AI patterns.
AI in Aceve products is an extra apprentice: it handles the paperwork so the skilled work stays human. It is embedded in the workflow, opt-in before default, and it always shows its receipt. These patterns are how that promise becomes pixels.
The frame
Why “apprentice” and not “assistant”.
Embedded, not chatbot-forward
For a master electrician, an AI that replaces judgement is a threat; an apprentice who handles paperwork is a gift. AI surfaces help at the moment of need inside the existing workflow — pre-filled fields, flagged anomalies, drafted texts — not from a chat window that interrupts. An assistant that is everywhere and always talking is an annoying colleague.
Trusted because it's checkable
Our users are accountable for regulated, money-bearing work, so the goal is calibrated trust — rely on the suggestion when it's right, override it when it's wrong. That only works when output is verifiable: sources shown, uncertainty visible, and a way back from every action. Trust that can't be checked is just persuasion.
Earn autonomy
Suggest → draft → act. One rung at a time, never skipping.
AI proposes; nothing changes until the user acts. The default for every new AI capability.
AI produces a first version — a quote line, an invoice text — that the user edits and approves.
AI completes a step on its own, within explicit boundaries the user set.
The same staging applies to rollout: every AI capability ships opt-in, becomes recommended after evidence, and only then becomes a default — always with a way back. See principle 06 and 08 on Design principles.
The AI layer
Seven patterns, in build order: controls → receipts → logs → undo.
Controls
Start, stop, pause. The user decides when the apprentice works — AI never runs without a visible switch.
Receipt card
PlannedEvery consequential suggestion shows its work: which documents it read, what it is and isn’t sure about, and a confirm step. No receipt, no action.
AI disclosure label
PlannedAI-generated content is marked as such — visibly for people, machine-readably for systems. Required, not optional.
Confidence indicator
PlannedUncertainty is shown in words a tradesperson can act on — never a bare “87%”. Low confidence looks different from high confidence.
Activity log
What the AI did, when, and on whose approval — reviewable after the fact. The audit trail is a feature, not a compliance chore.
Undo / rollback
Every AI action can be reversed. If a workflow can’t offer undo, it doesn’t get past the suggest rung.
Escalation to human
A clear path from “the AI is handling this” to “a person is handling this” — for the user’s judgement, not as a failure state.
The three marked planned are tracked on the component coverage board.
Writing for AI features
The voice rules apply — outcomes, plainly.
EU AI Act — a design standard, not a legal afterthought
Article 50 transparency obligations apply from 2 August 2026.
Users must be told when they are interacting with an AI system, and AI-generated content must be marked in machine-readable form. With customers across the Nordics, DACH and Benelux this is a product requirement for every Aceve surface. The disclosure label and receipt patterns above are how we comply by design — the law and the trust strategy point the same direction.