## Core Insight When developers began running several AI agents at once in 2025, a coordination problem appeared that Git was never built to solve. Detailed specifications became the mechanism that holds parallel agents together, because agents need explicit goals and constraints to avoid working at cross purposes. The developer's main output shifts from code to the spec: a document precise enough about intent, constraints, and success criteria that multiple agents can execute in parallel without colliding. This is what "the end of linear work" names. Traditional development ran in sequence: think, code, test, iterate. Multi-agent development runs in parallel: specify, delegate, coordinate, verify. The spec becomes the single source of truth that keeps agents aligned. Invest in spec quality and agent parallelism compounds the return; ship vague specs and it compounds the chaos. ## Evidence - 2025: developers started running multiple AI agents simultaneously - Coordination problems emerged that Git wasn't designed for - Clear specs became crucial because agents need explicit goals and constraints - Specs become "the main way teams align and collaborate" ## Connections - [[Software Engineering Bottleneck Shift from Code Writing to Context Transfer]] — specs as the primary channel for context transfer to agents - [[Systems Factory Identity Shift]] — building specification systems rather than code directly - [[Ghostwriter Author Accountability Model]] — specs as how the "author" keeps accountability - [[Developer as Context-Giver and Observer]] — specifications as the context-giver's primary output - [[Dark Factory as Eval Gated Autonomous Pipeline]] — spec-driven coordination enables autonomous execution with eval gates ## Implications Documentation, once an afterthought, becomes the primary work product. Engineers who write precise specs become more valuable than those who write fast code, and natural-language precision turns into a core engineering skill. Parallel execution pressures teams to formalize previously informal knowledge — architecture decisions, edge cases, performance constraints — into machine-readable specs agents can act on. --- *Source: [[The end of linear work]] — Augment Code, Feb 2026 — https://x.com/augmentcode/status/2021645539264410049*