Generative AI Fundamentals Map

Map/Software/Wave 7

Code review at scale

Checking AI-written code as fast as it is produced.

capabilityEmerging Open on the map →

Facts

Lane: Software
Wave: 7 of 12 (build order, not a date)
Needs: 1 boxes all the way back through hard links
Unlocks: 9 boxes downstream through hard links
Leverage: 2.5 points · stands in the way of 8 of 53 stages · rank 1 of 57
Cluster: Agents at work

Evidence as of October 2026

Writing code is no longer the slow step; reviewing it is. METR found experienced developers slower with AI tools even while they felt faster.

01METR: AI coding can make developers slower (The Decoder)

Around it

Needs · 2

Reliable verifierscapability · Built today
Contamination-resistant evalscapability · Emerging · helps, not required

Unlocks · 2

Software that maintains itselfstage · Speculative
Automated coderstage · Emerging

Brakes

Works against this box

Deskilling from AI usebrake · Emerging

Loops and off-ramps

Review is the new bottleneckbalancing loop

Openings researchers' judgement, October 2026 scouting

Startup room 78/100 · rank 4 of 124 Research room 61/100 · rank 25 of 124

Crowding5 Lab overlap5 Small-team fit2 Buyer today5 Research openness3

Depth: deep
Players: CodeRabbit ($143m C, $1.5bn); Greptile ($50m B, $750m); Qodo ($70m B); Anthropic, OpenAI Codex, Cursor Bugbot+Graphite, Copilot
Wedge: Independent, model-agnostic behavioural verification of agent-written code (tests, runtime, repeat runs) plus a neutral review-quality benchmark and calibration of when to stay silent. startup