A new AI tool launches every day, and developers are chasing each one at the cost of the thing that actually makes them engineers. Here's what the data says about the drift — and how to relocate rigor instead of losing it.
AI coding agents generate thousands of lines in minutes. But someone still has to review it all. Code review — not code generation — is now the bottleneck. Three strategies are emerging to deal with it.
Apps solve problems users already have. Games create experiences users didn't know they wanted. This fundamental difference means the playbooks for building each are not just different — they're inverted.
Four independent teams shipped persistent memory systems for AI coding agents in the same week. The convergence isn't a coincidence — it's the clearest signal yet about what separates useful agents from powerful ones.
Anthropic's Claude Code source code — 512,000 lines — leaked via npm. What the unreleased features tell us about the next 18 months, and 7 actions software engineers should take right now.
MCP hit 97 million installs the same week an AI agent nuked a production database. Three frontier models launched while Apple banned vibe coding apps. March 2026 crystallized the central tension of AI-assisted development.
Anthropic's Generator-Evaluator harness pattern was built for long-running AI coding sessions. But the same architecture maps surprisingly well to the two hardest problems in live-service games: scaling content production and optimizing monetization.