360, AI-native.

As soon as the repos are seeded: install the 360 plugin, run one proven flow (Gijs's Docker Fixer is the first), and drop one thing you tried, however half-baked, in the playground. That's the whole ask. No course, no gap to close, just a first rep.

Playground open to everyone from day one. Skills starts with one proven flow in it (Gijs's Docker Fixer), not an empty structure nobody fills. The plugin puts every proven flow in a consultant's terminal with one command.

An 18→19 environment reviewed for under €5 a run. Cheapest visible win we have, and every ingredient is already owned.

The first scoped, whitelisted key ships to the builder who asked for it. Oldest open promise on the list, from 6 July. More keys follow right after. It unblocks the chapter's MCP work.

One inbox, AI routes every invoice to the right entity. Two of the four entities first, reusing a flow we already built elsewhere, staging before go-live on 1 August. Our first AI flow in a client's production, which beats any demo.

A couple of colleagues sanity-check the draft first. Framed as show us what works, results published back within the week. Measure to boost people, never to audit them.

The all-in finance idea has been carried in the team for a year, and the flagship call finally gives it a lane. On paper first, then a working session to land it, including what it needs this time that it didn't get before.

Code pins in a month or two, so this starts now. OCA's OpenUpgrade for 20 trails by about half a year and our self-hosted base needs its own tooling regardless. What we learn flows back upstream to OpenUpgrade where it fits.

Which of the 36 on 16 and 17 go first, which are deep-custom and slow, which are near-standard and can prove the pipeline. Input to both the build week and the pricing call.

Gijs lifted his 19 MCP scope from the 20 codebase but couldn't verify this one. It sets how wide the safe-MCP window is. Hours of work for a bet-sized answer.

Three scenarios and the response we commit to, written before the reveal. That way September is execution, not scramble.

Start from the budget-vs-actual milestone data we already have, not from zero. Is the work genuinely faster, or do more tickets just come in? Nobody owns this number today, and the baseline rots while we wait. The full baseline follows the access ask, so the pricing calls wait for it.

Consented auto-harvest of prompts and flows into the repos, preview before send, never client data. Manual knowledge bases die on the contribution problem. This is the fix.

Gijs and Douwe plus a senior colleague at a receptive client. What's possible across their whole org, not a 360AI demo. The client asked us for exactly this after seeing the first flows, so the demand is already sitting there.

Gijs moves to where he's strongest: 0-to-1 R&D and the creative work. He proposed the split himself. Douwe picks up the client side and the internal connecting work. Agreed all round before it's official, and official before September.

The efficiency case, the migration pricing and the margin numbers all run on estimates until this lands. The one item on the page that needs a push from leadership.

Lock two or three people away for a week, come out with the scripts and the agent, trained on 19 and 20 source. Runs once the 20 code pins, with our migration specialists as the design centre: they call what the pipeline may touch, and review authority stays with them. Owners named in the September session.

Two or three, starting with near-standard clients off the 16 and 17 list. They prove the loop and they have a support deadline anyway. Every issue they hit feeds back into the agent.

Fixed price or hours. The cleanest fixed-price-arbitrage candidate we have: repeatable, scriptable, bounded. Decide before the wave, because after the first quote it's set.

One friendly existing client, high-volume and rules-based, with a clean outcome metric. Draft-create and an approve inbox first. No agent writes anything without the audit gate in front of it.

One or two, with the ownership instrument decided before the first offer goes out. Comp is table stakes at 360, so it isn't the pitch and it isn't the retention lever either.

A willing client with an outline outcome contract, plus a real cost-per-outcome number on one workflow we already deliver. This is what the 100-day review checks.

Both wait for the repos to carry real weight. The client-summary pull is the one people keep asking for unprompted: what does this client do, what's running, what are their flows.

30 to 40 projects on Gijs's estimate, and the install base says that's the floor. The year the tooling pays for itself many times over, or the year we find out we were late.

One existing per-seat customer re-papered to a hybrid outcome SLA: retainer plus consumption plus an outcome bonus. Never pure-outcome, never 3-year fixed.

An issue seen once is known forever, so migration 30 costs a fraction of migration 3. The compounding asset, and it only exists if we design it in from the start.

Every hire a manager of one, each paired with a domain expert. Headcount is the residual, not the target. We speak in revenue-per-FDE and agent reuse instead.

Per-seat caps out. The ceiling arithmetic lives in the strategy. Move the flagship to outcome pricing before Odoo 20 flattens the surface.