Why I build things with Claude Code on weekends

There's a moment in every executive's career where you realize you've become a consumer of technical work rather than a producer of it. You review designs. You approve architectures. You ask questions in meetings. You stop writing the code, the SQL, the model — and you manage people who do.

I didn't want that to be permanent.

What changed

About two years ago, I started using Claude Code on weekends to build things. Not to prove a point. Not as a hobby. Because I had real problems I wanted to solve, and I could suddenly close the gap between "I have an idea" and "this thing works" in ways that weren't possible before.

The first thing I built was a rough prototype of what became Samani.ai — an AI interview prep platform. I'd been thinking about the problem for a while. The gap between how people prepare for interviews and how interviewers actually evaluate them is enormous. I had opinions about how to close it. With Claude Code, I could build a working version of my hypothesis in a weekend.

That was new. That feeling — of being able to close the loop between thinking and building — was genuinely new.

What I actually do

I'm the Head of Data & Analytics at Vanta during the week. That work is technical — I'm building MCP-based self-service systems, LLM evaluation frameworks, semantic metrics layers — but it's deeply organizational too. I'm setting strategy, running metrics forums, managing teams, working with the CEO.

On weekends, I build alone. I'm back at the keyboard. No meetings, no stakeholders, no approval chains. Me and the problem.

What Claude Code does is collapse the distance between intent and artifact. I don't have to remember the exact syntax for a Next.js App Router layout, or the API shape of Supabase's JavaScript client. I can focus on what I actually care about: the product logic, the user flow, the data model. The agent handles the rest and flags when it needs judgment calls.

The four products

Over the past couple of years I've shipped four products:

Samani.ai — AI interview prep platform. Candidates practice with AI interviewers calibrated to specific roles. Recruiters can configure custom criteria. The evaluation engine was the hardest part — building something that gives useful signal, not just generic feedback.

Kate — AI phone answering for frontline businesses. Restaurants, service shops, clinics. Businesses that live and die by the phone but can't always pick up. Kate answers, books appointments, routes inquiries. This one forced me to understand voice AI at a level I hadn't before.

Coach Samani — School exam prep tutor. My kids are at the age where this matters. Building it gave me a visceral appreciation for how hard it is to get adaptive difficulty right.

DataMod Ventures — AI-native investment research platform. This started as a personal tool and evolved into something I actually use to track deals and develop theses.

All four are active. All have real users. None required me to hire anyone.

Why this matters for people like me

There's a version of seniority in tech where you get further and further from the actual work. Your leverage comes from organizational capital, pattern matching, strategic judgment — not from building things yourself.

I think that's a trap. And I think AI coding tools are a rare opportunity to escape it.

Here's what I mean: the most valuable thing a senior practitioner can do is apply judgment. The judgment about what to build, what problem actually matters, what the data is telling you, what the right abstraction is. For most of my career, that judgment was bundled with implementation work I had to delegate. Claude Code separates them. I can have the judgment and produce the artifact.

That's not just useful for side projects. It changes how I work at Vanta too. When I build MCP tools or write evaluation frameworks, I'm not just speccing things out for engineers. I'm often writing the first version myself. That changes the quality of the conversation. It changes what I can ask for. It changes what I catch.

What I'd tell other senior practitioners

If you're a VP of Engineering, a Chief Data Officer, a Director of Product — whatever the title — and you've drifted away from hands-on building, the barrier to getting back in is lower than you think.

You don't need to be current on every framework. You don't need to know the latest Tailwind utility classes or how to configure a Turbopack build. You need to be able to describe what you want, evaluate what you get, and make judgment calls when the agent asks for them.

That's something you're already good at. You've just been applying it to people instead of tools.

The site you're reading right now was built this week, in VSCode, with Claude Code. That's the demo.

---

Aamod Sathe is Head of Data & Analytics at Vanta and the founder of Samani.ai, Kate, Coach Samani, and DataMod Ventures.