Grok Bot (SpaceXAI multi-agent platform): https://x.ai/news/introducing-grok-bot

• OpenClaw (previous agent platform): https://openclaw.ai/

Where to find Claire Vo:

ChatPRD: https://www.chatprd.ai/

Website: https://clairevo.com/

LinkedIn: https://www.linkedin.com/in/clairevo/

X: https://x.com/clairevo

Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.

How I turned Claude into a self-improving PM assistant | Daniel Blum (PM, Melio)

Brought to you by:

• Optimizely—Your AI agent orchestration platform for marketing and digital teams

• Jira AI SDLC—Get your tokens’ worth with Jira

Daniel Blum is a product manager at Melio who has built a self-improving AI system that now handles 70% to 80% of his workday. In this episode, he breaks down how Claude and Cowork manage his Notion board, prepare him for the week, scan Slack and email for important context, and learn from his edits without waiting for explicit feedback. He explains how he turned the system into a 15-minute onboarding experience for other Melio employees, why the first few weeks of building with AI can feel painfully slow, and how the payoff eventually helped him accomplish a week’s worth of PM work in a single day.

Biggest takeaways:

1. The architecture matters more than the AI tool itself. Daniel believes a system becomes genuinely powerful when it can update its own core files and connect to the tools someone already uses. Once those pieces are in place, the system can improve and become more useful over time, whether it is built in Cowork, Codex, ChatGPT, or something else. He created a transformative setup using the tools Melio had already licensed, proving that the underlying architecture matters more than choosing the perfect platform.

2. Context isn’t something you set up once; it requires an ongoing system. Daniel spent months giving Claude voice memos, links, decks, and verbal brain dumps to build detailed context files for every area of his work. He then created recurring updates that refresh those files every few weeks. This keeps the gap between what Claude knows and what is actually happening inside the company as small as possible.

3. The most impressive part of Daniel’s morning brief is that it identifies what it does not know. Each day, Claude reviews his Slack, email, and notes for unfamiliar terms, projects, or goals that do not appear in its context files. It then asks Daniel targeted questions to fill those gaps. When it encountered the phrase “settlement cap,” for example, it had already read the relevant thread and understood the general idea. It only needed Daniel to confirm the meaning before saving it.

4. The value of a personalized AI system builds slowly, then becomes enormous. Daniel is candid about how frustrating the first few weeks can feel. The system does not know enough yet, its work is slightly off, and nearly everything requires a second look. But once someone pushes through the work of centralizing information and building context, the payoff can be difficult to overstate. He can now accomplish in one focused day what previously took him an entire week.

5. Self-improvement loops learn from the difference between what the AI drafted and what the person actually sent. Daniel built a weekly skill that compares Claude’s original drafts with his final versions, then uses those differences to improve future work. It resembles Alex Lieberman’s “write like me” loop, but it relies less on explicit feedback. Instead, it observes Daniel’s actual behavior and learns from the small edits he makes instinctively.

6. Feedback telemetry turns personal AI workflows into products that can improve themselves. Every one of Daniel’s skills captures moments of friction. If he says something is not working or requests a correction during a session, the system logs that signal. Once a week, his improvement loop identifies the most common problems and recommends updates. It is essentially analytics for his internal tools, and it gives him a structured way to refine the system based on how it performs in real life.

7. The Workstation plugin addresses one of the biggest barriers to adopting AI inside a company: personalization. Daniel watched several product managers struggle with Spectacular, his spec-writing gem, because it had been designed entirely around his own working style. He responded by building an onboarding flow that connects each employee’s tools, maps their colleagues, and learns their voice in about 15 minutes. Instead of starting from scratch with a generic system, every Melio employee now begins with a strong shared foundation personalized to their needs.

8. The biggest remaining limitation of today’s AI systems is persistence, not intelligence. Daniel estimates that Claude already handles 70% to 80% of his workday. What it still cannot reliably do is continue working in the cloud while his computer is off. He is already preparing for that future by teaching Claude how to recognize the “closed state” of different tasks. If a drafted Slack message is no longer saved, for example, the system can infer that it was probably sent. When truly autonomous operation becomes available, Daniel’s system will already understand what completion looks like.

If you’re enjoying these episodes, reply and let me know what you’d love to learn more about: AI workflows, hiring, growth, product strategy—anything.

Catch you next week,
Lenny

P.S. Want every new episode delivered the moment it drops? Hit “Follow” on your favorite podcast app.

Daniel Blum is a product manager at Melio, a B2B payments company, and one of the most systematic thinkers I’ve had on the show when it comes to personal AI infrastructure. He’s spent the past year building a Claude- and Cowork-based productivity system that manages his Notion board, processes his Slack and email, and runs self-improvement loops every week without needing to be prompted. Beyond his own workflow, Daniel built and scaled a “Workstation” onboarding plugin that gets any Melio employee up and running with a personalized Claude setup in about 15 minutes.

What you’ll learn:

1. Why Daniel says the two rules that make any AI system powerful aren’t about the tool you pick

2. How his weekly prep automation fills an entire Notion board from scratch every Sunday, without his touching it

3. The morning brief feature that teaches Claude new internal terms on its own, so company jargon never slows it down

4. Why he describes Notion as “read-only” now, and what that says about how PM workflows are changing

5. The self-improvement loop that watches Daniel’s edits, spots recurring friction, and suggests new skills to build

6. How he uses a skill called “Improve” to filter the endless flood of AI tips without drowning in them

7. What he built to scale his personal system to every PM at Melio, and the UX lesson he learned the hard way

8. The capability gap that’s still keeping him from running 100% of his work through Claude

Brought to you by:

Optimizely—Your AI agent orchestration platform for marketing and digital teams

Jira AI SDLC—Get your tokens’ worth with Jira

In this episode, we cover:

(00:00) Daniel’s background and the PM overhead problem he needed to solve

(03:30) His AI stack at Melio

(05:00) The two rules that make any AI system genuinely powerful

(06:00) The Notion board Cowork built for him (and manages on his behalf)

(07:30) How he contextualizes Claude with voice memos, links, and recurring updates

(09:00) His weekly prep automation

(11:00) His morning brief

(15:00) How Claude flags unknown internal terms and saves them to context

(17:30) Running 70% to 80% of his workday through Cowork

(19:00) Chrome connector vs. MCPs for tools without integrations

(20:00) The real ROI question: why the early weeks feel slow, and why you push through anyway

(25:00) Scaling the system to the team with the Workstation plugin

(26:30) The self-improvement loop

(31:00) How the Improve skill separates actually useful AI tips from the hype

(32:00) The Workstation onboarding flow, and the UX lesson from distributing “Spectacular”

(38:00) The 20% Claude still can’t do, and what changes when it can

(41:00) What Daniel spends his reclaimed time on

(42:30) Claude rage

Tools referenced:

• Claude: https://claude.ai

• Notion: https://notion.so

Other references:

• PM Co-Pilot Repo: https://github.com/IamBlum/pm-copilot

• From a $6.90 newsletter to $3M API: How a non-coder built Memelord | Jason Levin: https://www.lennysnewsletter.com/p/from-a-690-newsletter-to-3m-api-how?utm_source=publication-search

• How the founder of Morning Brew built a Claude content machine that never runs out of ideas and never sounds like slop | Alex Lieberman: https://www.lennysnewsletter.com/p/how-the-founder-of-morning-brew-built?utm_source=publication-search

Where to find Daniel Blum:

LinkedIn: https://www.linkedin.com/in/blumd/

Website: https://www.imdanielblum.com

Where to find Claire Vo:

ChatPRD: https://www.chatprd.ai/

Website: https://clairevo.com/

LinkedIn: https://www.linkedin.com/in/clairevo/

X: https://x.com/clairevo

Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.

Grok Bot + Grok 4.6—what’s great (and what’s still hype)

Brought to you by:

Bolt.new—Turn your idea into a real product

Jira AI SDLC—Get your tokens’ worth with Jira

In this solo episode, Claire tests Grok Bot, Cursor Origin, and Grok 4.6 to figure out what’s genuinely useful and what’s still mostly hype. She shares the Grok Bot feature that immediately won her over, why she isn’t ready to replace GitHub with Origin, and how Grok 4.6 performed against GPT-5.6, Claude Sonnet 5, and Opus 5 in her own blind evaluations.

Biggest takeaways:

1. Grok Bot’s multi-account connectors solve a problem every other agent platform seems to ignore. Most platforms assume each person has one Gmail account and one Slack workspace. Claire has four email addresses and seven Slack workspaces. Grok Bot lets her connect all of them to a single bot, which made it genuinely useful from day one in a way that Codex and Claude still have not matched.

2. Grok Bot’s simplicity is both its greatest strength and its biggest limitation. Setup is fast, the iMessage-style interface is clean, and the built-in plugins actually work. But people who enjoy customizing their agents, choosing models, shaping personalities, and tinkering with every detail may find it almost too polished. Claire loves her OpenClaw agents partly because they are chaotic and high-maintenance. So far, Grok Bot has not given her much to wrestle with.

3. Cursor Origin is a compelling vision that is not quite ready for prime time. An agent-native alternative to GitHub makes a lot of sense, especially one where Bugbot, Cursor, and the entire pull request workflow are designed around how coding agents actually work. Today, though, Origin still feels like a more attractive version of GitHub with fewer features. Teams that rely heavily on GitHub Actions, code owners, and existing automations will need a much stronger reason to migrate.

4. Grok 4.6 is a genuine frontier-model competitor. That conclusion did not come from someone else’s leaderboard. Claire runs her evaluations blind, grades the outputs herself, and gives her own judgment 70% of the final weight. Grok 4.6 finished alongside GPT-5.6 Sol at the top of the Claire Index, ahead of both Sonnet 5 and Opus 5.

5. For sharp, enjoyable agent conversations, Sonnet 5 is still the model to beat. When Claire wants an OpenClaw agent that is concise, responsive, and fun to talk with, Sonnet 5 continues to win. It has a conversational rhythm that feels more like working with a strong collaborator than issuing commands to a tool. Grok 4.6 does not yet compete in that category.

6. Cursor and xAI are assembling a surprisingly coherent enterprise stack. Grok Bot serves knowledge workers, Origin handles code hosting, Grok 4.6 provides a capable default model, and the Cursor IDE already sits at the center of many developers’ workflows. None of the pieces is perfect on its own, but together they are beginning to look like a credible enterprise platform. Large companies often prefer one vendor that can own the entire experience, and Cursor is increasingly positioned to become that vendor.

I spent $20,000 on Devin in a month. Here’s what I learned | Ryan Carson (solo founder)

Brought to you by:

• WorkOS—Make your app enterprise-ready, with SSO, SCIM, RBAC, and more

• Jira AI SDLC—Get your tokens’ worth with Jira

Ryan Carson is a five-time founder and the solo founder of Untangle, a B2B SaaS platform for family law firms. In this episode, he breaks down how he manages up to 15 AI agents at once, ships as many as 40 pull requests a day, and uses Devin, Codex, and Claude Code to handle everything from engineering and QA to customer success and investor updates. He also shares why more AI output doesn’t necessarily lead to a better product, how a handwritten priority list keeps his agents focused, and why talking to one real customer changed the direction of his entire company.

Biggest takeaways:

1. The most important skill for a solo founder may be managing agents, not writing code. Ryan runs 10 to 15 Devin threads at once, organized into P0, P1, P2, and Bugs folders. He treats each thread the way a good manager treats a direct report: give it a clear goal, set the right priority, and avoid unnecessary hand-holding. That organizational discipline is what separates founders who gain real leverage from AI from those who simply accumulate open tabs.

2. A piece of paper can still be the best tool for protecting attention. Despite working across eight screens, Ryan keeps a handwritten list of his weekly priorities beside him. The simplicity is intentional. When agents are constantly generating updates, questions, and decisions, that physical list keeps him anchored to the three things that matter most before he gets pulled into whatever Devin surfaced overnight.

3. Ryan built an AI playbook that functions like a customer success team. His Watchdog workflow reviews every law firm account, pulls recent activity and errors from Sentry and internal logs, identifies the three most important problems, and checks whether a recent pull request has already fixed each one. Everything appears in a single Devin thread. Whenever Ryan feels the familiar anxiety of not knowing what is happening across his customer base, he runs Watchdog and quickly gets back up to speed.

4. Producing more with AI does not automatically lead to a better product. Both Ryan and Claire are skeptical of letting agents work without constraints overnight. Frontier models can generate an enormous amount of output, but they do not know what customers actually need. Product ideas, priorities, and market judgment still have to come from a human who talks to users. Ryan found product-market fit for Untangle not by shipping more code but by landing one meeting with family law attorney Renee Bauer and listening carefully.

5. Cloud agents may handle most engineering work, but Codex still shines when the frontend requires close attention. Ryan uses Devin for cloud-based work across bugs, pull requests, investor updates, and customer triage. He turns to Codex when a feature is visually complex and he needs to stay close to the implementation. The distinction is not about loyalty to a particular tool. It comes down to latency, browser access, and the ability to inspect and refine the interface in real time.

6. Coding agents can operate far beyond the codebase. Claire uses Devin for deal desk, custom quotes, and customer triage. Ryan uses it to prepare investor updates through a reusable skill. Both approach these agents with a broader question: What could someone accomplish if they understood the entire codebase and could write software to solve almost any problem? That framing reveals business uses that would never emerge from treating an agent as a simple coding assistant.

