Twelve hours of negotiation. Three EU institutions in the room. One unresolved annex on conformity assessments for AI embedded in regulated products — medical devices, industrial machinery, connected cars. That was enough to collapse the entire Digital Omnibus trilogue on 28 April.
The AI Act's high-risk obligations are still legally due on 2 August 2026. The Omnibus was supposed to push that deadline to December 2027. Now the postponement is uncertain. A follow-up trilogue is scheduled for mid-May, but for the delay to take legal effect before August, negotiators need a political agreement, a Parliament vote, Council endorsement, and Official Journal publication — all in roughly ninety days.
That is not a comfortable margin.
The instinct in most organisations right now is to wait. If the deadline might move, why spend resources preparing for something that could land eighteen months later? The logic feels sound until you examine what happens in both scenarios.
If the Omnibus passes and the deadline shifts to late 2027, organisations that designed their governance early have a structural advantage. They spent months operating under a framework. They found the friction points. They know where their workflows break under scrutiny, and they fixed them before it mattered. The extra time becomes refinement time, not scramble time.
If the Omnibus fails and August holds, organisations that waited are facing immediate compliance obligations they have not prepared for. The IAPP's analysis of the trilogue failure is blunt: companies relying on the extended timelines would suddenly face requirements for which many have not adequately prepared.
In either scenario, the organisations that designed their governance come out ahead. The ones who waited lose — they just lose at different speeds.
This is not a new pattern. Regulatory uncertainty has been the default posture for AI governance since the AI Act was proposed in 2021. Timelines shift. Definitions evolve. Scope changes through amendments. The organisations that treated compliance as a date on a calendar — something to prepare for at the last moment — have been caught off guard repeatedly. The ones that treated governance as an operating layer embedded in their workflows have absorbed every shift without breaking stride.
The DLA Piper analysis of the Omnibus makes the underlying problem visible: the delay was proposed because EU member states were not ready. Standards were not finalised. Competent authorities were not established. The infrastructure for enforcement was incomplete. The regulation exists. The readiness does not.
That gap — between regulation and readiness — is precisely where governance design sits. Not the legal text. Not the compliance checklist. The design of how AI actually operates inside a workflow, who decides what it can do, what happens when it fails, how the rules connect to the work.
The conformity assessment dispute that killed the April trilogue was itself a design question. Should AI embedded in a medical device be assessed under the AI Act's framework, or the existing product safety regime, or both? The disagreement was not about whether oversight matters. It was about how oversight should be structured when the same system crosses multiple regulatory boundaries.
That is the question Wyrework exists to answer — not for regulators, but for the teams building with AI right now. When a workflow includes an AI agent, the governance design question is not "which regulation applies?" It is "how does this workflow operate responsibly regardless of which regulation applies?"
If the answer depends on a specific regulatory timeline, the governance is not designed. It is deferred.
The next trilogue is approximately 13 May. The Cypriot Council Presidency has until 30 June to close the file. After that, Ireland takes over. The calendar is tight and getting tighter.
But the calendar is not the point. The point is that the organisations treating governance as a date — waiting for the rule to land before designing how to follow it — are making the same mistake whether the date is August 2026, December 2027, or sometime next decade. Governance design is not compliance preparation. It is how work operates. And that does not wait for anyone.
One workflow at a time.
Sources: The Next Web, "EU and Parliament fail to agree on AI Act changes," 28 April 2026. IAPP, "AI Act Omnibus: What just happened and what comes next?" April 2026. DLA Piper, "The Digital AI Omnibus: Proposed deferral of high risk AI obligations under the AI Act," 2026. Modulos.ai, "EU AI Act Omnibus: The Trilogue Failed," April 2026.