Notes from building
with AI in practice.
Production AI systems, agentic software development and what the work feels like while it changes.
Oct 15, 2026 · ai · claude · agents · workflow
How to Audit Your Claude Code Instructions After a Model Upgrade
After moving to Opus 5.5, my Claude Code instructions went 22k → 13k tokens. Run the prompt audit, review with Claude, then test what you removed.
Oct 11, 2026 · ai · claude · agents · workflow
Claude Code Permissions and Hooks: Rules Your Prompt Can't Enforce
"Never push without asking" belongs in settings.json, not only in the prompt. How to set up Claude Code permissions and hooks, with real examples.
Oct 6, 2026 · ai · code-review · agents · workflow
AI Code Review Is Statistical: Why One Run Is Not Enough
Run an AI code review twice on unchanged code and the findings can differ. How I'd design for it: repeated runs, fresh context, tracking what each adds.
Oct 3, 2026 · ai · claude · agents · workflow
Graph Engineering for Coding Agents: A Simple First Workflow
Graph engineering for coding agents sounds complicated. A simple first workflow in Claude Code: draw it, write three agents and one SKILL.md.
Sep 30, 2026 · ai · agents · adoption · workflow
An Agent Marketplace Solves Distribution, Not Adoption
A marketplace ships agents and skills to every developer. It does not make anyone use them. Six questions I'd ask once the technical part works.
Sep 27, 2026 · ai · agents · claude · codex · workflow
One Source, Multiple Harnesses: Building a Portable Agent Marketplace
I run the same agents in Claude Code, Copilot CLI and Codex from one source. How the build adapts packages, hooks and roles, and what broke on the way.
Sep 17, 2026 · ai · claude · agents · workflow
How to Build a Personal Agent Marketplace for Claude Code
40+ agents and skills, five plugins in every project. How I package them as a Claude Code plugin marketplace, with my aw-review reviewer as the example.
Sep 9, 2026 · ai · strategy · legal · adoption
What a software factory taught me about AI strategy for law firms
I lived it in software: AI agents do the work and the advantage moves into the harness around the model. Now it is legal work. Nine AI strategy decisions.
Aug 12, 2026 · ai · code-review · codequality · programming · softwareengineering
If AI Writes and Reviews the Code, What Is the Human Approving?
AI agents write the code and independent AI code review gates it on evidence. What a PASS establishes about AI-generated code, and what a human approves.
Jul 23, 2026 · ai · adoption · finance · risk-management · compliance
What Oma Säästöpankki measured after four weeks of AI coaching
A four-week AI Coaching Program at Oma Säästöpankki. The pilot lead reported 15–50% time savings; one regulatory task went from four hours to minutes.
Jul 22, 2026 · ai · agents · workflow
AI Optimized My Throughput. It Did Not Optimize Me.
My AI agents raised my output more than 10x, and my workdays got heavier, not lighter. AI brain fry, decision fatigue, and becoming the bottleneck.
Jul 7, 2026 · ai · learning
AI took my job. Twice in one year. I still have one.
What a year of coding with AI did to my work, my role, and me. AI took over the writing, then the checking, and I had to relearn my role three times.
Jun 30, 2026 · ai · claude · architecture
5 months. 100% Claude Code. Zero architectural drift.
Every line of code in a five-month web service project was written by Claude Code. Here is the system that kept the architecture from drifting.
Jun 2, 2026 · ai · agents · claude
AI Has Made Me a Text Parser. And I'm Terrible at That.
Claude writes a ten-page spec in a minute. Reading it takes me two hours. The fix was giving the human a ViewModel instead of a database dump.
May 11, 2026 · ai · agents · workflow
FOMO, FOMAT, FOMUT, and the Bottleneck That Turned Out to Be Me
A year ago the AI agents were the bottleneck in my workflow. Now they run for hours unattended, and the bottleneck is me. Why that is a better problem.
Mar 16, 2026 · ai · testing · agents · security
How I Validate Quality When AI Agents Write My Code
The eight-gate quality pipeline of tests, validator agents, reviews and deploy verification that turns AI-generated code into evidence.
Mar 1, 2026 · ai · agents
I'm not working with a tool anymore
Eija Moisala asked me one question for her Ask a friend series: what have you been experimenting with, and what did it teach you?
Feb 7, 2026 · ai · agents · performance
Your AI Coding Agents Are Slow Because Your Tools Talk Too Much
A validator agent spent 6-9x longer reading test results than the tests took to run. The fix was not a better prompt. It was quieter tools.
Nov 20, 2025 · training · teams
Four training sessions with the Nokia team in Oulu
Back in Oulu with the same Nokia team: four sessions across half a year, and the one question that shows whether training actually stuck.
Nov 3, 2025 · ai · training · adoption
Turning AI into real results: what Elisa got right
What Elisa and a team led by Eija Moisala got right about AI adoption: leadership, permission, and making learning part of everyday work, with results in weeks.
Oct 20, 2025 · ai · training · adoption
Bringing AI to everyday work: speaking at Ilmarinen
A recap of my talk at Ilmarinen on making AI a real teammate at work: moving past pilots, into everyday workflows, and a culture that keeps learning.
Oct 7, 2025 · ai · training · adoption
AI training for employees: how to guide AI from pilot to practice
My Digiklinikka talk for the North Karelia Chamber of Commerce: why AI has to be guided, and how teams move it from small pilots to everyday practice.