Stack-ranked by Confusion Contribution — a blend of how many vendors are shouting the term, how far its meaning drifts between them, and how little a buyer can verify it in a demo. Tap any row for the gloss, a real vendor line, and the translation. Higher = foggier.
The buyer's core complaint, in one grid: who claims what. Red = the vendor actively markets this capability. Amber = adjacent / partial claim. The more red in a column, the less that word helps you tell anyone apart.
What's heating up in the messaging cycle, and what's quietly aging out.
Which vendor leans hardest on the word salad? Each is scored 0–100 on a Buzzword Reliance Score (BRS) — how densely they pack buzzwords, how many they coin or rebrand, how thin the substance runs underneath, and how much they just echo everyone else. Higher = guiltier. Tap a vendor for the rap sheet. Ranks move week to week.
Component key: Density = buzzwords per 100 marketing words · Coinage = new terms + rebrands shipped this week · Substance gap = how far the claims outrun verifiable capability · Me-too echo = share of messaging that just repeats a rival's line.
A new standing feature. The Offender Index measures how much a vendor leans on buzzwords. This measures something harder: whether a given word is even true — whether it actually describes the problem it claims to solve, or whether the label quietly smuggles in a claim the architecture can't support. Sameness tells you a word has no shortlisting value. This tells you a word may be actively misleading. Each issue we put one term on the table.
The claim. In its Top Strategic Technology Trends for 2027 (Oct 16), Gartner says an AI agent's identity is built from changeable parts — model weights, system prompts, tool capabilities and credentials — anchored to a unique workload identity. Because those parts shift, Gartner says enterprises need a "dynamic composite identity" that continuously correlates the agent's role, purpose, intent and behavior at runtime.
What the word gets right. More than most terms on this board. An agent really is different from a service account: change its prompt, swap its model or hand it a new tool and you have a different actor holding the same credential. "Non-human identity" hides that; "composite" names it. The anchor-plus-components picture is the most useful agent-identity model an analyst has published this year.
Where it slips. The second half folds intent and behavior into identity itself. Those are observations about what an identity is doing, not facts about what it is. It's the same collapse we flagged in Saviynt's IARA last issue, coming from the other direction: if identity is continuously re-derived from what the agent appears to intend, then the thing you hold accountable keeps moving. Gartner's own prediction shows the cost: by 2030, it expects agent mutability to "shatter" 75% of traditional compliance frameworks because enterprises can't prove what an agent did. A fluid identity makes that worse, not better.
| The word says | What it actually is | The honest name |
|---|---|---|
| "Composite" | A workload ID plus the model, prompt, tools and scopes the agent runs with | A versioned agent manifest. Accurate and useful: keep it. |
| "Dynamic" | The components change at runtime, so the identity is re-evaluated continuously | Fine if it means re-verify on every change. Risky if it means the identity quietly drifts. |
| "Intent" and "behavior" as identity | Runtime observations of what the agent seems to be doing | Behavioral monitoring — a detection control. Measure it against the identity; don't make it part of the identity. |
Change should trigger re-approval, not redefinition. Treat the composite as a manifest a named human approved: owner, purpose, model, prompt, tools, scopes. When any component changes, that is a new version, and it goes back through approval. Watch behavior against the approved version. That keeps the useful half of Gartner's idea and gives the auditor something fixed to point at. When a vendor says "composite identity," ask: what triggers a new version, and who signs it?
Last issue: Saviynt's "Intent-Aware Runtime Authorization" — verdict: fails the test (intent is evidence, not entitlement). The point of this feature isn't to single out a vendor or an analyst. It's to give buyers a second blade: not just who else says it, but whether the word is even true. Gartner quotes are from "Gartner's Top Strategic Technology Trends for 2027: AI Agent Identity" (G00855975, Oct 16 2026) and are short excerpts for commentary.
The Index is designed as a repeatable ritual, not a one-off. Each Friday, pull the week's vendor press releases, conference keynotes, product-page rewrites, and analyst notes, then score every term that appears on the four sub-metrics below. The composite is a weighted mean, clamped 0–100.
A term earns a spot on the leaderboard when its Confusion Contribution clears 55. New coinages get a NEW flag; a term drops off after two weeks below threshold. Rank moves (▲▼) compare against the prior issue.
Vendor Buzzword Reliance Score (BRS) is scored per vendor, same cadence:
Sources each week: vendor newsrooms (SailPoint, Okta, CyberArk, CrowdStrike, Palo Alto, Microsoft Entra), Black Hat / RSAC / Fal.Con / Oktane / Navigate keynotes, earnings-call transcripts, and independent buyer surveys. Buyer-confusion inputs are anchored to published practitioner data, e.g. the finding that roughly two-thirds of CISOs say aggressive marketing makes real innovation hard to distinguish from AI-washing.
The Buzzword Confusion Index runs in The Perimeter — market intelligence for CIOs and CISOs who have read the word "fabric" one too many times. New issue most weeks.