The closing session of each Clonmel weekend is the most important one. By the time you walk back to the car park you should have a written plan for your own organisation's brain — not a slide deck, not a vision document, a working specification you can hand to an engineer on Monday morning.
The plan fits on one page. If it doesn't fit on one page, it is not a plan; it is a wish list. Below is the template we use, with notes on what to put in each section.
Section 1 · The first concrete process you will operate through the brain
Pick one. Not five. The hardest part of getting an organisational brain off the ground is resisting the temptation to design the whole thing before shipping any of it. Choose the highest-pain repeatable process in the company today — the work your sharpest operator does in the morning that they shouldn't have to. That process is the seed.
Write it as one sentence: “The brain will run our [process] every [cadence], producing [output], replacing [n] hours of [role]'s time per [period].” If you cannot write that sentence cleanly, you are not ready to ship; you are ready to scope.
Section 2 · The agents
List the agents the seed process needs. Most seed processes need between three and seven. For each agent, write one line:
- Name — short, descriptive, role-specific.
- Job — one sentence, action-verb start.
- Model — which model runs this agent, and why. (“Sonnet because it needs to plan,” “fine-tuned Yi-7B because it's high-volume classification.”)
- Tools — comma-separated list of the tools this agent can call.
You should have one planner, one auditor, and one to five workers. If you have more, the seed process is too big.
Section 3 · The state layer
Three lines. Where structured state lives (postgres, sqlite, your existing OLTP) — usually the answer is “the database we already have.” Where unstructured state lives (object store, file system, a vector index) — usually the answer is “extend what we have.” The single canonical schema — what does the brain consider an “event,” what does it consider a “reflection,” what does it consider a “run.” Three lines, no more.
Section 4 · The tools
Every external action the brain can take is a tool. Every tool gets one line: name, what it does, who it talks to (gmail, stripe, mongo, slack), and whether it has a dry-run mode. If a tool with side effects does not have a dry-run mode, mark it red — that is the first thing engineering builds.
Section 5 · Governance
Three sub-lines. Who can change agent prompts in production (the answer is a named human, not “the team”). Where the audit log lives, and how long it is retained (regulator-grade if you are regulated, ninety days minimum otherwise). The kill switch — the single command that stops every agent and disables every tool. Test the kill switch before you ship.
Section 6 · The first three weeks
Twenty-one days, week by week, in plain English.
- Week one. State layer + first three tools, all with dry-run modes, all wired to your existing systems. End of week one: planner agent runs once, in dry-run mode, and produces a plan you can read. No actions taken yet.
- Week two. Worker agents wired up, end-to-end dry-run pipeline. End of week two: a full run produces an output you would be proud to show the operator the brain is replacing. Still in dry-run.
- Week three. Auditor wired in, kill switch tested, first live run in shadow mode (the brain runs the process; the human still does it; you compare). End of week three: live execution of the seed process, with the operator watching. Adjust. Then turn the operator off the work and reassign their week.
Section 7 · What gets added next
The bottom of the page. Three bullets, the next three processes you will run through the brain after the seed is stable. This is not a commitment; it is a north star. The actual order will change. But writing it down anchors the priority and signals to the team that the seed process is the start of something, not the whole thing.
What we'll do in the workshop
The closing session of each Clonmel weekend works through this template per attendee, in small groups, with hands-on coaching. Everyone leaves with a printed one-pager for their own organisation, signed off by their group, and a list of three things to start on Monday morning. We've run this format privately with a small number of teams over the last twelve months. The pattern is consistent: teams who leave with a one-page plan ship something within thirty days. Teams who leave without one don't.