I haven’t written code in more than 25 years.

TL;DR

I haven’t written code in more than 25 years. This week I tried Claude Cowork — agentic AI for non-coders — and shipped a working Chrome extension over lunch, then PDF upload support that afternoon. Agentic AI isn’t a chatbot. It’s a system that takes intent and executes. The new core skill is expressing outcomes precisely enough that an agent can execute them, and reviewing the result like an owner.

I haven’t written code in more than 25 years.

As a teenager, I taught myself to program and even made some money selling my own software. Then I studied biotech, went into pharma, and coding became something other people did.

Fast forward two decades. I’ve tested every productivity tool under the sun. Same outcome every time: commercial software is never quite your workflow. There’s always friction, always compromise.

So when AI started making software-building accessible again, I tried something simple: build what I actually need.

Here’s the itch: how many times have you been in a conversation thinking “I’ve just read about a company that does exactly that, but where?”

To solve that, I’ve been creating a personal knowledge database, a “second brain” that can surface insights from what I read and hear and find relevant: LinkedIn, STAT, X, podcasts, conferences, PDFs. Until now, it all lived across Notion and handwritten notes, relatively easy to store, painful to retrieve, and hard to make sense of.

I started hacking on it with Claude in the evenings. It worked, but it was slow. Lots of copy-pasting, no real flow, progress measured in weeks.

This week I tried Claude Cowork, Anthropic’s brand new research-preview version of Claude Code for non-coders.

Wednesday morning I thought: “I need a Chrome extension to quickly capture web content.” One prompt, I went to lunch, came back, and it worked.

That afternoon: PDF upload support. One prompt. A few minutes later, done, without me touching a single button.

That’s what “agentic AI” actually means to me. Not a chatbot answering questions, but a system that takes intent and executes: plans the work, creates the artifacts, tests, fixes, repeats.

In a world where software agents can draft the email, build the deck, run the analysis, file the paperwork, triage the inbox, negotiate the calendar, and yes, write the code, the bottleneck becomes the human side: clarity of intent, good judgment, and most of all knowing what “good” looks like.

The new core skill is learning to express outcomes precisely enough that an agent can execute them, and to review the result like an owner, not a passenger.

Key takeaways

• Commercial software is never quite your workflow. There’s always friction, always compromise.
• Agentic AI isn’t a chatbot answering questions. It’s a system that takes intent and executes: plans, builds, tests, fixes.
• One prompt, one lunch break, one working Chrome extension. The speed change is the story.
• The bottleneck shifts to the human side: clarity of intent, judgment, and knowing what “good” looks like.
• The new core skill: express outcomes precisely enough that an agent can execute them, and review the result like an owner, not a passenger.