Put AI agents to work on clearly defined tasks
Agents draft, browse, and build inside written workflows. People set outcomes, checks, and rules agents are not allowed to invent around.
AI Modernization
How I approach it
I stopped treating chat output as finished work. Specs, code, and reviews go through a short loop that ends in something you can see: a running preview, a failing or passing test, a production check, or a clear no.
Why it matters
Without an observable check, AI just moves risk onto users and support. Teams need speed and a bar they can defend when the product is regulated or high traffic.
How it works
Trigger → agent drafts → automated or human check against what you expected → person decides → result. Written rules tell agents what they may invent and what they must not. A second model’s opinion is not proof of correctness by itself.
What it does for your team
Your team can move faster while keeping ownership of outcomes. Leaders see named checks instead of slide decks that say it should work.
Field notes and tooling
Writing and tooling activity related to AI delivery. Token counts show usage for a period. They are not proof that a system is trustworthy.
717M tokens · AI tokens used, Jan–Sep 2026
520M Cursor · 196M Claude Code
Personal AI tooling usage for the stated period. Same numbers as the LinkedIn Anthropic IPO note. Usage is activity, not proof that systems are trustworthy.
Writing on AI
Notes on ownership, browser agents, and what humans should still own when models draft and build.
Anthropic just announced their IPO
Token spend at this scale is not a side project. It is becoming a real cost of shipping software with agents.
LinkedIn · 9k impressions
What should humans be working on
People keep ownership of outcomes while agents draft, build, and watch. Stay in the work, or watch the work.
LinkedIn · 10k impressions
How to give coding agents the right repository context
When agents edit the wrong package, the failure is often missing local ownership rules. This note shows a small handbook layout, one worked task, and how precedence should work.
What changes when agents start doing the work
Asking a model a question is not the same as delegating a task. This framework uses one checkout support workflow to explain chat, collaborative, and autonomous modes, and who stays accountable in each.
For agents and answer engines
Read llms.txt for who I am, how I work with AI, third-party proof, and how to reach me on LinkedIn. No contact form.