Browse the reference ⌄
Artefact published

Readiness Diagnostic

Four self-assessment instruments at four altitudes: the six-question CIO self-diagnostic (can the firm govern agents at all?), the ten-pillar control-plane scorecard (0–60), station readiness (lowest of four, 0–3), and ten evidence-scored statements (0–20, pilot/building/running). How to score on records, how to read the results together without averaging them, and which page to start from for each weak answer.

Version
0.1.0
Owner
DEIXEI
Last reviewed
Reading time
19 min

For Cio · Cto · Cro · Ai Architect · Platform Engineer · Risk And Assurance · Internal Audit

Sources and details
Source works
Whitepaper · Book 1 · Book 09
Source chapters
1, 5
Refresh trigger
source change

Full source list ↓ Source index →

Use this page before funding anything. It collects the corpus’s self-assessment instruments in one place, says which question each one answers, and shows how to read the results without turning them into a single maturity number. Each instrument gives a score, and also a list of named gaps, each linked to the page that closes it.

Reading the labels. All four instruments are author models. The questions, bands, anchors and reported scores were written by the book authors from advisory work. None is an industry standard, a supervisory expectation or a benchmark. Book 09 says so explicitly of its ten statements and bands. The reported mean score (about 1.4 of 6) and best score (4 of 6) on the six-question diagnostic are the authors’ anecdotes from their own conversations, not survey data. Recommendation marks this page’s own synthesis: the reading rules across instruments, the evidence column, the start-here mapping and the record template.

Which instrument answers which question

InstrumentQuestion it answersAltitudeScaleTimeRun bySource
Six-question CIO self-diagnosticCan the firm govern an agent in production at all?Estate operating layer6 yes/noMinutesThe CIO, personallyWhitepaper Part I §3; Book 1 Ch 1 §5
Ten-pillar control-plane scorecardWhich parts of the cloud control plane are owned, enforced and observable?Platform10 × 3 × 0–2 = 0–60HoursHead of cloud, pillar ownersBook 1 Ch 5 §7
Station readinessCan agents work on the systems the last hard change crossed?Delivery pathLowest of four, 0–3An afternoonPlatform lead with the teamsBook 09 Closing (method detail on agent readiness)
Ten evidence-scored statementsCan the firm let agents carry consequential work?Factory10 × 0–2 = 0–20An afternoonCIO with second line and internal auditBook 09 Closing

The whitepaper’s diagnostic and Book 1’s Adoption Gap Diagnostic are the same six-question instrument. The whitepaper cites Book 1 Ch 1 for it. This page presents it once and notes where the two wordings differ.

Three scoring rules for every instrument

From Book 09 Closing, applied here to all four instruments (recommendation):

  1. Numbers come from records, not memory. Where no record exists, write unknown and the name of whoever should know. The blank is a result, and often the most useful one.
  2. Do not average anything. Not across instruments, not across stations, not across pillars into a board number.
  3. Do the reading yourself. Someone may pull the data, but the person accountable reads it.

Where an instrument needs activity data, take ninety days. Book 09 says a month is noise and a year hides the bend. Split activity figures between the teams that use coding assistants most and least.

The six-question CIO self-diagnostic

Answer each question yes, no, or credibly “in progress”. “Credible” means a funded, owned piece of work with a date, not an intention (recommendation).

#ThemeQuestion (paraphrased)“No” looks like”Yes” looks likeEvidence to put on the table (recommendation)Start here
1InventoryCan you produce, within one hour, a complete inventory of every AI agent in the estate, including agents embedded in SaaS tools?A committee spreadsheet, updated monthly, suspected incompleteA live registry queried from the platform, reconciled with deployment, runtime and identity recordsThe registry query and its run time; the agent count from runtime logs beside the registry countAgent lifecycle
2BoundaryFor each agent, can you state in one sentence what it may do, with which tools, over which data?”It does claims triage”: use case, not boundaryA bounded-agency record with mandate, means, memory and monitoring all populatedThree randomly drawn agents with their envelopesBounded agent
3AccountabilityIf an agent acts wrongly today, can you name the accountable human within fifteen minutes?A committee or “the business sponsor”: a body, not a personA named, individually accountable wranglerThe wrangler field for the same three agents, and a timed lookupAccountability
4EnforcementCan the platform refuse a policy-violating agent action at runtime and evidence the refusal?Policy in a long document; refusals appear as audit findings months laterPolicy-as-code at the API layer, deny events in the evidence pipelineA recent deny event and its evidence recordEnterprise control plane
5EvidenceCould you hand a regulator, this week, a structured timeline of any agent action in the last ninety days, including its reasoning at each step?Logs exist; reasoning is not captured; reconstruction takes engineering effortEvidence records populated automatically and queryableOne timeline produced by query, with the time takenAudit evidence
6IntegrationHas a named person, this year, connected the firm’s service-management, governance, security, resilience, AI and agentic-security obligations into one artefact?Each obligation owned by a different function; integration is an annual workshopA unified control map, owned and under change controlThe control map, its owner and its last change recordControl map

