# Research

Quarterly state-of-the-art research on fast-moving topics. Each topic is re-researched roughly every three months.

- [Risks and Failure Modes of Agentic Coding Tools](https://gravitalforge.com/research/agentic-coding-risks.md) — updated August 21, 2026: Three independent measurements now attribute agentic coding failures to the harness rather than the model, inverting how these tools should be evaluated — and the field's most-cited productivity number has been retired by its own authors.
- [Agentic Coding Tools on Small Engineering Teams](https://gravitalforge.com/research/agentic-coding-team-usage.md) — updated August 19, 2026: The recommended first move flipped from building a shared skill library to writing the context file: a controlled eval found an AGENTS.md documentation index passed 100% while the same material as a skill passed 53% by default and went uninvoked in 56% of cases.
- [AI in Go-to-Market: From First Use Case to GTM Engineering](https://gravitalforge.com/research/ai-for-gtm.md) — updated July 24, 2026: Initial research
- [The Real State of AI-Assisted Coding](https://gravitalforge.com/research/ai-assisted-coding.md) — updated July 3, 2026: Initial research
- [Agentic Coding Tools for Data Analysis](https://gravitalforge.com/research/agentic-coding-data-analysis.md) — updated July 3, 2026: Initial research
- [Agentic AI in Non-Engineering Business Functions](https://gravitalforge.com/research/agentic-ai-business-functions.md) — updated June 19, 2026: Initial research