Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates and primary reference.
A countdown is good at creating urgency and bad at explaining what to do next. Once the date changes, the useful question is whether the team has built anything that survives the countdown.
Avoid the last-minute evidence scramble
Record evidence while the work happens. A test result should identify the version, conditions and outcome. A design decision should identify the choice and the reasoning. Reconstructing those details months later turns normal product work into archaeology.
An illustrative release review can ask for the current evidence rather than a newly polished summary. If the test belongs to an older configuration, say so. If the decision was never recorded, reconstruct it transparently instead of pretending the record existed at the time.
Preserve the difference between an observed result and an explanation written afterwards. Both can be useful, but they answer different questions. A retrospective narrative cannot prove that an unrecorded control operated during the original event.
Keep the record proportionate to the decision. The aim is not to document every keystroke. It is to retain enough context for someone else to understand why a consequential choice was made and what supported it.
| Decision | Evidence to request |
|---|---|
| At design time | Purpose, boundary and rationale |
| At test time | Version, conditions and result |
| At review time | Decision and outstanding uncertainty |
What to do with this
If the deadline disappeared tomorrow, this evidence would still help the next engineer, reviewer or customer understand the product. That is a better test of usefulness than the size of the compliance folder.
Record the decision, the responsible owner and the evidence still missing. The decision-record example shows a lightweight way to do that. Where a claim concerns system behaviour, connect it to evaluation evidence. These are proposed product practices, not a finding that a system complies with the law.
Primary timeline reference: European Commission, AI Act. The operational recommendations above are my analysis.