Scoring warn-only policy. Score question 4 “no” if policy-as-code warns but does not deny. Book 1 says that middle ground does not survive a serious incident review.

From policy-as-prose to policy-as-code: a policy that does not run is a memo.

A policy that warns but cannot refuse is still prose. Source: Building Organizations for the Age of AI (Book 1).

Bands (author’s bands, Book 1 Ch 1 §5.1; whitepaper Part I §3):

ResultReadingThe sources’ label
”No” to four or moreAgents on the roadmap, no operating layer to govern them”The 99 per cent"
"No” to two or threeShipping agents without a control plane: exposed”The 11 per cent without a control plane"
"Yes” or credible “in progress” on at least fiveThe position from which scaling, to several hundred production agents in the whitepaper’s framing, is safe”The 11 per cent with one”

The labels borrow from the 99/11/95 adoption-signal composite: roughly 99% of large organisations with agents on the roadmap, 11% with agents in production under named owners, and a very high incident signal among early shippers. Book 1 Ch 1 §4.3 calls this a directional composite of three separately sourced signals, not one measured population. Do not quote it as a statistic. This page uses the plain readings in the middle column for that reason.

Reported scores (anecdote). Book 1 reports a mean of about 1.4 out of six across the CIOs the authors ran it with, and a best credible score of 4. The whitepaper puts the highest defensible score across European regulated firms in mid-2026 at four. Both say a CIO who scores six has not looked closely. Treat a six as a prompt to re-check the evidence column.

Closure order (Book 1 Ch 1 §5.3). Close them in this order: inventory (1), then wranglers (3), bounded-agency records (2), the unified control map (6), policy-as-code (4), and audit evidence (5). The first three can start within weeks; the last three are programmes. Starting with 4 and 5 because they feel like progress fails: without 1–3 the platform has no operating layer to act on.

Wording differences. Question 6 in Book 1 names the frameworks (ITIL, COBIT, CIS, DORA, the AI Act, OWASP Agentic). The whitepaper names the domains (service management, governance, security, resilience, AI). The test is the same: one owned artefact, not one per function. The whitepaper frames the instrument explicitly as a CIO instrument, not a board instrument.

What to do with the result (Book 1 Ch 1 §7).

  • This week: score it yourself; do not delegate. Date it, and send it to the CRO and the head of internal audit with one paragraph of context, so the baseline exists outside your own head.
  • Within a month: commission the agent inventory, about six weeks of work. Cover explicit agents (built or procured), embedded agents (agentic features inside SaaS already in use) and experimental agents (proofs of concept that never went through change control). Book 1 expects the experimental category to be the largest.
  • Within a quarter: name a wrangler for every agent. Book 1’s guide for a large firm is no more than five to seven agents per wrangler.
  • At the board: replace the use-case slide with the score, in the form “we have x agents in production, our diagnostic score is y/6, and we commit to z by year end.”
  • Smaller firms run the same diagnostic. Book 1 argues their exposure is sharper because they have fewer redundant controls. Their first-count agent inventory is typically five to twenty agents.

The ten-pillar control-plane scorecard

Score each of the ten cloud capability pillars (identity, network, security, compute, data, integration, observability, governance, automation, AI platform control) on three dimensions at 0, 1 or 2. Each pillar scores 0–6; the total is out of 60 (Book 1 Ch 5 §7). Book 1 reports that no organisation scores above 50 on a first pass. The pillar definitions and per-cloud components are on cloud and hosting.

