Vol.062 — Week 8 of Going AI Native: The Proof Worked. Then We Asked If We Still Need the Software at All.
Date: 2026-08-08 | Newsletter
Key Summary
In Vol.062, the eighth weekly dispatch from Functional AI Partners’ public “Project AI Native,” Zenta reports the first true end-to-end proof — held this week rather than last Friday because he was on business in Japan. Three of the DCC Corporate Services data-entry accountants each took a single client and produced a complete set of management accounts without bending the work to fit the firm’s thirty-year-old accounting software. The expense-claim accountant broke the close into four skills, each a checkpoint whose output is verified before it feeds the next, carried forward the prior month’s data, and came back 98% accurate — “that is exactly how we submit to our clients.” The supplier-invoice accountant judged her simpler client needed no steps, ran one skill month by month, and, untaught, reinvented an agent loop — run, check, adjust, run again — above 95%. The journal-entry accountant brought back a leaner single-skill run. All three reached for engineering patterns nobody taught them, because they are the ones whose hands are in the work. The honest cost: to get depth they un-standardized — three bespoke one-offs, not a business yet, with the scale still to win back next week. And a question a month ago unthinkable is now live: do they still need the software at all? Do it simple. Let the people whose hands are in the work build the tools – and be willing to question even the software you have trusted for thirty years.