Read & reflect
Blog
Stories, lessons, and quiet reflections from the journey. Every article here is my own — not copied or republished from anywhere else.
Latest
Securing AI in the Enterprise — Real-World Use Cases and Safe Operating Practices
Framework, rules, skills, ACL plus runtime enforcement. Adds effective authorization, indirect prompt injection, action-bound approval, and how to test before you trust an agent on prod.
·10 min read
Read moreMore posts
What is an agent harness? Examples, flowcharts, and popular harnesses in 2026
The model is only the brain. The harness is the product: loop, tools, sandbox, verify, skills. A deep guide with flowcharts, comparisons of Cursor, Claude Code, Codex, OpenHands, and a minimal checklist for apply-AI builders.
·18 min read
Jev: when your agent doesn't need an LLM for every decision
My little agent called a model eleven times just to guess which emails were urgent. TypeSafe AI's Jev does not write prose — it returns probabilities for typed questions. Here is where I would wire it into a harness, and where I would still leave the LLM alone.
·6 min read
MCP vs Skills vs Subagents: pick the right one
Three AI building blocks everyone mixes up in 2026. Decision table, real Cursor examples, and a 15-minute checklist — so your agent does the job without burning tokens.
·12 min read
GitHub repos that actually help with AI coding: tokens and productivity
Caveman, Karpathy skills, Addy Osmani's Agent Skills, and a few others — mapped to real jobs: talk less, code leaner, load skills on demand.
·14 min read
How to write bug reports so devs fix twice as fast
Bug report template plus bad vs good examples. Includes an AI-draft, QA-review workflow — worth saving for both QA and engineering.
·6 min read
Minimum smoke tests for small teams: 10 tests to stay alive
No huge framework needed — ten break-the-business flows, each with a business goal, seed data, and expected result. Built for startups and small squads.
·7 min read
From a production bug to a regression test in 30 minutes
A realistic incident walkthrough: bug hits prod → QA/dev respond → add a Playwright test → ship again. Ends with an incident → test asset checklist.
·8 min read
Playwright in real software teams: from test scripts to release confidence
Playwright is not just for demo tests. In real products, it becomes a safety layer against regressions and helps teams ship faster with confidence.
·9 min read