The AI Hype Observatory

Tracking the slogans, buzzwords and predictions the AI field tells itself — and how fast they rise, peak, and die.

September 2026
63
AI Hype Index
Rising
Cooling
New this month
What changed · September 2026: Hype Index 66→63 — a fourth consecutive fall. The naming treadmill forked and its newest rung snapped: graph engineering was refuted within four days of coinage and goes straight to the graveyard, while skill engineering turns out to be the real successor to prompt → context → harness. Meanwhile cost and governance displaced capability as the story — Gartner sees inference cost per agentic workflow rising more than fivefold, KPMG found nearly half of leaders pulling back, and OpenAI paused frontier RL training for two weeks. New: Skill engineering, Software factory, Personal agent, Model welfare, Verification debt, Context rot, Slopaganda, Tokenmaxxing. Risers: AI capex bubble, Evals, AI slop, Physical AI, Lethal trifecta. Coolers: Agentic AI, Year of the agent, Test-time compute, Context engineering, Agent washing. Corrections: last month’s claim that evals had “no peak, no backlash and no rebrand” was wrong, “Agent = Model + Harness” is Osmani’s not Weng’s, and the 17%-deployed figure is Gartner’s April hype cycle, not a July CIO survey.
Panel 01

The Hype Cycle

Every tracked term plotted on the classic curve — from innovation trigger, over the peak of inflated expectations, down into the trough of disillusionment, up the slope of enlightenment, onto the plateau of productivity. Hover any dot.

Panel 02

The Prediction Ledger

The rolling "year of X" calls — and a verdict once reality is in. Accountability is the point.

Panel 03

The Lexicon

Filter by category. Each term carries its stage, momentum, a representative quote, and a link to its primary source.

Panel 04

The Graveyard

Buzzwords that have cooled, been superseded, or quietly retired. Today's slogan is tomorrow's epitaph.

The AI Hype Observatory
Harvested from public AI writing, podcasts, papers and curated X lists.
32 terms tracked. Monthly figures are mention-intensity estimates harvested from public AI writing, podcasts, papers and curated X lists; sparklines run April–September 2026 and month-over-month diffs begin from July 2026.
Curated by © Fred Pelard