## Core Insight Software engineering runs as a pipeline whose bottleneck has moved upstream. Code production — translating intent into executable instructions — used to be the hard part; with AI now writing over 90% of the code for some teams ("For us it's probably <10%" typed by hand), the constraint shifts to two things: making systems understandable to humans, and making real-world context ingestible by agents. Scott Wu (CEO, Cognition/Devin) argues against the "nihilist" view that better AI makes interfaces irrelevant — the opposite holds: "AI will soon be so good that the way you interact & knowledge-transfer will be the only thing that matters." DeepWiki and DeepWiki MCP exist specifically as context-transfer tools built on this premise. ## Cross-Domain Applications Engineers who make systems understandable — documentation, architecture, clear abstractions — become more valuable than those who excel only at code production. - [[Specification-Driven Development as Multi-Agent Coordination Mechanism]] — specs as formalized context transfer - [[Commoditization of Code Translation]] — code commoditization forces the bottleneck upstream - [[Shallow Familiarity Over Deep Technical Mastery]] — broad context beats deep syntax knowledge - [[Context Transfer as Engineering Bottleneck]] — same source, same thesis, independently synthesized - [[Engineering Judgment as Post-Automation Bottleneck]] — the coding-specific case of the same shift - [[Automation Shifts Bottleneck to Design]] — Norman's automation paradox generalized - [[Generalist Automators Replace Specialized Engineers]] — the organizational-power version of the same shift ## Source - [[Think of software engineering as a pipeline]] — Scott Wu, X/Twitter, 2026-02-12 — https://x.com/scottwu46/status/2021726120027230560