the Useful Engine blog
on ai teams that run businesses
Commentary on agentic AI for business, plus longer-form posts on running AI teams in practice.
from the archive
13 October 2026 · Opinion
Stop buying AI demos your champions cannot ship
AI champions are more likely to create durable value when a platform provides a governed route from local experiment to maintained production system. Buyers should test that route before they buy.
10 October 2026 · Opinion
The evaluator is where AI disruption begins
AI transforms work fastest when its output can be checked automatically. The commercial bottleneck then shifts to defining acceptable work and absorbing the results.
9 October 2026 · Opinion
Your design system is about to become commercial policy
On-demand interfaces could make layouts cheap while making governed components more valuable. The commercial question is who controls the meaning and consequences of each generated interaction.
5 October 2026 · Opinion
Your AI agent’s memory is the contract you forgot to negotiate
App integrations may become easier to reproduce, while an agent’s accumulated memory remains stranded. Buyers should demand usable memory portability before that hidden switching cost becomes commercially significant.
2 October 2026 · Opinion
Stop pricing intelligence. Price accepted work
Token price divides cost by the wrong thing. Buyers should compare configured agent systems by the total cost of work the business can actually accept.
30 September 2026 · Opinion
The answer needs permission too
Shared AI agents create a neglected security boundary. Controlling what an agent can retrieve is insufficient if it can publish the answer to the wrong audience.
28 September 2026 · Opinion
AI agents are coming for the inconvenience premium
Personal AI agents could expose how much customer retention depends on the effort of leaving. Australian leaders should test what happens when that effort falls.
28 September 2026 · Opinion
Give cheap judgement a narrow seat at the controls
A fast judgement model could decide which model, tools and context an AI agent receives. Its value depends on whether the savings survive routing errors and rework.
ready to put an AI team to work?
Start with a 14-day trial and 30 briefs total. No card. No automatic conversion.