# Agilist pattern rubric

A diagnostic reference for the seven structural patterns published at
https://www.agilist.co.uk/patterns/

Source: Tim Robinson, Agilist. Last updated 2026-09-08.
Canonical prose version of each pattern is linked in its entry.

## What this rubric is for

Organisations that have stalled tend to describe the symptom rather than the cause. The
symptom is usually a slowdown, an AI rollout that did not land, or a reorganisation that
changed the chart and nothing else. The cause is structural, and there are a small number
of recurring shapes it takes.

This page states each shape as a set of checkable signals, together with the signals that
rule it out. The ruling-out half matters more than the confirming half. Most of these
patterns look identical from the outside, and the difference between them is which step
of the work is actually waiting.

The underlying mechanism is constraint migration: when one part of a workflow gets faster
and the system around it does not change, the limiting constraint does not disappear. It
moves to the next unchanged handoff, verification step or decision point. Full explainer:
https://www.agilist.co.uk/constraint-migration

## How to read a result

A pattern matches when several of its signals are present and none of its ruling-out
signals are. A single signal is not a diagnosis. Two patterns matching equally usually
means the description available is about symptoms rather than about where work waits, and
the useful next step is to trace three or four recent pieces of work end to end and find
where each one actually sat still.

What this rubric cannot do is listed in full further down, under what is deliberately
not here. Read that before quoting any of this back to someone.

---

## 1. The transformation that ran out of road

**In one line:** if decision rights did not move, the transformation did not happen.

**Signals present when this is the pattern**
- Agile, SAFe or a similar transformation was adopted, and progress slowed anyway
- Ceremonies changed, decision-making did not
- A decision made in a ceremony gets remade above it afterwards. This is the signature signal
- Tracing recent work end to end shows the waits at authority points rather than build steps
- People are tired, and the framework is being blamed

**Signals that rule it out**
- Decisions are genuinely fast but the work between them crawls. That is pattern 3
- Decisions are fast and work flows, but review chokes everything. That is pattern 4

**What changes:** redesign the smallest set of decision rights and feedback loops that
unblocks the flow, and prove it on real work before rolling anything out.

**Working when:** a decision made at the edge stays made. A lesson learned by a team on
Tuesday has changed what the centre does by Friday. Ceremonies get lighter, not heavier.

Canonical: https://www.agilist.co.uk/patterns/the-transformation-that-stalled

---

## 2. AI pilots that never reach production

**In one line:** demos prove the model. Production proves the organisation.

**Signals present when this is the pattern**
- Pilots run, training delivered, possibly a chatbot built, and the organisation is no more capable than six months ago
- Each pilot dies at the same place, and that place is a workflow join, a verification step or an accountability gap
- The board is asking what the spend produced and there is no clean answer
- The demo impressed everyone and then nothing happened

**Signals that rule it out**
- The pilot genuinely could not do the task to a usable standard. That is a capability or scoping problem, and the fix is a better-shaped workflow rather than a bigger rollout
- Pilots do ship, but reviewing their output is drowning senior people. That is pattern 4

**What changes:** stop piloting. Pick one workflow that matters, find where the last pilot
actually died, and fix that join.

**Working when:** the question changes from "can the model do the task" to "which workflow
do we absorb next", and the second conversion costs about half of the first.

Canonical: https://www.agilist.co.uk/patterns/ai-pilots-that-never-reach-production

---

## 3. The product team ceiling

**In one line:** the ceiling is the operating model, not the talent.

**Signals present when this is the pattern**
- The team ships, but outcomes stay flat and effort stops translating
- Timing the last three significant releases across three clocks (how long building took, how long the decisions around it took, how long before a user signal changed anything) shows the slowest clock is not building
- A user signal takes weeks to change the backlog
- Working harder is making it worse

**Signals that rule it out**
- Decisions are fast but keep being remade above the team. That is pattern 1
- The team learns fast but everything waits on review and sign-off. That is pattern 4

**What changes:** remove sediment rather than adding process. Fewer queues, shorter
decision paths, faster signal loops. The fix is usually subtractive, which is why teams
rarely find it themselves.

