AI That Stays in Its Lane
The premium is a machine that cannot cross the line into the judgements that are yours to make.
The unease isn't only about privacy
What people fear is a machine deciding things it has no business deciding.
- An off-switch answers surveillance — and nothing of overreach.
- On a Big Tech stack there is no lane, only a model that reaches as far as it's allowed.
- The organising question: which decisions must never leave human hands?
The line that isn't a matter of taste
Some judgements can be systematised; values, meaning and agency cannot.
- Automating a value smuggles in someone else's value as the default.
- The boundaries, plainly: values verified not automated; agency respected not simulated.
- The decisions a community most needs help with sit right on this line.
It carries what can be carried, and stops at the rest.
A boundary you only hope for is not a boundary
A guardrail written as a wish doesn't constrain what the system actually does.
- Most 'responsible AI' is a sign asking you not to cross, not a fence.
- We built the line as architecture: reserved decisions are gated to humans by the code path.
- A check runs before an action takes effect — and can't be talked out of its job by a prompt.
What the machine does instead
It proposes; a person disposes — and the seam between them is visible on purpose.
- Surfaces history, summarises threads, drafts routine text — real work taken off the plate.
- Stops at the judgement itself; never crosses into what's reserved.
- Every assist is logged: what was used, which model, where it ran.
The bound is what lets you bring it close.
Why bounded is the more useful machine
A machine bounded by construction can be trusted close, not kept at arm's length.
- The bound is what lets you relax and hand it the tedious work unsupervised.
- We don't claim it out-thinks the biggest models — we claim a better arrangement.
- A community needs a capable assistant it can trust not to govern it. That's shipped.
Who needs the boundary most
The people who carry responsibility they cannot delegate.
- Boards, where a decision blamed on 'the system' is a decision no one is accountable for.
- Clinicians and caseworkers, where overreach is harm, not inconvenience.
- The principled — who wanted not the machine's absence but the certainty it stays in place.
The principle the rest of the series stands on
Knowing its place is the design discipline that makes everything else safe to own.
- The unrewritable record works because the AI can't author a director's reasons.
- The in-house model is governed by the same line about what it may decide.
- A system worth trusting knows which of the two it is: instrument, not authority.
You don't have to choose between Big Tech AI and no AI. You can hold your own.