Sam built a full-stack internal web app in six hours with Claude Code and Supabase without writing code, then tested Claude Cowork live. What we learned.
Yes, someone who has never written code can build a working full-stack web app in about six hours with Claude Code. On Episode 29 of Built This Week, Sam Nadler showed an internal evaluation tool he built in Claude Code and Supabase between meetings, then tested Claude Cowork live. Jordan Metzner's verdict on Cowork versus the terminal: the interface matters far less than using the best tools of the moment.
Jordan opened by pointing out a trend on the show: more and more of the tools we demo are built by Sam, and they go deeper into the code base instead of relying on tools that build everything for you. This week Sam showed an internal evaluation tool. Here we focus on how he built it.
Sam says he did it in Claude Code in the terminal. He started in Cowork and likes it, but says it pretty quickly made sense to keep working in the terminal. The database is Supabase, which he finds very easy to understand. For deployment, he used the stack we already run, Amplify, AWS and GitHub, and says that once it is all connected it is pretty simple to get an app live on a company subdomain.
The timeline: he started on a Monday afternoon and was close to going live by Wednesday. Working in between meetings, he puts the total at about six hours. Every analysis in the app runs on one model. When Jordan asked what powers it, Sam answered: "Opus 4.5 for everything."
I haven't written one line of code in my life, full stack web app development in six hours that has some pretty cool features.
— Sam Nadler, Episode 29
Sam was also clear that six hours is not the end. Users will give feedback and catch bugs. He had pushed a similar product to a different team that Monday and already made three or four bug fixes from live testing.
Watch Sam test Claude Cowork live in Episode 29 (starts at 10:56) · Episode 29 notes and transcript
Cowork had gotten a lot of hype in the days before we recorded. Sam describes it as basically a wrapper around Claude Code and an entry point for almost anyone to start building their own products. At the time, it lived in Anthropic's Mac desktop app, next to chat and Claude Code, with the chat interface people are already used to.
The difference, Sam says, is context. You give Claude Cowork access to a folder, so it can read your documents. He had written a short PRD for a Hollywood transformation trainer: a personal trainer app focused on aesthetic results and progressive overload rather than overall health. He pointed Cowork at the folder and typed "build this." After he told it to check the PRD, it summarized the idea as a mobile web app for upper body hypertrophy.
Then it asked how he wanted to build it. Sam likes that Cowork proposes these trade-offs in a clean UI, though he notes the terminal does the same thing. He picked React and Supabase, and chose a real LLM API for the training responses. Cowork then showed its plan and progress in a more polished view, so you can see what it is doing and where it is in the process. The build is on our Hollywood Transformation Trainer page.
Sam called Jordan a Claude Code terminal guy. Jordan says he has used Cowork and thinks it is a great tool, and he loves that you can dump a bunch of documents into it. But he doesn't think the interface matters much. What matters, in his view, is using the best tools at the time, and right now he thinks those are Anthropic's. Web, terminal or desktop app, it gets you building.
His reason for preferring the terminal is parallel work. With Claude Code you can run as many sessions as you want at once. Sam didn't think Cowork could open multiple windows at the same time. Sam agreed, and described how he works with four or five terminals open on different projects, switching between them to give context and test:
It's literally just kinda managing a team of little Claudes out there, and, like, the pace when you're doing it like that is way faster than kinda waiting for this to go through the progress and the plan.
— Sam Nadler, Episode 29
Jordan also shared an idea for teams: he has a skill that holds the design features and layouts he likes, and Sam can use it to copy them. He expects us to have enterprise skills shared across everyone, so the products people build work together. His last point was about change. Today's interface may not be tomorrow's, so you have to be willing to keep changing how you work to keep up. We compare the two in more depth on our Claude Code vs Claude Cowork page, and we talked about the rise of multi-agent AI in Episode 30.
The news was Anthropic closing a funding round above $10 billion, which Sam said sounded like it was still open. He noted that about a month and a half earlier Google had been on a tear, and in the last 30 to 60 days it was Anthropic's turn.
Jordan said we are shifting more of our spend to Anthropic, and less to Cursor and ChatGPT. Hearing Anthropic say publicly that it is focused on enterprise, not ads, makes him feel good as a CEO, because he expects it to keep building tools to support him rather than tools to market to him. But his loyalty has a clear condition:
And their, I mean, their models are top notch. And at the end of the day, that's all that really matters.
— Jordan Metzner, Episode 29
If Google's Gemini tools became top notch tomorrow, he says, we would probably move there. Sam mentioned that he had heard a premium tier just dropped 50%, and Jordan noted there is now a $100 Max plan alongside the $200 one. Even so, Jordan says, it might be getting cheaper but we are spending more money using it, which he called kind of a gateway drug. His reasoning for sticking with Anthropic for now: coding is everything, and tools like Cowork are wrappers around Anthropic's coding models.
Sam asked how Jordan thinks about this at the level of the whole organization, now that almost anyone can build their own tools. Jordan says everyone is becoming an agent manager. The plan is to give everybody Claude Cowork or Claude Code so they can build tools and automations for their own jobs.
So far, he says, it has not led to any reduction in workforce in our case. It has increased efficiency: the finance team creating invoices faster, and the sales team writing RFP responses and proposals faster, better or more exactly. He expects spend on Claude to rise, along with many more internal tools like the one Sam showed.
His next question is what happens when those tools feed one dataset. As we automate more of them and build a bigger data model with AI on top, he wants to see what insights we can draw across the whole pipeline.
The live Cowork session, the terminal versus desktop debate and our take on Anthropic's funding are all on camera. Watch Episode 29 of Built This Week, where the full transcript is also on the page.