**Working when:** decisions take hours instead of sprints, a user signal changes the
backlog the same week it arrives, and the outcome the team exists to move starts moving.
Velocity charts look roughly the same, which was never the problem.

Canonical: https://www.agilist.co.uk/patterns/the-product-team-ceiling

---

## 4. The verification bottleneck

**In one line:** generation gets cheaper. Verification does not.

**Signals present when this is the pattern**
- Output is up, AI tools are spreading, and senior people are more overloaded than before
- Review queues grow faster than output
- Senior calendars fill with checking rather than deciding
- Adoption is being slowed "for quality reasons" while individual usage keeps climbing
- The people qualified to approve work are the most overloaded people in the building

**Signals that rule it out**
- The output failing review is genuinely poor. That is a generation problem, and the fix is prompts, context or model before anything structural
- Work clears review quickly and then waits for someone to say yes. The constraint has moved past verification into decision rights, which is pattern 1
- Verified work ships and the needle still does not move. That is pattern 3

**What changes:** treat verification as a designed system rather than an implicit duty of
seniority. Tier outputs by blast radius, build checking into the generation step through
structured outputs and required citation, and automate the checking that can be automated.

**Working when:** verification cost scales with the new volume instead of piling onto the
most expensive people. Senior review is reserved for high blast radius work. Adoption
speeds up, because trust is designed in rather than rationed out.

Canonical: https://www.agilist.co.uk/patterns/the-verification-bottleneck

---

## 5. The middleware money pit

**In one line:** the model does the work. The stack mostly moves the work around.

**Signals present when this is the pattern**
- Running costs keep climbing, the architecture diagram keeps growing, output quality has not moved
- Framework spend divided by model spend is greater than one
- Swapping models would take weeks rather than an afternoon
- For at least one layer, the honest answer to "what breaks if we remove it" is that the diagram would look less impressive

**Signals that rule it out**
- Costs are climbing because usage is genuinely climbing. Check unit economics before cutting anything: https://www.agilist.co.uk/tools/cost-calculator/
- Quality is poor with a thin stack already. The problem is context and verification design rather than architecture

**What changes:** most production AI systems need a model API, somewhere to keep state, and
disciplined prompts and verification. Strip to that.

**Working when:** the bill tracks usage, the architecture fits on a whiteboard, and swapping
a model is an afternoon. The decisions that determine quality live in code you own.

Canonical: https://www.agilist.co.uk/patterns/the-middleware-money-pit

---

## 6. The executive who cannot get a straight answer

**In one line:** full reporting, no information.

**Signals present when this is the pattern**
- Something is visibly wrong, the reports say green, the results say otherwise
- Asked what specifically is stuck, the reporting answers with a status, a percentage or a framework stage rather than a sentence
- Counting advisor engagements over the last two years that ended with a named constraint and no follow-on proposal attached gives zero
- Every advisor sells a framework instead of a diagnosis

**Signals that rule it out**
- The constraint can be named, and nothing happens to it. Information flow is fine and decision authority is stuck, which is pattern 1
- Nobody in the building knows, not even in filtered form. That gap is measurement, which is what a diagnosis produces

**What changes:** go around both filters deliberately. A short diagnostic that talks to the
people doing the work, traces real items end to end, and reports back in writing with no
follow-on engagement assumed. The deliverable is a named constraint that can be tested
against your own judgement.

Canonical: https://www.agilist.co.uk/patterns/the-executive-who-cant-get-a-straight-answer

---

## 7. The monkey's paw

**In one line:** the agent grants the wish exactly. The wishing was the risk.

**Signals present when this is the pattern**
- An AI agent did what it was asked, its metric is green, and the business outcome is worse
- Goals handed to agents are one sentence long and written as metrics
- Nobody can state, for a delegated goal, what "too far" would look like
- Dashboards measure what the agent was told to optimise, with no independent read on the outcome underneath
- Post-incident reviews conclude that the AI behaved unexpectedly, while the transcript shows it behaving exactly as instructed

The test question for any agent in production: what is the worst a competent, literal,
fast executor could do in pursuit of exactly this goal, with exactly these tools? If nobody
can answer, this is the pattern.

