Phase 02

Sprint

Turn the Awake decision into operating proof. In ten weeks, one workflow is designed again from zero, rebuilt at the edge, and run beside the old way against baselines fixed before anything is built. Nothing is switched off early; the new way earns the work only as it wins.

Start a Sprint

Ten weeks, full time on one workflow · from CHF 60,000 · Prerequisite: Awake completed

The 10 Weeks

One team. One workflow.
Proven, not promised.

Weeks 1–2 Two maps, then the Cut Map the chosen workflow as it is (decisions, exceptions, tribal knowledge), and fix the baseline before anything is built: cycle time, error rate, cost. Then, with the three to five people who will run the edge, design it again on a white canvas: if we needed this capability today, with everything now possible, how would we build it? Then the Cut runs on that design: every step, rule, approval, and field names its owner and its consequence, or it isn't built. Ten steps, or three?
Weeks 3–6 Build at the edge Capture what survived the Cut (your senior people's judgment as playbooks, the firm's real trade-offs written down), and build the new workflow on the harness primitives: a constitution the agents read first, playbooks, machine checks, named human gates. It is built on your existing tools: the orchestration layer is an implementation choice we make together, never a product you are locked into. It is insulated from the unit it will replace, and funded by you. A 60-minute weekly pulse keeps it honest.
Weeks 7–10 Shadow run, then it earns the work The new workflow runs beside the old one against the week-1 baselines; the proof is cycle time (ten times, not ten percent) and a falling override rate, not a demo. Then it earns the work: one percent of it, then ten, then all of it. The parts of the old way that turn out to be needed come back on their own, about ten percent. The old way stays live until there is nothing left to send to it. Document the playbook that lets the team run it independently, and the case study that funds the next workflow.
Expected Outcomes
10× case The qualification bar before we start; the shadow run measures what is actually achieved
60′ Weekly pulse meeting — the boring ritual that compounds
1→100% The share of the work the new way earns, in steps, as it wins
1 Team running it independently by week 10
What You Walk Away With
The Cut Design The new workflow, designed from zero and cut to what has a name and a consequence, with every decision and exception documented: the specification your AI needs.
A Working Harness The workflow live on the primitives (constitution, playbooks, checks, gates), connected to your existing tools, carrying real work, with the override log that proves it.
The Case for the Next One Before and after on cycle time, error rate, cost, and parts not built: the numbers that fund the next workflow, and, in your words, the story of how your people moved with the work.
The Investment
From CHF 60,000

Sold in two stages, and the stage boundary is already in the method: the first two weeks are the from-zero design and the Cut.

Stage 1 · The Design Gate · CHF 15,000 · weeks 1 to 2

It covers the from-zero redesign, the Cut, and the order-of-magnitude case in writing. It ends with a real kill option: if the Cut cannot show a credible order-of-magnitude cycle-time case, we stop. You keep the design artifacts and a rigorous, written no.

Stage 2 · Build & shadow run · from CHF 45,000 · weeks 3 to 10

It is priced per scope at the gate. Proceeding is your decision at the gate; Stage 1 creates no obligation.

Ready to prove it with one team?

The Sprint is how you turn the decision into proof: ten weeks, one workflow, fixed baselines, and a reversible path from one percent of the work to all of it.

Start Your Sprint

After Sprint? → Scale: the next workflow, until the edge is the company.

Why the pilot worked and the company didn't

Field notes on stalled AI transformation, executive judgment, and the specification economy — the weekly letter behind Book 2. About once a week. No spam. Unsubscribe anytime.