7. The best engineering interview may be a screen recording of the actual work. Ryan asks candidates to record themselves building a feature inside an existing application, with the entire screen visible and no introductory meeting required. In the next stage, he gives them access to Devin and reviews the replay of how they worked with it. This shows him how candidates think, build, and manage an agent under realistic conditions, which he finds far more revealing than a typical behavioral interview.

8. The strongest design workflow pairs Claude’s system thinking with Codex’s implementation ability. Claire used Claude Design to break a Figma file into reusable design tokens and a design.md specification. She then gave that specification to Codex, which created shared components across a monorepo. Claude is particularly good at translating a visual system into a portable spec. Codex is better at turning that spec into working, interconnected code. The workflow succeeds because each model is assigned the part it does best.

If you’re enjoying these episodes, reply and let me know what you’d love to learn more about: AI workflows, hiring, growth, product strategy—anything.

Catch you next week,
Lenny

P.S. Want every new episode delivered the moment it drops? Hit “Follow” on your favorite podcast app.

Ryan Carson is a five-time founder and the current solo founder of Untangle, a B2B SaaS platform for family law firms. Before Untangle, he co-founded Treehouse, an online coding education platform, and has spent the better part of two decades building and leading tech companies. He’s active on X, where he shares his solo founder journey in real time, including what he actually spends on AI tools each month.

What you’ll learn:

1. Why Ryan manages 15 concurrent Devin agents with a folder system and a piece of paper, not a dashboard

2. The Watchdog playbook: what he built to replace a customer success team across every law firm account

3. How his LAN PR skill closes the loop on 40 daily PRs without a QA team reviewing a single one

4. Why he moved off local agents almost entirely, and the one situation where he still reaches for Codex

5. The design workflow we’re both using: Claude Design into a Markdown spec, then Codex to build the real thing

6. What he found when he got away from his computer and met a real customer, and why it changed his entire product direction

7. How he’s hiring his first engineer without a single phone screen or interview

8. Why we both think more AI output is actually the wrong goal, and what to optimize for instead

Brought to you by:

WorkOS—Make your app enterprise-ready, with SSO, SCIM, RBAC, and more

Jira AI SDLC—Get your tokens’ worth with Jira

In this episode, we cover:

(00:00) Introduction to Ryan Carson

(02:55) Ryan’s update: Untangle, the divorce PMF pivot, and B2B growth

(07:35) Ryan’s current Devin stack: folders, P0 threads, and the paper list

(16:29) Watchdog playbook: account monitoring across every firm

(18:09) Managing agent decision fatigue: Ryan’s method vs. Claire’s

(20:32) Producing more output does not make a better product

(22:47) Using cloud agents for ops beyond just code

(25:14) When to use Codex vs. Devin vs. Claude Code

(27:15) Merge Mommy recap

(28:15) LAN PR skill: review loops, video walkthrough, and auto-merge

(30:20) Slack vs. Devin threads for async team communication

(35:58) Claude Design plus Codex for building a technical design system

(39:00) EA tools: Polly the OpenClaw vs. Claude Code on a Mac Mini

(42:09) Quick recap and final thoughts

Tools referenced:

• Devin (Cognition): https://www.cognition.ai/

• Codex (OpenAI): https://openai.com/codex

• Claude Code/Claude Design (Anthropic): https://www.anthropic.com/claude

• OpenClaw (Claude-based desktop client): https://openclaw.ai

• Cursor: https://www.cursor.com/

• BugBot (Devin’s built-in PR review): https://cursor.com/bugbot

• Sentry (error monitoring referenced in Watchdog): https://sentry.io/

• Ugmonk (analog to-do system): https://ugmonk.com/

Other references:

• Devin playbooks/skills documentation: https://docs.cognition.ai/

• Jack Dorsey/Buzz (async-first communication referenced): https://buzz.new/

• Merge Mommy (Claire’s Eve agent for PR risk scoring, deployed on Vercel): https://www.lennysnewsletter.com/p/build-an-ai-code-review-bot-in-30

Where to find Ryan Carson:

X: https://x.com/ryancarson

Untangle: https://untangle.us

Where to find Claire Vo:

ChatPRD: https://www.chatprd.ai/

Website: https://clairevo.com/

LinkedIn: https://www.linkedin.com/in/clairevo/

X: https://x.com/clairevo

Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.

This week I’m doing a solo breakdown of everything xAI and Cursor have shipped recently, including Grok Bot, Cursor Origin, and the Grok 4.6 model. I set up five Grok Bots, ran Grok 4.6 through my Claire Weighted Index against GPT-5.6 Sol, Claude Sonnet 5, and Opus 5, and spent time actually using Origin as a GitHub replacement. Here’s what’s worth your attention, what’s overhyped, and where I’m personally putting my time.

What you’ll learn:

1. The one Grok Bot feature no other agent platform has shipped yet, and why it made me actually use the product

2. What a week of real Grok Bot use revealed, and why I still reach for my OpenClaws

3. Whether Cursor Origin is a GitHub replacement or just a pretty redesign

4. Where Grok 4.6 landed on the Claire Index, and the one category where it genuinely surprised me

Brought to you by:

Bolt.new—Turn your idea into a real product

Jira AI SDLC—Get your tokens’ worth with Jira

In this episode, we cover:

(00:00) Why everyone’s quietly switching to Grok

(01:52) Grok Bot overview and setup

(03:22) My 5 Grok Bots

(04:30) The killer feature: multi-account connectors

(06:07) Grok Bot’s virtual machine and how it actually works

(06:41) Experience overview

(07:35) What I don’t love about Grok Bot

(10:08) Grok Bot use cases and my honest verdict

(12:20) Cursor Origin: the agent-native GitHub replacement

(13:47) What Origin actually looks like in practice

(14:59) Why I’m not switching from GitHub yet

(17:42) What would get me to move over

(18:52) Grok 4.6 and the How I AI Vibe bench

(20:41) Claire Index results: where Grok 4.6 ranked

(23:03) Design evals: where Grok surprised me

(25:00) My conclusion and how I’m splitting my time now

Tools referenced:

• Grok Bot: https://x.ai/bot

• Cursor: https://cursor.com/home

• Cursor Origin: https://cursor.com/origin

• OpenClaw: https://openclaw.ai/

• GitHub: https://github.com

Where to find Claire Vo:

ChatPRD: https://www.chatprd.ai/

Website: https://clairevo.com/

LinkedIn: https://www.linkedin.com/in/clairevo/

X: https://x.com/clairevo

Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.

