Vol.052 — Week 3 of Going AI Native: The Hard Part Isn’t Building the Skill. It’s Caring Enough to Make It Right.
Date: 2026-06-26 | Newsletter
Key Summary
In Vol.052, the third dispatch from Functional AI Partners’ public “AI Native” project, Zenta reports that the four domain experts at DCC Corporate Services — his Singapore accounting firm — each built their first Claude skill this week: one sorts physical expense receipts into a categorized claim, one turns a supplier invoice into a ready-to-key payables voucher, one prepares month-end adjusting journals, one reconciles the fixed-asset schedule against the books. Read closely, he writes, you stop seeing “automation” and start seeing judgment written down — the expense skill knows a meal is only claimable when the client’s name is on the receipt; the journal skill carries one client’s quirky single-accrual convention. None of it was written by an engineer, but by the people who do the work, two weeks after first touching an LLM. The week’s hurdle is the part token-sellers skip: building a skill is the easy half, and a skill that runs is not the same as a skill that is right. The vendor sells the five-minute build, never the slow, unglamorous fine-tuning against messy real files. The moment of the week was human: one data-entry accountant checked her finished skill against the original paper, found two receipts the AI had missed entirely, and — though fine-tuning was Week 4’s job — went straight back in and started teaching the skill, a week early, on her own, because being approximately right was not acceptable to her. That, Zenta argues, changes who should build your AI: not whoever is most technical, but whoever cares most about the outcome being correct. An agent is maybe one-fifth tool and four-fifths encoded judgment plus the will to keep correcting it — and both already live in the person who has done the work for ten years. He is honest about the limit: nothing is in production, the skills are uneven, and the hard wiring to live systems comes in Week 4 and beyond. Do it simple. Anyone can build a skill in an afternoon. Only someone who cares about the answer will make it right.