AI engineering is 90% building evals and 10% prompting Claude Code
paulabartabajo_ · x · 2026-10-16
paulabartabajo breaks down the real job of an AI engineer: 90% is building the right evals for the business problem and the evaluation framework, and only 10% is going back and forth with Claude Code to build the software that climbs those evals. The takeaway: in the AI coding era, the craft has shifted from writing code to designing evaluation systems.
More from coding & agent
- Mealsight Open-Sourced: Photo-Based Meal Analysis With a Confidence Gate That Admits Unknowns — HowDevelop · 2026-10-16
- Vitalik builds "paranoid mode" agent wrapper with Tor, zkAPI and local prompt rewriting — DavideCrapis · 2026-10-16
- Reflection releases Beam 1.0, its first open-weight model for coding, reasoning and agents — paf1138 · 2026-10-16
- Agent fine-tunes Qwen3.5-0.8B with synthetic data: 4.7%→76.3% accuracy, 11.9× faster — ben_burtenshaw · 2026-10-16
- Beam-501B agent demo: solo beats 5-agent team on ProgramBench rebuild task — ben_burtenshaw · 2026-10-16
- Beam agent builds BeamSpeak, a local Parakeet dictation app for macOS — ben_burtenshaw · 2026-10-16