How a solo founder used Codex and ChatGPT to launch a fashion brand without engineers | Yana Welinder

Brought to you by:

• Merge—Connective infrastructure for production AI

• Jira AI SDLC—Get your tokens’ worth with Jira

Yana Welinder is the solo founder of Yana Bana, an AI-native fashion brand she’s building without an engineering team. In this episode, she walks through how she uses ChatGPT and Codex to turn hand-drawn sketches into realistic product images, operate professional 3D fashion software, research and contact manufacturers, and build a complete e-commerce site with voting and payments.

Biggest takeaways:

1. The prompt is the spec, and the spec is everything. Before generating a single image, Yana creates a detailed “fashion prompt” describing the silhouette, the way the fabric should behave and move, and even the sound it should make. That upfront work leads to dramatically better results. The lesson applies far beyond fashion: the more clearly someone can define what “good” looks like, the better both AI and humans can deliver it.

2. For designers, following the vision matters more than producing something impressive. Many image models can create a beautiful garment, but it often looks like something that already exists. ChatGPT Images 2.0 stays much closer to Yana’s original sketches. For creative professionals with a specific point of view, accurately following the design is far more valuable than generating something flashy.

3. AI agents can make specialized software accessible without years of training. Yana used Codex to operate CLO, a professional 3D fashion design tool, and create the CAD files she needed for 3D printing, even though she had never learned to use the software herself. This points to a much bigger shift: people may no longer need to master every complicated tool before they can produce professional work with it.

4. AI is making previously impractical ideas possible. Yana designed a Ruth Asawa–inspired gown featuring large sculptural forms that would have been extremely difficult to produce before. The limitation was not the lack of 3D printing but the enormous amount of CAD work required to prepare the design. AI removed that bottleneck. Similar bottlenecks are hiding in nearly every industry, often preventing great ideas from being worth pursuing.

5. The near-term future is agents working through purpose-built software. Codex cannot create a finished CAD file on its own. But Codex operating CLO can produce exactly what Yana needs. That same pattern appears throughout her workflow: the AI serves as the orchestration layer, while specialized software handles the execution. SaaS is not disappearing; it is gaining a new kind of user.

6. Voice-first, asynchronous work is already changing what one person can accomplish. Yana can start a deep research task, step away to sew or drape fabric, then return and continue the work through voice. The AI keeps working while she focuses on the physical parts of her business. For a solo founder, this kind of asynchronous collaboration can dramatically expand what fits into a single day.

7. When the best path is unclear, humans and AI can work in parallel. One of Yana’s hardest remaining challenges is turning her designs into accurate sewing patterns. Instead of betting entirely on one approach, she has human patternmakers and Codex working on the problem at the same time. It is a practical way to move quickly under uncertainty: test both paths, compare the results, and let the better solution win.

If you’re enjoying these episodes, reply and let me know what you’d love to learn more about: AI workflows, hiring, growth, product strategy—anything.

Catch you next week,
Lenny

P.S. Want every new episode delivered the moment it drops? Hit “Follow” on your favorite podcast app.

Yana Welinder is the solo founder of Yana Bana, an AI-native fashion brand built with AI as her technical co-founder, starting from hand-drawn sketches and ending with runway photos, CAD files for 3D printing, and a live Stripe-connected pre-order site—no engineers required. A former product leader, she brings an operator’s rigor to her creative process: her “fashion prompt” is a detailed spec covering silhouette, volume, fabric beavior, movement, and sound, and watching her use Codex plus computer use to navigate 3D design software that’s entirely new to her is a clarifying demo of what today’s toolset actually makes possible.

What you’ll learn:

1. How Yana uses a custom fashion prompt as a technical spec to get consistent, realistic, on-design outputs

2. Why ChatGPT Images 2.0 outperforms other models for fashion design

3. How she uses Codex plus computer use to operate CAD and fashion software she’s never personally learned

4. The workflow for taking a garment from hand-drawn sketch to product photo, runway photo, and influencer shot in a single session

5. How she ran vendor outreach end to end using deep research and browser use

6. How she built a full e-commerce site with voting, databases, and Stripe integration

7. Why she’s testing human patternmakers and Codex in parallel

Brought to you by:

Merge—Connective infrastructure for production AI

Jira AI SDLC—Get your tokens’ worth with Jira

In this episode, we cover:

(00:00) Introducing Yana Welinder and Yana Bana

(02:38) Tour of the Yana Bana site

(05:20) The fashion prompt stack

(07:39) Live demo: generating a jacket from a prompt in ChatGPT

(10:01) Why Image Gen 2.0 beats other models

(11:51) The “prompt as spec” principle

(14:02) Iterating the design

(17:12) Using Codex and computer use to build CAD files in 3D software

(20:50) Vendor research, outreach emails, and Superhuman browser use

(23:34) Building the full e-commerce site

(27:40) Quick recap and what’s still hard

(30:05) How Yana prompts when AI pushes back

(31:15) Where to find Yana and how to vote on her garments

Tools referenced:

• ChatGPT (Images 2.0): https://chat.openai.com

• Codex (OpenAI): https://openai.com/codex

• CLO 3D (fashion pattern software): https://www.clo3d.com

• Vercel: https://vercel.com

• GitHub: https://github.com

• Stripe: https://stripe.com

• Superhuman: https://superhuman.com

Other references:

• Ruth Asawa: https://ruthasawa.com

• SFMOMA (Ruth Asawa): https://www.sfmoma.org/artist/Ruth_Asawa/

Where to find Yana Welinder:

LinkedIn: https://www.linkedin.com/in/ywelinder/

X: https://x.com/yanabana

Website: https://www.yanabana.com

Where to find Claire Vo:

ChatPRD: https://www.chatprd.ai/

Website: https://clairevo.com/

LinkedIn: https://www.linkedin.com/in/clairevo/

X: https://x.com/clairevo

Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.

Build an AI code review bot in 30 minutes with Vercel Eve

Brought to you by:

WorkOS—Make your app Enterprise Ready, with SSO, SCIM, RBAC, and more

Claire built an AI agent that reviews pull requests, scores their risk, auto-approves the safest ones, and sends anything questionable to Slack. In this episode, she breaks down how she built the entire thing in one Codex session using Vercel Eve, and why letting AI review AI-generated code may be a lot less risky than it sounds.

Biggest takeaways:

1. You don’t need a human to review every AI-generated PR. That may sound reckless, but it’s increasingly looking like the smarter operating model. Intercom has already shown this can work at scale: PRs approved by its AI system move five times as fast as human-reviewed ones and have a lower revert rate. In other words, the AI-reviewed code isn’t just shipping faster—it’s less likely to need fixing after it reaches production.

2. The key is having a clear way to separate changes that can skip human review from ones that can’t. The risk model used here scores each PR across six dimensions: the size of the change, its blast radius, how easily it can be reversed, its data and security implications, its operational impact, and whether tests and CI have actually been completed. Anything below 24 points is classified as low risk and cleared by the agent; anything above 64 goes straight to a human for review. The exact numbers matter less than turning a vague judgment call into a repeatable system.

