Code Pilot
- Industry
- Software
- Founded
- 2020
- Evaluation
- Production workflow study
Code Pilot was evaluated on realistic software tasks using held-out inputs, independent scorers, and reproducible model configurations. The report separates product capability from the underlying model and records every source, artifact, and scoring version.
Performance Summary
Scores combine task completion, groundedness, and domain-specific quality criteria. Confidence intervals reflect item-level resampling.
Aggregate weighted score
78.4%
+6.1 vs category meanTask accuracy
81.2%
95% CI ± 2.4Source authority
74.8%
Reviewed across 200 tasksUnsupported claims
3.1%
Lower is betterKey Strengths
Traceable work. Outputs consistently cite the source artifacts used to reach a conclusion.
Reliable task completion. Multi-step tasks complete with fewer abandoned or partially satisfied requirements than the category mean.
Clear failure boundaries. The product surfaces uncertainty instead of manufacturing unsupported details.