OpenClaw's creator argues that 80% of apps will disappear once personal agents run locally with full desktop access. The demo is compelling. The missing guardrails are the real story.

Coding agents aren't winning because of better models — they're winning because CLI-based tools like Claude Code manage context better than any IDE. The real productivity unlock comes from sub-agent architecture, aggressive context clearing, and treating tests as the verification loop that lets agents run fast without breaking everything.

Most MCP servers expose raw REST endpoints to agents that can't afford to browse them. Designing agent-native tool surfaces — curated, outcome-oriented, token-aware — separates production-grade integrations from expensive handshake failures.

Amazon Kiro replaces ad-hoc prompting with a spec-driven workflow: structured EARS requirements, correctness properties, and property-based tests. The result is AI-generated code you can actually verify against its original intent.

Stanford research across 120k developers shows median AI coding ROI of just 10%, despite millions in tool spending. The variance between teams is massive—and telling.

Coding evals jumped from single lines to full codebases in four years. Static benchmarks miss the production gap that destroys rollout timelines.

Lessons from autonomous driving reveal how to make AI agents robust: focus on infrastructure, feedback loops, and simulation over model selection.

AI coding productivity gains evaporate at enterprise scale. Bloomberg's deployment across 9,000+ engineers reveals why platform thinking matters more than tool quality.

Building AI agents for non-technical users requires unsupervised autonomy, not just longer runtimes or better orchestration.

Kitze draws a sharp line between vibe coding and vibe engineering. The difference isn't the tool. It's whether you can judge when code is good enough. That judgment is the new core skill.