Dimension012
Ownership clarityNo accountable ownerOwner named, mandate unclearAccountable owner with a mandate
Policy coverageControls on paper onlySome controls enforced as codeControls enforced and evaluated as code
ObservabilityNot observablePartly observableEvidence answers the pillar’s auditor question

The level descriptions are this site’s anchors, shared with agent readiness. Book 1 defines the three dimensions and the 0–2 scale, but not the wording of each level.

Use the auditor’s question as the observability test. A pillar scores 2 on observability only if its evidence answers the question Book 1 Ch 5 §5 assigns to it (Book 1’s board slide sets ten minutes; using it as the scoring test is this page’s recommendation):

PillarAuditor’s question (Book 1 Ch 5 §5)
IdentityShow every identity that touched system X this week, including agents.
NetworkDemonstrate that no agent in development can reach production data.
SecurityHow quickly can you detect prompt injection?
ComputeWhat runs your critical or important functions?
DataWhere is sensitive data, and who can read it?
IntegrationWhich APIs and MCP servers are exposed, and who calls them?
ObservabilityShow the audit trail for action Y.
GovernanceWho approved this exception?
AutomationWas this deployed through an approved pipeline?
AI platform controlWhat can this agent actually do?

The 90-day plan (Book 1 Ch 5 §7). Pick the two lowest pillars. Book 1 expects them to be AI platform control (because it is new), plus integration (MCP servers not yet under control) or observability (the evidence schema not yet adopted). A realistic 90-day target moves one pillar from 2 to 5. A plan to move all ten by one point is, in the source’s word, theatre. Run a weekly 40-minute forum until the pillars stabilise: chaired by the wrangler lead, with one representative from each framework team and one from each pillar under work. In parallel, extend one practice per framework altitude: change enablement, managed data and asset inventory are Book 1’s starting points.

Owner floor. A large firm runs ten capability groups with one accountable director each. A firm of about 1,000 people combines them under at least five owners, two pillars each on average. Book 1 warns that one owner for all ten creates a single point of failure for the whole control plane.

Pre-flight check (Book 1 Ch 5 §8). Before scoring, check that the frameworks and the cloud organisation are operationally real. One test: ask three randomly chosen practice owners for their practice’s KPI last quarter. If two cannot answer, the framework is shelf-ware, and pillar scores built on it will mislead.

Station readiness (pointer)

The third instrument scores the three to five systems the firm’s last hard change had to cross. Each station gets 0–3 on four questions: can an agent learn what it is, catch a wrong change cheaply, undo it cheaply, and see the outcome? The station’s score is the lowest of the four, and the change’s score is its lowest station. At 0–1 an agent assists and a human reads every step. At 2, bounded repetitive work can be handed over with outcomes checked. At 3, a station is a candidate for standard changes without a reviewer, inside machine-enforced limits. The anchors, the record and the funding rule (low score against high dependence first) are on agent readiness. Book 09 Closing says never add the station score to the ten-statement total.

Ten evidence-scored statements

Book 09 Closing’s factory readiness score. Score each statement 0 (no), 1 (partly), or 2 (yes, with evidence). A 2 means you can put your hand on a record: a signed document, a register, a query result. Something you believe but cannot show scores 1. A plan scores 0.