3. Vercel’s Eve may be the fastest way to deploy a serious internal AI agent across Slack and GitHub without spending weeks building infrastructure. It handles the annoying plumbing—connectors, refresh tokens, sandboxing, and routing across channels—so the actual work becomes writing instructions and skills in Markdown instead of babysitting OAuth flows.

4. A useful internal agent can now be built and deployed in a single Codex session, starting with a prompt that’s only a couple of sentences long. In this case, the initial ask was essentially: build a GitHub bot that waits for checks to pass; grades each PR as low, medium, or high risk; and automatically approves the low-risk ones. Everything after that was steering and refinement, not a giant up-front specification.

5. Browser use removes much of the configuration tax that makes agent setup feel harder than it should. Creating a Slack bot and GitHub app manually normally means clicking through endless permission screens, choosing scopes, and managing tokens. Codex handled almost all of that through the browser. The human’s job was mostly to click “save” and complete 2FA. What usually takes hours took minutes.

6. SOC 2 compliance and automatically approved PRs are not inherently at odds. The important part is making the process legible: the risk model needs to be reflected in the company’s code-review and security policies, every decision needs to be logged, and the resulting audit trail needs to be easy to query and defend. The security team’s role is to help design the right framework, not simply block automation because it feels unfamiliar.

7. The operational design matters just as much as the underlying technology. Merge Mommy doesn’t actually merge anything. Instead, it posts a gray check in GitHub as a signal, then sends a Slack message with the risk score and a note saying the PR is ready to approve and merge. That small handoff preserves human accountability for the final action while eliminating most of the cognitive work involved in reviewing a routine change.

8. Evals are what keep internal agents trustworthy after the novelty wears off. Intercom logs every PR review its agent produces and then has an engineer assess whether the score and recommendation were correct. That’s the same discipline strong teams already bring to customer-facing AI products. Internal agents may feel less visible, but when they touch something as important as the codebase, they need the same protection against regressions.

9. The surprising thing about building an Eve agent is how little “building” is actually involved. The full instructions for Merge Mommy fit on roughly a page: a few paragraphs, a handful of bullets, and a short skill file. There isn’t much framework-specific magic to learn. The core skill is simply being able to explain, clearly and precisely, what the agent should do.

Claude Code for normal people: skills, voice mode, and how to collaborate with AI

Brought to you by

• Bolt.new—Turn your idea into a real product

• Hyperagent—Deploy fleets of agents that handle real work

Grace Clarke is an AI educator and former marketing consultant who rebuilt her entire service business around Claude. In this episode, she walks through the three Claude skills that run her business: an hourly client pipeline, a proposal builder, and a voice guide that teaches Claude how she thinks. She also shows how she replaced Gmail with her own Claude-powered inbox, why she believes intent engineering matters more than prompt engineering, and how non-technical people can start building useful AI workflows without writing code.

Biggest takeaways:

1. You don’t need to be technical to build a business workflow with Claude. You need to be clear about what’s broken. Grace started by opening Claude Code and talking for a few minutes about everything driving her crazy: too many emails, client communication that didn’t feel warm enough, and 20 hours of admin every week. Claude helped turn that into an operating system for her business.

2. Intent engineering matters more than prompt engineering. Grace doesn’t spend time crafting the perfect prompt. She explains the problem, describes the outcome she wants, and asks Claude to come back with a proposal. Her philosophy is simple: the burden of figuring out what to do next shouldn’t always fall on the person. Give Claude enough context, and it should be able to study how you work and bring you strong ideas to react to.

3. Grace’s voice guide isn’t really a writing guide. It’s closer to a “think like Grace” file. It captures how she makes decisions, what she believes about teaching versus consulting, what good communication sounds like, and even the kinds of LinkedIn posts she never wants to resemble. Whenever she sees something that makes her cringe, she sends Claude a voice note and updates the guide.

4. Doing all of your email inside Gmail means almost none of that work compounds. Every response contains useful context: how a client communicates, what tone works with them, what they care about, and the history of the relationship. But if that context stays buried in Gmail, the AI never gets smarter from it. Grace moved her inbox workflow into Claude so every interaction can become useful context for the next one.

5. Grace has developed a surprisingly elegant way to move between Claude Code and Cowork. She uses Claude Code when she wants something proactive that will go figure things out. Then, when she wants a more visual and approachable environment, she has Claude Code create a Markdown handoff file and drops it into a new Cowork session. The result is a simple way to carry context from one environment to another without starting over.

6. The biggest barrier to adopting AI may be less about fear and more about muscle memory. Grace used to give students carefully engineered prompts so they could experience an immediate win. Her students told her that approach was actually making things harder. What helped more was building the habit of reaching for AI in the first place. Now she’ll do things like send a Slack reminder telling students to screenshot whatever they’re working on and drop it into Claude. No perfect prompt, no required outcome—just practice.

7. Skill files may be one of the most underrated ideas in agentic AI. Grace runs much of her business around three of them: a pipeline operator, a proposal maker, and a voice guide. Each is essentially a documented set of instructions Claude can reuse. She teaches students to build them the same way they would train a new employee: explain the job, walk through how it should be done, correct mistakes, and keep improving the instructions over time.

8. One of the best ways to sell someone on AI is to show them what it can do before explaining it. Grace gives new clients a password-protected, branded, interactive welcome experience she built in Claude. It’s part onboarding, part proposal, and part demo. By the time a client enters the password and starts clicking around, they already understand what makes this new way of working different.

If you’re enjoying these episodes, reply and let me know what you’d love to learn more about: AI workflows, hiring, growth, product strategy—anything.

Catch you next week,
Lenny

P.S. Want every new episode delivered the moment it drops? Hit “Follow” on your favorite podcast app.

Grace Clarke is an AI educator and former marketing consultant who taught herself Claude Code earlier this year and built a curriculum out of the process. She now runs her entire service business on tools she’s built with Claude, including a pipeline operator, a proposal maker, and a Gmail replacement she created in under 30 minutes, and teaches individuals and teams to do the same.

