Code up 2,450%, releases only 30%: AI Engineer keynote says confidence, not code, is the bottleneck
0xkarasy · x · 2026-11-06
After watching Tariq Shaukat's AI Engineer World Fair keynote, cassidoo highlights the line "code is abundant, confidence is scarce."
Key data points:
- Agentic coding has pushed lines of code up 2,450% and commits up 240%, but production releases are only up 30% — the bottleneck sits late in the SDLC.
- Human review isn't a silver bullet: reviewers agree with AI 93% of the time when it's right, but still 80% of the time when it's wrong.
- LLM-as-a-judge fares no better: identical prompts get different verdicts 7–17% of the time, and asking "are you sure?" flips answers 7–34% of the time.
The talk frames the gap with a confidence formula: the industry's scarce resource is trustworthiness, not code volume.
Related event: AI Engineer Conference: Trust, Not Code, Is the New Bottleneck(2 posts)→
More from coding & agent
- OpenClaw Ecosystem Hits 47 Products; Users Say Agents Outgrew the Chatbot Label — heyneighbor · 2026-11-06
- Full workflow: idea to finished AI video with Claude Code/Codex + Seedance, minimal re-rolls — dotey · 2026-11-06
- New vibe coder replicates Autodesk Maya in 3 hours with ChatGPT, asks for better workflow — Kiiaro · 2026-11-06
- Anthropic engineer confirms he's officially building Claude Code Desktop — maxleiter · 2026-11-06
- "Did your agents patch today's CVE?": dev shows agent running in the dark — aniketmaurya · 2026-11-06
- Dev fixes trivial bugs just to stop Claude nagging about them — jmugan · 2026-11-06