Field notes in agentic engineering

What building real software with agents actually looks like.

I’m an AI/ML engineer and tech lead — ten years of production software. I build complete systems with frontier models: backends, deployments, monitoring, infrastructure — and publish the whole record, including the roadblocks that never make it into demos.

Read the build logs → Get the digest

From the workshop: Analytics Without Cookies, a Banner, or Losing Half the Data to Ad-Blockers · 21 Jul 2026 →

Click to flip

The record

Build logs

Analytics Without Cookies, a Banner, or Losing Half the Data to Ad-Blockers Feature21 Jul 2026 How I run PostHog on this site: cookieless, no consent banner, EU-hosted, and reverse-proxied through my own domain so ad-blockers barely touch it. The real Cloudflare Function and client config, both liftable. I Built a Fintech Startup in 2020. What AI Actually Changed Case study20 Jul 2026 In 2020 I launched a fintech stock screener — three years live, 2,000+ users, built by hand. I rebuilt one with AI agents this year. Here's what changed. Can You Use yfinance Commercially? No — Here's the License Research Roadblock16 Jul 2026 Is yfinance data free to use commercially? No — here's the primary-source license research, done before monetizing was ever on the table, plus the 50-KPI stock screener it protects. My OpenSpec Workflow: Spec-Driven Development With AI Agents Tutorial16 Jul 2026 The OpenSpec workflow I use with AI agents: spec, adversarial plan review, isolated build, red-team code review, then verify. Real PR numbers inside. Python Writes, TypeScript Reads: One SQLite File, Zero Drift Feature16 Jul 2026 Python for ingestion, TypeScript for the app, one read-only SQLite file as the seam — and one JSON Schema generating matching Zod and Pydantic types.

Browse Features & architecture Tutorials & workflows Case studies Roadblocks

Projects

Selected work

QuantScreen

My quant equity screener — point-in-time discipline, 503 tickers, ~50 KPIs. Every PR planned by spec and adversarially reviewed; the full build is the flagship case study.

Live · read the build overview →

Mage Survivor

A bullet-heaven incremental RPG I’m building the same way — different domain, same discipline, same public record.

In development

The Course

Rigorous agentic engineering, taught from the QuantScreen build as it actually happened — dead ends included. No launch date I’d have to invent.

In production →

Correspondence, not a feed.

AI & Tech distilled into one considered digest, plus every build log first. Used for exactly one thing; unsubscribe works immediately, always.

Signup lands here once it’s wired to the list — not before. Meanwhile the RSS feed carries every build log, and the newsletter page has the details.