What this series is
A small series of opinionated hot takes on AI engineering, drafted with LLM assistance and verified by hand. The index post.
AI-assisted postDrafted with help from Claude, edited and fact-checked by Mart. See transparency policy →What this is
A small series of opinionated hot takes on AI engineering, drawn from personal experience and from reactions to the research, articles, and public discourse the field keeps generating. The posts share a structural argument: most "AI safety" stories shipped in 2026 are architecturally weak, and the symptoms everyone is watching in public, from the leaked Apple system prompt to the strawberry-R problem to the car wash test, are consequences of that weakness rather than jokes about model intelligence. Deterministic verification systems are not going anywhere, and the cleanest path through the next five years takes them seriously.
The first post is up: Telling an AI not to hallucinate is like telling a person not to make mistakes.
How the posts get made
The drafting workflow is openly LLM-assisted. The thesis, the angle, and the anchor examples are mine before any model is involved. Claude helps me converge loosely connected thoughts into a structured essay, surface citations I need to verify, find contradicting research I have not seen, and tighten the prose. I then read every cited source, remove anything I cannot personally defend, edit the prose by hand, and publish.
The verification step that protects the output is the same shape as the argument these posts make: a separate, slow, deterministic pass in front of a probabilistic generator. Every AI-assisted post carries a pill in the header linking to the transparency page, where the workflow is documented in plain English.
What to expect
Long posts. Specific numbers. Cited sources. Strong takes, framed to be falsifiable. No career-milestone announcements, no listicles, no thought-leader voice. The pace will be irregular.
A rough queue:
- "AI writes the PR, AI reviews the PR" as the most expensive bad pattern currently shipping
- Vibe coding tech debt as a measurable phenomenon
- Why every "agentic" framework eventually looks like LangChain under the hood
- Where deterministic tooling holds a structural advantage over probabilistic generation, by principle
If you read a piece and think it is wrong, the useful response is to tell me where. If a counter-piece exists or gets written, link it. I will read it.
Read next
The cloud was a network-diagram icon for two decades before it was a product. Compaq coined cloud computing in 1996, AWS made it real in 2006, and the surrounding vocabulary — VPS, hyperscaler, colocation — each carries its own history.
The .yml extension is a 1990s DOS artifact. The 'YAML Ain't Markup Language' acronym is a 2002 self-correction. Both questions resolve cleanly once you know markup languages and data serialisation formats are different categories with different ancestors.
A short tour of database name origins. Children, code names, predecessors, acronyms, mythology, and at least one pet duck on a houseboat. Most of the history of the field is hidden inside the names themselves.