LeadMastery mode

What EU AI Act Posture Means for a Delivery Team in 2027 — Desk-Level

Most teams can name the Act. Few can name what Friday's paste and recap habits will look like under it.

Ihor NesterenkoIT System Analyst6 min read · Oct 5, 2026
One page. Five lanes. Not a statute digest. AI-generated cover image — Author.
One page. Five lanes. Not a statute digest. AI-generated cover image — Author.

Most delivery leads I meet in 2026 can say the EU AI Act exists. Few can say, in one page, who their programme is under the Act's roles, which AI uses already sit on the desk, which human checks they own before ship, what evidence they can produce without counsel, and when they must stop and escalate. That gap is the posture problem — not a missing statute digest.

This piece is a practitioner horizon scan for 2027. It is not legal advice. It does not tell you how to comply. It ships a formal picture — a Desk Posture Map — so a lead can name desk behaviour before Annex III high-risk rules land and before Compliance Theatre fills the calendar with unread decks.

Compliance Theatre — Why Decks Fail Desk Reality

Call the failure mode Compliance Theatre: a long Act summary, a RAG status, and a slide titled "AI Act readiness" that never maps to the ASN paste, the Copilot workshop recap, or the Silent Accept in Friday's pair review.

If the phrase is new, keep this version: literacy without lanes.

I have watched logistics programmes celebrate "Act awareness" while the junior still pastes client ASN samples into a model, the BA still forwards an AI workshop note as minutes, and nobody can say who verified the overnight-batch rule. The deck looked mature. The desk did not.

Three symptoms show up together. The readiness slide names chapters nobody on the delivery team owns. The RAG turns green because a workshop ran, not because a reject still works under time pressure. The chat history of AI uses lives in private threads the lead cannot inventory.

Anticipate the pushback: "Legal owns the Act; we wait for their policy." Waiting is a choice. While you wait, the fastest paster still legislates. A Discovery Norm Map already binds paste, recap, pack, and escalation as provisional team law. The Act does not replace that map. It asks a harder adjacent question: can you state your desk posture in Act-facing language without pretending you are counsel?

Commission implementation timelines put transparency rules and broad enforcement in 2026, with Annex III high-risk system rules applying from 2 December 2027 under the adjusted path. That date is why "2027" is not futurism for a delivery lead. It is the planning horizon for programmes that already live with AI-assisted discovery.

The Desk Posture Map

Call the practice a Desk Posture Map: one dated page, five lanes, named lead owner, review date. Provisional. Supersedeable by Legal. Not a compliance certificate.

Plain version: role, inventory, oversight, evidence, escalation — written so Friday can obey it.

Lane 1 — Role. For each material AI use on the programme, say provisionally whether you behave more like a provider, a deployer, or a user of someone else's system. Most delivery teams using Copilot for mappings and recaps are not building high-risk AI products — and panic that assumes they are wastes the year. Role honesty cuts complacency too: if you do deploy a high-risk tool (for example an employment-screening assistant), deployer duties are not a slide title. Commission material on deployer obligations for high-risk systems stresses use per instructions, competent human oversight, monitoring, and log retention where logs are under your control. Your Role lane names who on the team must treat those themes as desk questions — without drafting counsel's opinion.

Lane 2 — Inventory. List the AI uses that already touch delivery artefacts: paste into models, workshop summarisers, mapping suggestions, pack drafting, status-update polish. If it is not on the inventory, it is not governed. Inventory is boring on purpose. Surprise uses are how Theatre returns. On the logistics programme the first inventory missed "status polish in Teams" — half the board pack had already been AI-smoothed without a named check.

Lane 3 — Oversight. Name the human checks that must happen before an AI-touched artefact becomes the record or the ship candidate. This is where existing habits plug in: Spoken Accept in live pair review; Claim Sheet Pass for absentees; pack examinations before workshop closed. Oversight is not "AI literacy" as a course badge. It is a named reject that can still fire under time pressure.

Lane 4 — Evidence. Write what the team can already show without Legal: who verified, what was rejected, which prompt was visible, which hedge stayed open. Deployer-side logging themes in high-risk contexts exist for a reason — traceability fails when only the model remembers. Desk evidence is the trail you can produce from tickets and spoken calls. Certifiable legal files stay in Escalation.

Lane 5 — Escalation. Stop conditions that leave the delivery lead's authority: missing Legal position on recording; client data the paste agreement forbids; a use that may be high-risk and nobody has role clarity; a vendor claim you cannot verify. Escalation without a date is theatre. Put a named owner and a next review on the page.

That is the formal picture. Five lanes. One owner. One review date before your 2027 planning freeze. Ugly is fine. If you can delete a lane and the page still "looks complete," you rebuilt Theatre.

One Week to Install

Experience, not a programme.

Day 1. Draft the five lanes in a shared note. Put your name and a review date inside thirty days. State in the header: provisional desk posture — not legal advice; supersedes nothing counsel will later require.

Day 2. Role lane: pick the three heaviest AI uses on the programme. Write one provisional role line each. Flag any "might be high-risk / unknown" for Escalation, not for hallway debate.

Day 3. Inventory: circulate the list. Ask the team to add uses you missed. Freeze a v0.1 by end of day. Include the boring ones — polish, rename suggestions, "quick" summarisers.

Day 4. Oversight: bind Spoken Accept, Claim Sheet, and pack checks to Inventory rows. If a row has no check, it is a gap — close it or escalate it.

Day 5. Evidence and Escalation: name two artefacts you can already produce (for example a Spoken Accept note on a mapping reject; a Claim Sheet with an open hedge). Name two stop conditions. Dry-run one stop in standup — say the words "we stop here" once, so the room hears what Escalation sounds like.

By Friday the map exists. It will be incomplete. Completeness is counsel's job later. Naming lanes before 2027 planning is yours now. The Regulation itself applies in stages — literacy and prohibitions early; high-risk obligations on a later clock. Use that stagger as permission to build desk posture early, not as a reason to wait until the last applicable date.

Where Desk Posture Stops

The Desk Posture Map does not replace Legal, InfoSec, DPO, or a formal AI policy when the organisation requires one. It does not classify your systems with legal effect. It does not authorise a pilot. It does not settle whether a vendor's "AI-powered" line makes you a provider.

It answers a narrower question: while those artefacts are unfinished, what is this delivery team's provisional law for AI uses that already sit on the desk — written so 2027 does not arrive as a surprise?

When Legal's opinion lands, date the map "retired" or "superseded." Do not run two competing standards. When a use may be Annex III high-risk — employment tools, essential-service decisions, or other sensitive categories that sometimes touch delivery programmes at the edges — stop role cosplay and escalate. Edges are Escalation's job.

Compliance Theatre will offer another deck. Desk Posture offers a page Friday can still obey.


Name the role.

Name the check.

Name the stop — or inherit whoever pastes first.

Related

← Back to hub