
Cortex Workflow: Signal, Shape, Build and Review

Building Tools for Defenders 🛡️ — with Rahul Raina
Topic 6 of 6
Published Sep 18, 2026 · Clip 41:57–59:59 (18:02)
Rahul walks through the end-to-end Cortex workflow. Signals are collected from sales calls, customer success feedback and surveys, correlated by AI agents into a ranked markdown 'signal packet' reviewed by product-focused builders. In the Shape stage, the AI interviews the shaper, is connected to the codebase to surface feasibility, and produces a dynamic mock built on real components deployed in a preview environment — instead of Figma mocks. In the Build stage, engineers task background agents writing code, shipped via preview environments with validation; PMs can comment on PRs or edit branches themselves. A communication cortex auto-generates news posts and training artifacts; maintain agents scan for vulnerabilities weekly and a 'first responder' triages support tickets; a reflect agent updates prompts and harness. On review: an auto-approver classifies ~30% of low-risk changes to go through automatically, while humans review higher-level, human-reviewable artifacts (skills like 'grill me' and 'mind the gap' create tests and visuals) rather than reading every line of code. He argues the industry should build tools to compress and accelerate human understanding, not just code output, citing Linear's concept-level PR review as an example.