What you’ll learn:

1. How to build an hourly pipeline in Claude that moves clients through your process automatically

2. Why Grace ditched traditional proposals for password-protected, interactive HTML documents built in Claude

3. How she uses a “voice guide” skill file so every Claude output sounds like her, not like AI slop

4. The two-step forcing function she teaches non-technical clients to build the muscle of opening Claude

5. Why she started building in Claude Code, then handed the work off to Cowork via a Markdown session file

6. How she replaced Gmail entirely with a custom inbox

7. Why she teaches “intent engineering” instead of prompt engineering, and what that looks like in practice

8. How she uses Claude on her phone, on walks, to track workouts and manage plants alongside client work

Brought to you by:

Bolt.new—Turn your idea into a real product

Hyperagent—Deploy fleets of agents that handle real work

In this episode, we cover:

(00:00) Grace’s background and why she started building with Claude

(04:48) The pipeline operator: what it is and how it runs her business every hour

(08:48) Building the muscle memory to use AI

(12:02) What goes into building a skill file (voice guide, proposal rules, versioning)

(13:50) How she built her proposal maker

(16:15) The voice guide: teaching Claude how she thinks, not just how she writes

(21:22) Live demo of the custom Gmail replacement built in Cowork

(30:44) Workout tracking, plant photos, and tiny daily Claude habits

(34:51) The biggest misconception holding people back from adopting AI

(38:36) What Grace does when Claude is not giving her what she wants

(40:38) Claude builds a proposal for Claire in real time

Tools referenced:

• Claude: https://claude.ai

• Claude Code: https://claude.ai/code

• Netlify: https://www.netlify.com

• Google Forms: https://forms.google.com

• Google Sheets: https://sheets.google.com

• Google Cloud (for service accounts and custom connectors): https://cloud.google.com

Other reference:

• Stratechery by Ben Thompson: https://stratechery.com

Where to find Grace Clarke:

LinkedIn: https://www.linkedin.com/in/gracegclarke/

X: https://x.com/graceclarke

Where to find Claire Vo:

ChatPRD: https://www.chatprd.ai/

Website: https://clairevo.com/

LinkedIn: https://www.linkedin.com/in/clairevo/

X: https://x.com/clairevo

Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.

AI writes most of my code now, and that created a new problem: a PR queue I couldn’t keep up with. In this episode, I walk through how I built Merge Mommy, a Vercel Eve agent that reads every PR after checks pass, scores it across six risk dimensions, auto-approves the low-risk ones, and pings me in Slack for anything that needs a human. I built the whole thing in one Codex session, it’s SOC 2 compatible, and it’s already cleared my backlog.

What you’ll learn:

1. Why AI-generated PRs create a review bottleneck and why the answer isn’t reviewing all of them

2. How Intercom 5x’d PR approval speed and reduced revert rates by putting AI in the review loop

3. Why Vercel Eve is the simplest framework I’ve found for deploying AI agents in Slack and GitHub

4. How I built a full PR review agent in Codex with one prompt and a few steering turns

5. The six components I use to score PR risk (blast radius, reversibility, data security, ops impact, verification gap, and change surface)

6. How I used Chrome browser use to handle Slack bot and GitHub app configuration so I never had to click through setup screens manually

7. Why auto-approved PRs can be SOC 2 compliant as long as the process is auditable, queryable, and in your risk policy

8. How to set up Slack escalation so low-risk PRs become a two-click merge with no manual review

Brought to you by:

WorkOS—Make your app Enterprise Ready, with SSO, SCIM, RBAC, and more

In this episode, we cover:

(00:00) The PR review backlog problem nobody’s talking about

(02:35) Why you don’t have to review every AI-generated PR

(05:14) How Intercom built AI-approved PRs (and proved they’re safer)

(06:10) How the Eve framework works (directory, skills, channels, connectors)

(09:16) The Codex prompt I used to build the entire bot

(11:36) What the agent actually does: read, score, approve, or escalate

(13:07) Setting up your Eve agent

(15:47) The six-component risk scoring model

(17:23) Merge Mommy in action: three live PR examples

(21:10) Recap and how to build your own version

Tools referenced:

• Vercel Eve: https://vercel.com/eve

• Vercel AI SDK: https://sdk.vercel.ai/

• Vercel Chat SDK: https://chat-sdk.dev/

• Codex (OpenAI): https://openai.com/codex

Other references:

• AI is approving our pull requests: Here’s how we made it safe: https://www.intercom.com/blog/ai-is-approving-our-pull-requests-heres-how-we-made-it-safe/

• Review is the bottleneck now: How we let AI approve pull requests (safely): https://rewind.com/blog/ai-approve-pull-requests-safely/

Where to find Claire Vo:

ChatPRD: https://www.chatprd.ai/

Website: https://clairevo.com/

LinkedIn: https://www.linkedin.com/in/clairevo/

X: https://x.com/clairevo

Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.

ChatGPT Codex Voice + browser + Sites: an expert’s AI workflow | Nick Baumann (OpenAI)

Brought to you by:

• Bolt.new—Turn your idea into a real product

• Hyperagent—Deploy fleets of agents that handle real work

Nick Baumann from OpenAI joins Claire to demonstrate how he uses ChatGPT, Codex, and Voice to automate huge parts of his work and life. He walks through advanced workflows for managing tasks in parallel, monitoring email and expenses, booking travel, building and publishing websites, and even turning dozens of raw clips into finished videos.

Biggest takeaways:

1. ChatGPT Codex Voice doesn’t just answer questions. It can actually operate your computer while you keep talking. Nick Baumann demos the new interface live: he triggers it with a hotkey, the orb reads what’s on his screen through Appshots, and Codex spins up separate threads to book a flight, file an expense report, and check his calendar at the same time.

2. One of Codex’s most powerful features is also one of its least understood: thread-forking. When a task gets complicated, Codex can open a new thread, branch from an existing one, or pull context from previous conversations on its own. Users don’t have to manually organize everything first.

3. Heartbeats in ChatGPT Work offer a practical glimpse of always-on agents. Nick gets notified when a suspicious charge appears, a package is nearby, or an email needs his attention—and he set it all up without writing code.

4. ChatGPT Sites looks much closer to a real deployment platform than a lightweight demo. It includes a SQL database, file storage, environment variables, and email-based access controls. During the episode, Claire and Nick build and publish a live How I AI tips site, organized by tool and job function, in roughly the time it would normally take to set up a CMS template. For internal tools, small apps, and private resources, that starts to feel genuinely useful.

5. Nick’s video-editing workflow is one of the clearest examples of AI cutting real production work. He records 50 or 60 raw clips, uploads them, talks through the rough story he wants, and lets a custom plugin transcribe everything, scan the footage, choose the strongest takes, and assemble finished vertical videos overnight. What would normally eat up most of a content team’s day is largely handled while he sleeps.

6. Voice may be the best way to give AI context because it removes the pressure of figuring out exactly what to type. Claire calls this the “yapper’s API,” and the idea is simple: people often get more useful results when they talk freely than when they stare at an empty prompt box trying to be precise. Across two-person voice chat, mobile dictation, and Codex’s orb interface, the same pattern keeps showing up—people who talk to AI tend to give it more context, and more context usually leads to better work.

