https://oxagen.sh/blog Research notes on ontologies, AI agents, coding agents, self-improving models, and self-evolving agents. What the papers show, where it breaks, and what it takes to govern it. en-us Wed, 16 Sep 2026 00:00:00 GMT https://oxagen.sh/blog/steering-a-run-you-are-not-watching https://oxagen.sh/blog/steering-a-run-you-are-not-watching Wed, 16 Sep 2026 00:00:00 GMT When a long run goes wrong, most teams can kill it or argue with it. Both are bad. A steer is a third option: a message with a delivery mode, a status, and a record. Autonomous agents AI agents https://oxagen.sh/blog/the-agent-time-horizon-is-doubling https://oxagen.sh/blog/the-agent-time-horizon-is-doubling Wed, 16 Sep 2026 00:00:00 GMT METR measures how long a task an agent can finish on its own. That number has doubled roughly every seven months since 2019. What a week-long run does to supervision. Autonomous agents AI agents https://oxagen.sh/blog/the-problem-is-not-slop-it-is-your-process https://oxagen.sh/blog/the-problem-is-not-slop-it-is-your-process Wed, 16 Sep 2026 00:00:00 GMT The worry about AI slop is about output quality. The measurements point somewhere else: at teams handing an agent a human workflow and expecting it to hold. Autonomous agents AI agents https://oxagen.sh/blog/what-an-agent-should-be-told-before-it-starts https://oxagen.sh/blog/what-an-agent-should-be-told-before-it-starts Wed, 16 Sep 2026 00:00:00 GMT A prompt is a bad container for a standing rule. It has no owner, no date, no scope, and no record that the agent read it. Context records fix the container. Autonomous agents AI agents https://oxagen.sh/blog/the-limits-of-self-correction-and-model-collapse https://oxagen.sh/blog/the-limits-of-self-correction-and-model-collapse Wed, 09 Sep 2026 00:00:00 GMT Self-correction fails without external feedback, and training on your own output degrades the model. What the collapse and verifier research says you should do instead Self-improving models AI agents https://oxagen.sh/blog/deterministic-coding-agents-every-turn-on-the-record https://oxagen.sh/blog/deterministic-coding-agents-every-turn-on-the-record Wed, 09 Sep 2026 00:00:00 GMT A coding agent changed your tree and nobody can replay how. What the research says about execution feedback, sampling variance, and turns you can audit Coding agents AI agents https://oxagen.sh/blog/self-improving-models-from-star-to-self-rewarding https://oxagen.sh/blog/self-improving-models-from-star-to-self-rewarding Wed, 09 Sep 2026 00:00:00 GMT A vendor says the model improves itself. Here is the research lineage behind that claim, what signal drives each loop, and the exact thing that stops it Self-improving models https://oxagen.sh/blog/governing-an-agent-that-rewrites-itself https://oxagen.sh/blog/governing-an-agent-that-rewrites-itself Wed, 09 Sep 2026 00:00:00 GMT An agent that edits its own code needs a gate, not a hope. What the safety literature says about reward hacking, sandboxes, oversight, and typed contracts Self-evolving agents Ontologies https://oxagen.sh/blog/graph-grounded-retrieval-vs-vector-search https://oxagen.sh/blog/graph-grounded-retrieval-vs-vector-search Wed, 09 Sep 2026 00:00:00 GMT Dense retrieval finds the passage that looks right. Graph-grounded retrieval finds the fact that is right. What the research says about the difference Ontologies https://oxagen.sh/blog/self-evolving-agents-what-the-evidence-shows https://oxagen.sh/blog/self-evolving-agents-what-the-evidence-shows Wed, 09 Sep 2026 00:00:00 GMT What actually changes when an agent improves itself, what judges the change, and the measured gain, across eight systems from Voyager to AlphaEvolve Self-evolving agents AI agents https://oxagen.sh/blog/the-science-of-ai-agents-from-react-to-tool-use https://oxagen.sh/blog/the-science-of-ai-agents-from-react-to-tool-use Wed, 09 Sep 2026 00:00:00 GMT Planning, tool use, memory, and reflection each came from a paper that measured something. Here is the research line, and what it still cannot do AI agents https://oxagen.sh/blog/what-an-ontology-buys-an-agent https://oxagen.sh/blog/what-an-ontology-buys-an-agent Wed, 09 Sep 2026 00:00:00 GMT Agents fail confidently when they answer from the wrong context. Here is what classes, relations, constraints, and time-aware facts actually buy you Ontologies AI agents https://oxagen.sh/blog/what-swe-bench-measures-and-what-it-misses https://oxagen.sh/blog/what-swe-bench-measures-and-what-it-misses Wed, 09 Sep 2026 00:00:00 GMT SWE-bench turned coding agents into a leaderboard race. What a resolve rate proves, where test-based grading leaks, and what it cannot tell you Coding agents https://oxagen.sh/blog/why-agents-fail-measuring-reliability-and-cost https://oxagen.sh/blog/why-agents-fail-measuring-reliability-and-cost Wed, 09 Sep 2026 00:00:00 GMT AgentBench, WebArena, GAIA, SWE-bench, and tau-bench measure five different things, and none of them report what a run costs. What that hides AI agents Coding agents