[404] – Developer Not Found: The Continuing Developer Evolution • Derek Bingham • YOW! 2025

The talk argues that the shift isn’t “AI writes code,” it’s “teams engineer context and process.” Prompt engineering is a dead end at scale; context engineering (what code, rules, architecture, tools, and how much of it to include) determines quality. The practical signal: manage context windows, chunk work into tasks, ruthlessly reset chats, and avoid tool bloat (especially via MCP) that silently degrades output.

Spec-driven development (requirements → design → task plan → implement) is a workable mitigation. It constrains generation, enables verification at each stage, and protects the context window. It’s not magic—humans still must review diffs and own core business logic—but it prevents the “vibe coding” trap where agents refactor beautifully and break everything.

Claims that “AI replaces juniors” are labeled as naive. Juniors are still required for review, maintenance, and to become tomorrow’s seniors. Jevons paradox applies: cheaper code generation predicts more software demand, not less headcount. Leaders should plan for redistribution of effort: less boilerplate, more architecture, verification, and model-change governance.

Hidden risks most will miss: MCP can expand attack surface and context uncontrollably; lazy reliance on agents leads to confident nonsense (ProBake example). Teams need clear guardrails, security posture for tool invocation, and model-change measurement (A/B, evals) because agent quality varies with context and updates.

The bigger shift is organizational: move from typing speed to systems thinking, collaboration, and ethics (cost, energy, water). Decide explicitly what the AI writes (boilerplate, tests, cleanup) versus what humans author (core algorithms, critical decisions).