Generative AI Fundamentals Map

Map/Agents and organisations/Wave 7

Correction propagation

A fix made once reaches every agent and stays fixed.

capabilityEmerging Open on the map →

Facts

Lane: Agents and organisations
Wave: 7 of 12 (build order, not a date)
Needs: 3 boxes all the way back through hard links
Unlocks: 3 boxes downstream through hard links
Leverage: 0.4 points · stands in the way of 3 of 53 stages · rank 22 of 57
Cluster: Agents at work

Evidence as of October 2026

The NANDA "learning gap": tools that do not retain feedback keep repeating the same mistakes.

01Why 95% of AI pilots fail, on the MIT NANDA report (VC Cafe)

Around it

Needs · 2

Persistent memorycapability · Emerging
Shared ontology and wikicapability · Emerging

Unlocks · 1

Continual-learning agentsstage · Emerging

Loops and off-ramps

Company knowledge loopreinforcing loop

Openings researchers' judgement, October 2026 scouting

Startup room 68/100 · rank 8 of 124 Research room 73/100 · rank 4 of 124

Crowding1.5 Lab overlap3 Small-team fit4 Buyer today2 Research openness5

Depth: deep
Players: No dedicated vendor; CopilotKit CLiF (Oct 2026); Snowflake VQR; TencentDB Team Memory; Anthropic shared stores; TRACE, ShareMem papers
Wedge: Correction registry: one fix becomes a scoped, reviewed rule with generated tests, pushed to every agent and regression-tested on each model change; no benchmark yet measures fix persistence. startup