Taj Pelc

pronounced: Thai Pelts

I make AI ship production code. Even in 16-year-old codebases.

15 years shipping software, leading engineering teams since 2016. The thing is, the model isn't your bottleneck. The harness around it is. I build harnesses that let AI agents verify their own work. Then I install them in software teams.

What I think about

AI-augmented engineering

The model is the same for everyone. The harness isn't. A team with a verification loop outships a team without one.

Building products, not just systems

Users, pricing, retention. Not just code. Technology serves the business outcome.

Shipping velocity

Ship, learn, iterate. The elegant architecture comes after product-market fit.

Selected writing

Nov 19, 2024

Code generation is changing my approach

AI writing boilerplate forces better architecture. Complexity grows, AI struggles, you refactor. Built something in a week that would take a month before.

Jul 3, 2024

Optimize for going from idea to production

Merge product and execution teams early. Prototype and test with users during ideation. Optimize for idea-to-production speed.

Jan 3, 2025

Shipped is better than perfect

Until you have users, be bold not careful. Learn through playtest feedback loops, not isolation. Ship, get feedback, improve, repeat.

236 posts on engineering leadership

I'm writing again. New posts on AI-augmented engineering, shipping with tiny teams, and building products.

Work with me

I take a small number of AI-enablement engagements. Hands-on, with your team, in your codebase. Harness, tests, agentic loops. Currently open.

taj@pelc.cc