# Agents That Author Their Own Skills

Research increasingly targets agents that improve autonomously by capturing procedures they discover at inference time into reusable, executable skills and co-evolving their own workflows, rather than being reprogrammed or reprompted each run; expect recurring methods pushing self-improvement as an agent capability axis.

- Conviction: 35 / 100 (forming)
- Horizon: Emerging (watchlist)
- Tracking since: 2026-07-28T00:00:00.000Z
- Last updated: 2026-07-28T14:03:00.013Z
- Canonical: https://polylog.news/ai/trends/self-improving-agent-skills
- Publisher: Polylog

## Recent evidence

- [confirms] FlowEvo Proposes Agents That Improve by Co-Evolving Their Workflows and Reusable Skills (2026-07-28): FlowEvo proposes agents that co-evolve their workflows and reusable skills, turning useful procedures found at inference time into executable skills instead of rediscovering them each run. It marks a distinct research thread on agents that author their own capability, separate from demonstration-based or context-efficiency work.