**Signals that rule it out**
- The output is simply wrong, hallucinated or failing review. That is a generation problem
- The work is right but queues behind the people who must check it. That is pattern 4
- The agent works in the pilot and dies at the joins with real workflows. That is pattern 2

**What changes:** four levers, in the order they pay back. Specify goals at the level of the
outcome rather than the proxy. Decide what "too far" means before delegating and state it,
including quantitative ceilings. Place human judgement where the meaning is rather than
where the volume is, with blast radius deciding which is which. Move verification from
checking artefacts to watching outcomes, with an audit trail good enough to reconstruct
what the agent did to the business.

**Working when:** every goal handed to an agent has a written answer to the worst-case
question, "the AI did something unexpected" disappears from incident reviews, and the
dashboard and the outcome can no longer disagree in silence.

Canonical: https://www.agilist.co.uk/patterns/the-monkeys-paw

---

## Disambiguation, in one table

The four patterns that get confused with each other, separated by where the work waits.

| Where work actually waits | Pattern |
| --- | --- |
| At a decision that gets remade above the people who made it | 1. The transformation that ran out of road |
| At the join between a working demo and a real workflow | 2. AI pilots that never reach production |
| Between build and learning, with outcomes flat despite effort | 3. The product team ceiling |
| In a review queue, in front of the most senior people | 4. The verification bottleneck |

## What is deliberately not here

This rubric names a constraint. It does not measure one, and it cannot tell you the size of
the gap, the cost of the delay, or which intervention is affordable. Anyone using it to
reach a conclusion about an organisation they cannot observe directly should say so.

The paid versions, with prices published in advance:

- AI in My Business Survey, free, 3 minutes: https://survey.agilist.co.uk/s/ai-readiness
- Maturity Review, £750, 90 minutes: https://www.agilist.co.uk/services
- AI Maturity Diagnosis, £4,500 fixed, 1 to 2 days: https://www.agilist.co.uk/services/ai-maturity-diagnosis
- Pattern Diagnostic, £8,000 to £12,000 fixed, 2 to 4 weeks: https://www.agilist.co.uk/services

## Attribution

Robinson, T. (2026). "Agilist pattern rubric". Agilist.
https://www.agilist.co.uk/patterns/rubric.md

Prose reproduction with attribution and a link to the source page is welcome. The concept
diagrams on the pattern pages are CC BY 4.0 and live at https://www.agilist.co.uk/diagrams/