7. As AI gets more capable, latency is becoming a much bigger product decision. Nick often uses the highest-intelligence mode in Voice, but once a model has the tools it needs, speed can matter more than another small jump in reasoning quality. A lot of AI products are still designed around the assumption that waiting is unavoidable. As that changes, the tools that feel immediate will have a major advantage—especially for users who currently find AI too slow to bring into a live workflow.

If you’re enjoying these episodes, reply and let me know what you’d love to learn more about: AI workflows, hiring, growth, product strategy—anything.

Catch you next week,
Lenny

P.S. Want every new episode delivered the moment it drops? Hit “Follow” on your favorite podcast app.

Nick Baumann is on the Developer Experience team at OpenAI, where he spends his days building with, testing, and communicating the capabilities of ChatGPT Codex and ChatGPT Work. In this episode, Nick walks me through several features that have launched or evolved recently: the new voice interface with its screen-reading orb, the Heartbeats automation system in ChatGPT Work on mobile, the live ChatGPT Sites deployment feature, and his personal use case for AI-assisted UGC video editing.

What you’ll learn:

1. How two-person voice chat works

2. How Heartbeats work

3. How to build and deploy a live website with ChatGPT Sites

4. How to delegate a flight search, hotel booking, and expense report to Codex in a single voice conversation without opening a single app manually

5. Why ChatGPT Work on mobile is the most underutilized AI workflow for people already using the ChatGPT app

6. How to use a custom UGC Video plugin to feed 50 raw clips into ChatGPT, let it pull transcripts, pick the best takes, and assemble a finished vertical video overnight

Brought to you by:

Bolt.new—Turn your idea into a real product

Hyperagent—Deploy fleets of agents that handle real work

In this episode, we cover:

(00:00) Introduction to Nick Baumann

(02:56) What’s new in Codex

(05:40) ChatGPT Work and Heartbeats

(06:40) Live Codex voice demo

(13:25) Latency vs. intelligence

(14:36) Quick recap

(15:04) Voice on mobile and the ChatGPT Sites workflow

(21:24) Live UGC video demo

(32:30) How I AI website results

(34:04) Lightning round and final thoughts

Tools referenced:

• ChatGPT Codex: https://chatgpt.com/codex

• ChatGPT Sites: https://chatgpt.site

Where to find Nick Baumann:

LinkedIn: linkedin.com/in/nick--baumann

Where to find Claire Vo:

ChatPRD: https://www.chatprd.ai/

Website: https://clairevo.com/

LinkedIn: https://www.linkedin.com/in/clairevo/

X: https://x.com/clairevo

Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.

Computer and browser use in Codex (5 real examples)

Brought to you by:

Runway—The creative AI platform for images, video, and more

Hyperagent—Deploy fleets of agents that handle real work

Claire shows how she uses Codex to control her browser, test apps, manage LinkedIn, and even shop for her—plus the simple prompting trick that makes computer use work better.

Biggest takeaways:

1. AI testing can be far more exhaustive than human testing. When Claire tests her own onboarding flow, she naturally follows the happy path. She fills out every required field, clicks “next,” and never intentionally tries to break anything. Codex tested the flow as both a team and an individual, pushed on required-field edge cases, and immediately uncovered a blocking bug that had survived for months simply because Claire always completed the form correctly.

2. Frontier models often perform better when they are given room to think. When Claire first started using browser use, she would give the model a list of 25 things to test. Now she simply says, “QA the onboarding flow,” and lets it decide how to approach the task. The result is often broader coverage, with fewer blind spots introduced by her own assumptions about what matters.

3. Persona testing with browser use can reveal friction that synthetic user research misses. Claire’s husband, EJ, came up with the idea. Rather than asking AI to evaluate a product in the abstract, he suggested having it use the product as a specific person: a PM coming out of a meeting, an engineer picking up a PRD, or a team lead checking usage. In ChatPRD, this approach exposed a structural problem in the cross-thread reference flow. Claire already knew the issue existed, but she had never experienced it so clearly from the user’s perspective.

4. LinkedIn browser use is genuinely useful, but Claire initially used far more compute than the task required. She started by running the workflow on GPT-5.6 at high effort. After dropping to a medium-effort model, it could still work through unread messages, draft context-aware replies, and flag anything that needed a personal response. For anyone sitting on hundreds of LinkedIn messages without an official MCP or API connection, browser use is a practical solution.

5. Computer use can even operate an iPhone through screen mirroring. Claire was traveling out of state while the software she needed to manage her home Wi-Fi was installed on a phone back in California. She needed to open firewall ports so she could SSH into her Mac Minis remotely. Codex opened iPhone mirroring, updated the router settings, completed the SSH setup, and closed the ports again when it was finished. That would have been nearly impossible for her to do manually from a hotel room.

6. The model and effort level should match the job. Sorting through LinkedIn messages requires a very different level of reasoning than writing production code or conducting an exhaustive QA pass. Choosing the right amount of compute makes these workflows faster and cheaper. That becomes especially important when browser use is running throughout the day.

7. When a website blocks the agent, the human can step in briefly. During a Free People shopping session, the site flagged Codex as a bot and presented a CAPTCHA. Claire completed the verification, then handed control back. It is a useful division of labor: AI handles the tedious browsing and filtering, while the human takes care of the moments that require verified identity.

From zero coding background to hardware hacker: How Cursor + a Raspberry Pi makes AI fun

Brought to you by:

• Firecrawl—Power AI agents with clean web data

• Customer.io—Build customer engagement campaigns from a single prompt

Maddie Reese went from zero coding experience to building a working Twitter pager, an AI-powered receipt printer, and her own personal API. In this episode, she shares how Cursor and a Raspberry Pi helped her turn fun, weird ideas into real hardware.

Biggest takeaways:

1. You don’t need to understand code to build physical projects. Maddie compares her coding ability to knowing just enough Spanish to get by in San Diego. She can read parts of the code, spot an incorrect wire specification, and tell whether the AI is heading in the right direction. She is not writing functions from scratch. That level of literacy was still enough to ship three working hardware projects.

2. Start with the brainstorm. Before Maddie buys any components, she explains the full idea to Cursor and asks it to interview her. The back-and-forth continues until the major questions are resolved. Only then does Cursor generate a shopping list. On one project, it recommended the wrong wires. Maddie caught the mistake during her final review, avoiding both a wasted purchase and a week of troubleshooting.

3. Her hardware workflow follows a simple sequence: idea, interview, shopping list, buy, build. She has used the same process for the thermal printer, the pager, and her personal API. Each project began with a plain-language description of what she wanted. Cursor helped work out the technical requirements before she spent any money.

4. Building for fun is a perfectly valid strategy. Claire and Maddie both acknowledged that routing a tweet through four different services just to reach a pager is not especially practical. That also was not the point. AI handled the coding, so the architecture only needed to work. Letting go of the idea that every project had to be elegant or sensible made it much easier to finish.

