06-reference/research

agentic assessment framework competitive landscape

2026-07-15·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
phdatacafcompetitive-landscapeai-readiness-assessmentagentic-frameworks

The Assessment-to-Build Landscape: Everyone Sells the Diagnostic, Almost Nobody Couples It to the Build

The question

How do other consultancies / AI firms structure agentic "assessment-to-build" frameworks versus phData's CAF, and what is the competitive landscape for AI-readiness assessment methodologies? Context: Ben is the named technical PM for CAF (his first official PM role) and phData is his main bet as of 2026-06-08 — this benchmarks CAF's structure against public competitor frameworks while he restructures it.

What we already know (from the vault)

What the web says

Convergences and contradictions

Synthesis for RDCO

The single most important structural finding: the readiness-assessment layer is a commodity and is racing to zero, while the assessment-to-build coupling is where the defensible product lives. The evidence is unusually clean. AWS — the vendor with the most to gain from owning enterprise AI adoption — has let its own CAF-AI whitepaper go to "historical reference only." McKinsey, Accenture, KPMG, and Deloitte are all scoring the same five-ish dimensions (strategy, data, technology, organization/talent, governance) with different nouns on top. And a long tail of content marketers is giving away 8-dimension frameworks and scoring quizzes as lead-gen. If CAF's restructure spends any energy trying to win on "better readiness dimensions" or "a more rigorous maturity index," it is competing in a market where the going rate is $0 and the category leader has abandoned the field. That layer should be table stakes inside CAF, never the hero. The hero is the thing nearly none of them have: an assessment whose output is an executable input to the build rather than a PDF that a human re-types into a scoping doc.

But the differentiation seam is narrower than the vault currently claims, and it is not "assessment-to-build" in general — it is portfolio-layer assessment-to-build. Thoughtworks AI/works is the real comparison set, and it is doing the coupled thing: Reverse Engineering → Dynamic Spec Development → Spec to Code, with Control Plane and Runtime Ops behind it. That is a genuine assessment-to-build engine with a machine-consumable artifact at the center, and it invalidates any pitch premised on "nobody productizes the front of the funnel." The distinction that survives is one of altitude. AI/works operates at the application SDLC layer — it takes a decided-upon system and drives it from legacy/spec to running code. It answers how do we build this thing well and fast. CAF operates one level up, at the use-case portfolio layer — it answers which agents should exist in this estate at all, in what order, at what autonomy, and what does the governed build of each look like, and then hands a manifest down into delivery. Thoughtworks starts after the decision; CAF's first half is the decision, made repeatably. That's the seam, and it's real — but it must be stated at that altitude, because "we do assessment-to-build" is now a contested claim rather than a whitespace claim.

Three structural moves worth stealing, one worth naming, and three worth deliberately rejecting. Steal: (1) A time-boxed clock as the product promise. Thoughtworks' 3-3-3 (3 days to concept, 3 weeks to prototype, 3 months to production MVP) is doing enormous work — it converts an abstract methodology into a purchasable, falsifiable commitment, it forces delivery discipline internally, and it is the single best answer to Accenture's scale story, because a mid-market buyer cannot verify CoE depth but can absolutely verify a calendar. CAF has no public clock. It should get one, and the vault's existing "built in a weekend" proof point suggests the honest number is aggressive. (2) Autonomy/authority as a first-class assessment axis. The Big-4's five dimensions are pre-agentic — they score data, talent, process, tech, and none of them score how much authority this agent may hold. Thoughtworks saw this and built a whole framework for it (manual / semi-automated / automated oversight tiers). CAF's routing-by-autonomy instinct is already aligned with where the frontier is going; it should be elevated from an internal routing rule to a named, client-visible scoring axis, because it is simultaneously the most agentic-specific and the most governance-legible thing in the framework. (3) A named, exportable client-facing artifact. "Scope of Authority blueprint" is a good noun. CAF's build manifest is currently an internal object with an internal name; a client-facing name turns it from plumbing into the deliverable the client believes they bought.

Reject, deliberately: the CoE + certified-headcount scale story (unwinnable per the vault — Accenture's 5,000 certified people are a procurement moat, not a vanity metric); the maturity-index-as-hero (commoditized to free, and it is a diagnostic dead-end that terminates in a roadmap instead of a build); and the "N modules / M agents" arms race (the vault already called this — phData loses a module-count contest to Deloitte on day one). And take the naming question seriously as PM: CAF collides head-on with AWS CAF, which has ten years of SEO, a published AI extension, and reflexive recognition among exactly the cloud architects who sit in these rooms. Ben owns this decision now. The cheap version is to keep CAF internally and give the client-facing artifact and clock distinct, ownable names — which conveniently is the same move as (3) above. That's one restructure decision that pays off twice.

Why this is in the vault

This directly informs Ben's CAF restructure as named technical PM — it invalidates one specific line the vault was carrying ("nobody publicly productizes assessment-to-build"), identifies Thoughtworks AI/works as the real comparison set at a different altitude, and surfaces the AWS CAF name collision as a live naming decision he owns. It also answers the "assessment-as-named-product" follow-up parked in [[2026-06-28-snowflake-si-cortex-positioning-caf-gap]].

Open follow-ups

Related

Sources

Vault:

Web (primary-verified — fetched):

Web (secondary — search-result summaries, NOT primary-verified; do not quote to clients):