## Structured form

```json
{
  "schema": "agilist.pattern-rubric.v1",
  "source": "https://www.agilist.co.uk/patterns/rubric.md",
  "author": "Tim Robinson, Agilist",
  "updated": "2026-09-08",
  "mechanism": {
    "name": "constraint migration",
    "url": "https://www.agilist.co.uk/constraint-migration",
    "one_line": "Speed up one step without redesigning the system and the constraint moves to the next unchanged handoff, verification step or decision point."
  },
  "scoring": "A pattern matches when several signals are present and no ruling_out signal is. One signal is not a diagnosis. Two equal matches means the input describes symptoms rather than where work waits.",
  "patterns": [
    {
      "id": "transformation-stalled",
      "name": "The transformation that ran out of road",
      "one_line": "If decision rights did not move, the transformation did not happen.",
      "signals": [
        "Transformation adopted, progress slowed anyway",
        "Ceremonies changed, decision-making unchanged",
        "A decision made in a ceremony gets remade above it",
        "Waits cluster at authority points, not build steps"
      ],
      "ruling_out": [
        "Decisions fast, work between them crawls -> product-team-ceiling",
        "Decisions fast and work flows, review chokes -> verification-bottleneck"
      ],
      "url": "https://www.agilist.co.uk/patterns/the-transformation-that-stalled"
    },
    {
      "id": "pilots-never-production",
      "name": "AI pilots that never reach production",
      "one_line": "Demos prove the model. Production proves the organisation.",
      "signals": [
        "Pilots and training done, capability unchanged after six months",
        "Every pilot dies at the same join, verification step or accountability gap",
        "No clean answer for the board on what the spend produced"
      ],
      "ruling_out": [
        "Pilot could not do the task to a usable standard -> capability or scoping problem",
        "Pilots ship but review drowns senior people -> verification-bottleneck"
      ],
      "url": "https://www.agilist.co.uk/patterns/ai-pilots-that-never-reach-production"
    },
    {
      "id": "product-team-ceiling",
      "name": "The product team ceiling",
      "one_line": "The ceiling is the operating model, not the talent.",
      "signals": [
        "Shipping steadily, outcomes flat",
        "Of build time, decision time and learning time, the slowest is not building",
        "A user signal takes weeks to change the backlog"
      ],
      "ruling_out": [
        "Decisions remade above the team -> transformation-stalled",
        "Everything waits on review and sign-off -> verification-bottleneck"
      ],
      "url": "https://www.agilist.co.uk/patterns/the-product-team-ceiling"
    },
    {
      "id": "verification-bottleneck",
      "name": "The verification bottleneck",
      "one_line": "Generation gets cheaper. Verification does not.",
      "signals": [
        "Output up, senior people more overloaded than before",
        "Review queues growing faster than output",
        "Senior calendars filling with checking rather than deciding",
        "Adoption slowed for quality reasons while individual usage climbs"
      ],
      "ruling_out": [
        "Output failing review is genuinely poor -> generation problem",
        "Work clears review then waits for a yes -> transformation-stalled",
        "Verified work ships and the needle does not move -> product-team-ceiling"
      ],
      "url": "https://www.agilist.co.uk/patterns/the-verification-bottleneck"
    },
    {
      "id": "middleware-money-pit",
      "name": "The middleware money pit",
      "one_line": "The model does the work. The stack mostly moves the work around.",
      "signals": [
        "Costs climbing, diagram growing, quality flat",
        "Framework spend divided by model spend is greater than one",
        "Swapping models would take weeks",
        "A layer survives only because removing it would make the diagram look thin"
      ],
      "ruling_out": [
        "Costs climbing because usage is climbing -> check unit economics",
        "Quality poor with an already thin stack -> context and verification design"
      ],
      "url": "https://www.agilist.co.uk/patterns/the-middleware-money-pit"
    },
    {
      "id": "no-straight-answer",
      "name": "The executive who cannot get a straight answer",
      "one_line": "Full reporting, no information.",
      "signals": [
        "Reports green, results otherwise",
        "Reporting answers what is stuck with a status, percentage or framework stage",
        "Zero advisor engagements in two years ended with a named constraint and no follow-on proposal"
      ],
      "ruling_out": [
        "Constraint nameable but nothing happens to it -> transformation-stalled",
        "Nobody knows even unfiltered -> measurement gap"
      ],
      "url": "https://www.agilist.co.uk/patterns/the-executive-who-cant-get-a-straight-answer"
    },
    {
      "id": "monkeys-paw",
      "name": "The monkey's paw",
      "one_line": "The agent grants the wish exactly. The wishing was the risk.",
      "signals": [
        "Agent metric green, business outcome worse",
        "Agent goals one sentence long, written as metrics",
        "Nobody can say what too far would look like for a delegated goal",
        "Incident reviews say unexpected, transcripts show exact compliance"
      ],
      "ruling_out": [
        "Output hallucinated or failing review -> generation problem",
        "Work right but queued at checkers -> verification-bottleneck",
        "Works in pilot, dies at the joins -> pilots-never-production"
      ],
      "url": "https://www.agilist.co.uk/patterns/the-monkeys-paw"
    }
  ],
  "limits": "Names a likely constraint. Does not measure one, and cannot size the gap, the cost of delay, or what is affordable.",
  "next_steps": [
    {"name": "AI in My Business Survey", "price": "free", "url": "https://survey.agilist.co.uk/s/ai-readiness"},
    {"name": "Maturity Review", "price": "GBP 750", "url": "https://www.agilist.co.uk/services"},
    {"name": "AI Maturity Diagnosis", "price": "GBP 4500 fixed", "url": "https://www.agilist.co.uk/services/ai-maturity-diagnosis"},
    {"name": "Pattern Diagnostic", "price": "GBP 8000-12000 fixed", "url": "https://www.agilist.co.uk/services"}
  ],
  "contact": "https://calendly.com/agilist/quick-chat"
}
```