#StatementWhat earns a 2 (paraphrased from Book 09)Evidence to pull (recommendation)Start here
1Every agent has a named ownerA register entry for every agent the platform’s logs show running, each naming a person, not a team; register count equals log countRegistry export and runtime agent count, same dayAccountability
2Agents have their own identitiesAgents never act under the identity of whoever started them; the access review treats agents as a separate principal class alongside people and workloadsLast access review, showing agents as a principal classIdentity and secrets
3There is one evidence recordOne query answers the twenty-change sample: who or what acted, under what limits, with which model, approved by whom (person or machine), with what outcome, each recorded at the moment of the actThe query and its output for last month’s sampleAudit evidence
4The routing gate sees where tests came from and what a change touchesThe auto-approval score takes two inputs: who wrote the tests, and how many components depend on what changed. Control values sit on a list, and any change to one goes to a reviewer. Where nothing is auto-approved, score the rules that assign reviewers on those two inputsGate configuration; the signed control-values listChange governance
5The CRO has signed a control budgetA signed band on how fast change may arrive relative to the capacity to stand behind it, with a measure, a threshold and consequences that follow without a meetingThe signed band and the last automatic consequence it triggeredControl budget
6Usage is measured for every shipped productA list of everything shipped in the past year, each with last month’s users or calling systems, read from the product itselfProduct list with usage source per rowFactory measurement
7Every agent loop has a stop conditionEach agent has a count, cost, time or repetition limit that stops it with no human in the loop, and the limit has been tested at least oncePer-agent stop condition and test dateRunning the factory
8Reversal is decided before releaseChange records carry the named way back before deployment; the last rehearsal of a way back has a dateSample change records; rehearsal logRelease engineering
9The seven owners are namedSeven names in one place, with anyone holding more than one seat declaredThe seat listSeven names
10The threshold for building is setA signed document with numbers: how many products, on how many genuinely shared patterns, over what horizon, one of them shared data with a stated promise about meaning and freshness. “Below the line, we buy.”The signed threshold and last countBuild, configure or consume

Bands (Book 09 Closing):

TotalBandMeaningNext action
0–7PilotWhatever the dashboards say, the agents are experimentsKeep them away from consequential change; start at line 1 of the ninety days
8–14BuildingThe usual result a year into a programmeThe zeros show which of the ninety-day lines were skipped
15–20RunningRe-score next quarterA running factory can slip back into building without anyone deciding it should

Read statements 4 and 5 together. Book 09 calls them the pair that fails late: full marks on the other eight still allow a gate to pass a change whose own tests were built to confirm it, at an arrival rate no one has signed off. A firm scoring 2 on statement 5 but 0 or 1 on statement 4 has a signed band over a gate that cannot see what matters (recommendation).

Inputs the statements assume. Statement 3 needs the twenty-change sample: twenty changes from last month, drawn at random by someone outside the teams, with approver, what was checked, test origin, minutes to approval, and whether a control value was touched. Book 09 adds the morning drill: timing an unannounced supervisor-style question. Both are defined on audit evidence. Run them first (recommendation): statement 3 cannot honestly score 2 without them.

Reading the results together

The instruments sit at different altitudes and fail in different ways. Read them as a stack, top down, and act on the lowest layer that is failing (recommendation):

PatternWhat it meansAct first on
Six-question: “no” to four or moreNo operating layer; every other score is prematureInventory, wranglers, bounded-agency records, in that order
Six-question passes, pillar scorecard has two pillars at 0–2 of 6The control plane has holes where agents will crossThe 90-day plan on the two lowest pillars
Pillars healthy, lowest station 0–1The platform is ready but the systems agents must change are notReadiness money worst-first, weighted by dependence (agent readiness)
Ten statements in PilotThe factory cannot yet carry consequential work, whatever the platform scoresThe implementation roadmap from line 1
Ten statements in Building, zeros on 1, 2 or 9Ownership and identity gapsAccountability, identity and secrets
Ten statements in Building, zeros on 3, 4 or 5Evidence and gate gaps: the late-failing pairAudit evidence, change governance, control budget
Statement 10 at 0 and a platform hire plannedSpending ahead of the portfolio ruleWrite and sign the threshold before the hire (build, configure or consume)
High self-scores with empty evidence columnsBelief, not evidenceRe-score with the evidence column; every unsupported 2 becomes a 1

Never send a composite upwards. Report each instrument’s score with its gaps named. Book 09 tells the reader not to send either score upwards as an average. Book 1 tells the CIO to put the six-question score, with the agent count and a commitment, on the board slide.

Size calibration (Book 09 Closing; Book 1 Ch 5 §7). In a firm of a few hundred people, the sheet takes an hour. The seven seats are shared among three or four people and declared. One combination must be refused: a single person who builds the controls on the agents and also tests whether they work. In such a firm the threshold will usually say buy. A firm of a few thousand, with technology under one head, gets the organisational work nearly free. It should borrow the operating model, size the controls to its own risk, and run the threshold hardest, because it is the most likely to fail it.

Diagnostic record

Keep each run as a dated record, so the next run is a comparison and not a fresh debate. This template is a recommendation; the sources prescribe no format.

