wardley-maps.sgit.ai / Doctrine / Explorer / Villager / Town Planner

Explorer / Villager / Town Planner

Original contribution Founding brief 14 February 2026 W1 · glossary

Simon Wardley's Pioneers–Settlers–Town Planners is well known as an organisational pattern for human teams. What follows is something else: PST used as the architecture of an AI agent system — three separate directory trees, three session contracts, three evolution-stage mandates, and a documented handover between them. It has never been written up anywhere, including in the five agentic-setup guides written by the project that runs it.

Lead with the filesystem

The theory is available in a hundred places. This is the part that is new, and it is three directories:

.claude/CLAUDE.md              .claude/explorer/CLAUDE.md      (816 words)
                               .claude/villager/CLAUDE.md      (1,836 words)
                               .claude/town-planner/CLAUDE.md  (942 words)

team/roles/              <-- THE EXPLORER TEAM  (note: no "explorer/" prefix)
team/villager/roles/
team/town-planner/roles/

There is no team/explorer/ directory. The Explorer team is the unprefixed default — team/roles/ — annotated in the root session contract as # Explorer team role-based review documents. That asymmetry is not an oversight to apologise for. The Explorer is the origin state. Villager and Town Planner are named departures from it. Everything starts unprefixed and at Genesis, and acquiring a prefix is what evolution looks like in a filesystem.

The operative table

From the root session contract, verbatim — the most authoritative statement of the mapping anywhere in the corpus, preceded by "As of v0.5.8, the project operates with three teams based on Wardley Maps methodology":

Each team's own contract restates its stage in the second person — "You operate at the Genesis → Custom-Built stages of the Wardley evolution axis" — and that sentence is the mechanism that makes this more than a diagram. The agent is told which stage it is standing on before it does anything. A prompt that begins by fixing the evolution stage constrains everything downstream: what counts as done, what counts as good enough, whether a missing test is a defect or a deliberate deferral.

Which roles sit in which population

From the founding brief of 14 February 2026, verbatim headings:

• Primarily Explorer — Architect ("Designing new components, experimenting with approaches"), Designer, Dev, Journalist, Ambassador, Advocate, Sherpa
• Primarily Villager — DevOps ("Deployment, infrastructure, performance, monitoring"), QA, GRC, DPO
• Across both — Conductor ("Orchestrates both teams, manages the handover"), CISO, AppSec, Librarian, Historian, Cartographer ("Maps the system at both stages")

The Villager-exclusive translator

Described in the Villager contract as "Villager-exclusive… the first role that exists only in the Villager team". This is Wardley-consistent and rather elegant: translation is a productisation activity. Nothing at Genesis needs translating, because nothing at Genesis has a second audience yet. A role that can only exist at one stage of the axis is the strongest evidence in the whole setup that the stage assignment is doing real work rather than decorating an org chart.

The Town Planner team was created a week after being deferred

The founding brief is explicit: "❌ Town Planners Team — not yet… Attempting to create the Town Planners team now would be premature. We don't have anything at commodity stage yet." It was created on 20–21 February 2026, with the Alchemist as its founding role — earlier than the brief's own stated criterion, and for a business rather than a technical reason.

That is published here on purpose. A documented deviation from your own plan, with the date and the reason, is exactly the kind of evidence the doctrine assessment demands of every rating it makes. A pattern write-up that only contains the parts that went to plan is a brochure.

The theft, and why the agentic version is different

The key dynamic is what Wardley calls the "theft" — the villagers take (pinch) what the explorers have built and turn it into a product.

That is standard doctrine and belongs to Wardley. The agentic version is not.

In a human organisation the theft is a political event. Somebody's work is taken from them, ownership changes, credit is contested, and the whole thing needs managing by a person with standing. Between agent teams the theft is a file move plus a version bump — and it can be logged, dated, and reviewed in a diff. There is no ego to manage, because there is nobody to have one.

That is a genuine advantage of running PST with agents rather than people, and it is claimed here rather than hinted at. The corresponding disadvantage is equally real: the political friction in a human handover is partly a quality gate. Somebody defending their work forces the taker to understand it. A file move does not.

The strongest extension: PST applied to artefacts, not people

the lifecycle of a skill is to start as massive LLM, held together with string and gum, very expensive, lots of tokens, all the way to a product, all the way to code, and it almost becomes a commodity, eventually disappearing, because maybe it is absorbed by something else.

The mechanism is stated in the same brief — "the natural progression of a skill is to start in English and end up in code. As you refine the skill, you reduce the scope, you reduce the variability" — and the consequence is the most useful sentence in the entire corpus:

Anybody who spends a lot of money on tokens has an engineering problem. They are using explorer-type code and solutions in a commodity environment.

That turns a strategy framework into an operational cost diagnostic. Your token bill is a measurement of where your capabilities sit on the evolution axis: expensive means Genesis, and Genesis in a place where a commodity answer exists means you have an engineering problem rather than a pricing problem. It is the single most actionable thing this site can hand a working engineer, and it was unpublished until now. 4 June 2026.

Two related pieces from the same period. The same skill differs by character — "a skill, the same skill for a town planner, is very different for an explorer, for a villager" — which is what makes the three-team structure more than three folders. And the seniority inversion: the vibe-coding brief assigns Villager work to "junior and mid-level engineers" and Town Planner work to "the most experienced engineers on the team", inverting the usual assumption that the exciting new thing goes to the seniors. The same brief argues that "the vibe coded MVP is the specification. It is more precise, more concrete, and more usable than any document the business stakeholder could have written" — which is the Explorer output the theft acts on.

Thirty-one of thirty-three roles never got the memo

Asserted against implemented

The Cartographer's own reference document supplies the verdict on that sixth row: "a map that was accurate three months ago and hasn't been updated is worse than no map." It is now six months. Which raises Q8 — if the discipline failed with a named role and a stated responsibility, what would make it work?

Where it came from

On 16 June 2020, six years before any of this, Dinis Cruz sat in an Open Security Summit working session called Team Topologies & PST & Squads & Tribes — with Simon Wardley in the room, alongside Tony Richards and Luke Robbertse. The recording is on the chronology page. Nobody has ever put the 2020 session and the 2026 filesystem side by side. That is the most satisfying connection in the whole research, and it is the reason this pattern is not a rediscovery.

What would make this page better

• Draw the map of the pattern itself. Where do the three teams sit on an evolution axis? Where does translator sit? It does not exist yet, it is cheap, and it is an obvious original artefact. Logged as T3.
• Measure it. Q7: does the PST→agent-team mapping actually improve outcomes, or is it a legible metaphor? There is no measurement, and one operationalised role definition out of 33 is thin evidence either way.
• Push the stage mandate down into the roles. The defect above is a find-and-fix, not a redesign.