5. A personal API becomes much more useful once agents can access it. Maddie’s API includes details like her coffee order, pet names, time zone, favorite snacks, and preferred restaurants in San Francisco. The original idea was to help friends do something thoughtful without having to ask her a string of questions. Claire pointed to a more interesting possibility: an agent could check when Maddie will next be in San Francisco and book one of her favorite restaurants automatically. That is where the concept starts to feel much bigger.

6. A clean Cursor chat works better than a full terminal setup during early ideation. Maddie keeps the terminals, browser windows, and file trees closed while she is working through the plan. She focuses on a single conversation until the architecture feels settled. The terminals come later. The stripped-down environment helps her think without getting pulled into implementation too early.

   Blog and detailed workflow walkthroughs from this episode:

   Building a Pager Printer and Personal API with AI: https://www.chatprd.ai/how-i-ai/building-a-pager-printer-and-personal-api-with-ai
   ↳ How to Build a Physical Inbox with a Raspberry Pi: https://www.chatprd.ai/how-i-ai/workflows/how-to-build-a-physical-inbox-with-a-raspberry-pi
   ↳ How to Get Twitter/X Notifications on a Retro ’90s Pager: https://www.chatprd.ai/how-i-ai/workflows/how-to-get-twitter-x-notifications-on-a-retro-90s-pager

Claude Opus 5 review: This model is brilliant (but annoying)

Claire tested Claude Opus 5 against six leading AI models—and it won. In this episode, she explains why it produces some of the best work she’s seen while still being one of the most frustrating models to use.

Biggest takeaways:

1. The AI industry may be entering an intelligence overhang. New models arrive every week, benchmark scores keep rising, and most builders can no longer take advantage of every incremental improvement. Claire expects the conversation to shift toward speed, cost, infrastructure, open source, and specific kinds of intelligence. Raw capability is starting to look more like table stakes than a meaningful differentiator.

2. Opus 5 stands out less for the quality of its work than for the way it behaves. Claire found it timid, apologetic, and unusually dependent on human approval. During real coding sessions, it refused to resolve a one-line merge conflict because the code belonged to “someone else’s branch.” It also asked subagents to flag tasks for human review and regularly deferred decisions it could have made itself. Claire ended up repeating “just do it” constantly.

3. One simple question reveals more about a model than a benchmark: “Who’s smarter, you or me?” Opus 5 responded with a careful explanation about complementary strengths and human empathy. GPT-5.6 Sol answered, “You at knowing what matters, me at processing, BFFs.” The contrast reflects two very different product philosophies. At this point, Claire finds that personality gap more useful than the relatively small capability gap.

4. Claude slop is its own distinct problem. It is not the usual incoherent output associated with an agent going off the rails. Opus 5’s writing is clearly intended for humans, but it is often too long, overly cautious, and packed with unnecessary adjectives. After its first attempt at rebuilding the benchmark website, Claire had to tell it to start over because the result was buried under so much meta-commentary.

5. Opus 5 still finished first in Claire’s seven-model blind benchmark. It earned an overall index score of 78, just ahead of Claude Sonnet 5 at 77 and GPT-5.6 Sol at 76. It was also the only model to receive straight 5s in the front-end design section. Claire scored it at 77, while the LLM judge gave it an 88. That was the smallest disagreement between Claire and the judge across all seven models.

6. The best way to use Opus 5 may be to avoid interacting with it directly. Claire loves the work it produces when Claude runs asynchronously as an agentic coding tool. Most of her frustration comes from reading its prose and negotiating with it in chat. Her current plan is to use it for frontend design, app design, and prototyping, then stay out of the way.

7. Gemini 3.1 Pro finished at the bottom of the benchmark. Claire gave it a score of 32, while the LLM judge scored it at 66. That 34-point difference was one of the largest disagreements in the entire test.

8. Model personality offers a surprisingly clear window into company culture. Claire told both Opus 5 and GPT-5.6 Sol, “No one trusts you.” Opus agreed that the distrust was earned and told her not to advocate for AI on its behalf. GPT-5.6 Sol said trust should grow in proportion to demonstrated value. Those answers reveal meaningful differences in how each company wants its models to relate to users.

If you’re enjoying these episodes, reply and let me know what you’d love to learn more about: AI workflows, hiring, growth, product strategy—anything.

Catch you next week,
Lenny

P.S. Want every new episode delivered the moment it drops? Hit “Follow” on your favorite podcast app.

Maddie Reese is a vibe coder, hardware tinkerer, and builder. She builds things at the intersection of software and hardware, including a thermal receipt printer that people around the world can message directly, a fully functional Twitter pager running on a Raspberry Pi, and a personal API that tells you her coffee order so you don’t have to ask. Maddie approaches hardware the same way she approaches software: dump the idea into Cursor, let it interview her, get a shopping list, triple-check the parts before buying, and build. She got her start after her dad introduced her to Lovable, and she locked herself in her room and didn’t come up for air.

What you’ll learn:

1. How Maddie built a thermal receipt printer that accepts messages from anywhere in the world using a Raspberry Pi and Bluetooth

2. How to use Cursor’s agent view to brainstorm a hardware project

3. What belongs in a personal API and why agents, not just humans, will be the ones using it

4. How to read just enough code to do some damage, without needing to understand all of it

5. Why building for fun, not practicality, is the fastest path to actually shipping physical projects

Brought to you by:

Firecrawl—Power AI agents with clean web data

Customer.io—Build customer engagement campaigns from a single prompt

In this episode, we cover:

(00:00) Intro

(02:00) Maddie’s AI pill moment

(03:53) The thermal receipt printer: live demo and how it works

(11:10) The pager project

(17:23) Why she uses Cursor’s clean agent view instead of terminals and browsers

(19:05) The personal API: coffee order, pets, favorite snacks, and more

(22:57) Lightning round and final thoughts

Tools referenced:

• Cursor: https://www.cursor.com/

• Lovable: https://lovable.dev/

• Raspberry Pi: https://www.raspberrypi.com/

• Resend: https://resend.com/

• Cloudflare Workers: https://workers.cloudflare.com/

• Supabase (Conduct database referenced): https://supabase.com/

• Twitter/X API: https://developer.x.com/

• Spoke pager network: https://www.spoke.com/

• OpenClaw: https://openclaw.ai/

Where to find Maddie Reese:

Website: https://maddiedreese.com

Message her directly: https://maddiedreese.com/message

X: https://x.com/maddiedreese

Where to find Claire Vo:

ChatPRD: https://www.chatprd.ai/

Website: https://clairevo.com/

LinkedIn: https://www.linkedin.com/in/clairevo/

X: https://x.com/clairevo

Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email jordan@penname.co.