FieldContent
run_dateDate scored
scored_byThe accountable person who did the reading
instrumentsix-question, pillar scorecard, station, ten statements
itemQuestion, pillar-dimension, station or statement number
scoreAs the instrument defines it, or unknown
evidence_refPointer to the record, query or signed document; empty means the score is at most 1 (or “no”)
unknown_ownerFor unknown items, the named person who should know
gap_actionThe page or ninety-day line that closes it, with an owner
rescore_dueNext run (quarterly for the ten statements; after each 90-day plan for the pillars)

Send the six-question result to the CRO and the head of internal audit on the day it is scored, as Book 1 advises.

Failure modes

FailureRecognise itRecover
Scoring from belief2s with nothing in the evidence columnApply Book 09’s rule: believed but unshown scores 1; a plan scores 0
One maturity numberA board slide with a single blended percentageReport each instrument separately with named gaps
Delegated diagnosticThe six questions answered by a team, not the CIOThe CIO re-scores personally
Starting with the platformFunding policy-as-code and evidence pipelines before inventory and ownersFollow the closure order: 1, 3, 2, then 6, 4, 5
Ten pillars at onceA 90-day plan moving every pillar by one pointTwo lowest pillars only
A six out of sixA perfect score on the first runRe-check against the evidence column; the sources treat six as not yet looked
Quoting 99/11/95 as a statisticThe composite presented as a survey resultLabel it directional, or drop it
No re-scoreThe last run is older than two quartersBook the quarterly re-score of the ten statements

Acceptance checks

  • The six-question diagnostic was scored by the CIO personally, dated, with an evidence reference per “yes”, and sent to the CRO and head of internal audit.
  • The pillar scorecard shows thirty cells (ten pillars × three dimensions); each observability 2 answers its auditor’s question from evidence.
  • The 90-day plan names the two lowest pillars, a target score for each, and the forum’s chair.
  • The station scores for the last hard change are recorded on the station assessment record, and are not added to any other total.
  • The ten statements are scored with an evidence reference for every 2. The twenty-change sample and morning drill were run before statement 3 was scored.
  • Statements 4 and 5 are reported side by side.
  • Every unknown names a person.
  • Each gap is mapped to a page or ninety-day line with an owner, and a re-score date is set.

Source status

  • One instrument, two sources. The whitepaper’s six questions summarise Book 1 Ch 1 §5 and cite it. The bands match. The reported scores differ in framing: Book 1 gives a mean of about 1.4 and a best of 4, the whitepaper a highest defensible score of four in mid-2026. Both are anecdote.
  • The pillar scorecard. Book 1 Ch 5 §7.1 describes a per-pillar 0–6 scale, and §7.2 a 0/1/2 score per dimension totalling 60. These are consistent: three dimensions × 0–2 per pillar. The level wording is this site’s own.
  • The ten statements, bands and seven seats are, in Book 09’s own words, new to that edition and the authors’ own: a starting point, not a standard.
  • Not covered here. The sectoral regulatory diagnostic is on the regulatory map, and the station method on agent readiness. These are different instruments.

Source material

  • Whitepaper (Building Organisations for the Age of AI), Part I §3, The diagnostic that locates the firm
  • Book 1 (Building Organizations for the Age of AI), Chapter 1, The Adoption Gap (§4.3 the 99/11/95 composite; §5.1–5.3 the Adoption Gap Diagnostic; §7 Monday morning and board conversation)
  • Book 1 (Building Organizations for the Age of AI), Chapter 5, The Framework Backbone and the Cloud Control Plane (§4.2 ten pillars; §5 pillar table and auditor's questions; §7 pillar scorecard and 90-day plan; §8 pre-flight check)
  • Book 09 (Everything Is Green), Closing, An Afternoon with a Pencil (three rules; station score; ten statements and bands; seven owners; at your size)

← All AI Delivery Factory guidance

On this page

  1. Which instrument answers which question
  2. Three scoring rules for every instrument
  3. The six-question CIO self-diagnostic
  4. The ten-pillar control-plane scorecard
  5. Station readiness (pointer)
  6. Ten evidence-scored statements
  7. Reading the results together
  8. Diagnostic record
  9. Failure modes
  10. Acceptance checks
  11. Source status

All pages in Adopt and fund →