https://macanderson.com Mac Anderson is the founder and CEO of Oxagen, the author of Engineering Deterministic AI Coding Agents, and the creator of Stella, an open-source coding agent. He researches and builds coding agents that ship software on their own. en-us https://macanderson.com/research/continuous-delivery-of-fine-tuned-open-weight-modelshttps://macanderson.com/research/continuous-delivery-of-fine-tuned-open-weight-modelsFri, 09 Oct 2026 16:00:00 GMTEvery coding-agent session your team runs produces a trace. Kept and graded by an oracle the agent cannot touch, those traces become the training set for a model you own. This book covers the oracle, the air gap, the one-bit verdict, the harness hooks that collect traces for free, how much data a fine-tune needs, and the pipeline that delivers new weights every week. https://macanderson.com/research/agents-are-waiting-on-a-process-built-for-peoplehttps://macanderson.com/research/agents-are-waiting-on-a-process-built-for-peopleWed, 30 Sep 2026 16:00:00 GMTAn agent can write a change in minutes. Then the change waits for a person to read it. This post measures that wait, what eight companies changed about it, and how Oxagen ships at every hour. https://macanderson.com/manualhttps://macanderson.com/manualSat, 19 Sep 2026 16:00:00 GMTA field manual in 21 parts, by Mac Anderson of Oxagen. Parts 1 to 13 show how to build the deterministic system around a coding agent, with published evidence in each part. Parts 14 to 20 show how to operate the agents you now run: identity, authority, budget, equipment, and a record another person can read. Every part ends with steps for this week and metrics to track. https://macanderson.com/research/steering-a-run-you-are-not-watchinghttps://macanderson.com/research/steering-a-run-you-are-not-watchingWed, 16 Sep 2026 16:00:00 GMTWhen a long run goes wrong, most teams can stop it or type at it. Both work badly. A steer is a third option. It is a message with a delivery mode, a status, and a record. https://macanderson.com/research/the-agent-time-horizon-is-doublinghttps://macanderson.com/research/the-agent-time-horizon-is-doublingWed, 16 Sep 2026 16:00:00 GMTMETR measures how long a task an agent can finish on its own. That length has doubled about every seven months since 2019. This post covers what week-long runs mean for supervision. https://macanderson.com/research/the-problem-is-not-slop-it-is-your-processhttps://macanderson.com/research/the-problem-is-not-slop-it-is-your-processWed, 16 Sep 2026 16:00:00 GMTThe worry about AI slop is about output quality. The measurements point to a different cause. Teams give an agent a workflow built for people and expect it to work. https://macanderson.com/research/what-an-agent-should-be-told-before-it-startshttps://macanderson.com/research/what-an-agent-should-be-told-before-it-startsWed, 16 Sep 2026 16:00:00 GMTA prompt file has no owner, no date, no scope, and no record that the agent read it. So it is a poor place for a standing rule. Steering records give each rule those things. https://macanderson.com/research/the-limits-of-self-correction-and-model-collapsehttps://macanderson.com/research/the-limits-of-self-correction-and-model-collapseWed, 09 Sep 2026 16:00:00 GMTWithout outside feedback, self-correction fails, and a model trained on its own output gets worse. This post covers what the collapse and verifier research says to do instead. https://macanderson.com/research/deterministic-coding-agents-every-turn-on-the-recordhttps://macanderson.com/research/deterministic-coding-agents-every-turn-on-the-recordWed, 09 Sep 2026 16:00:00 GMTA coding agent changed your code and no one can replay how. What the research says about feedback from running code, random sampling, and turns you can audit. https://macanderson.com/research/self-improving-models-from-star-to-self-rewardinghttps://macanderson.com/research/self-improving-models-from-star-to-self-rewardingWed, 09 Sep 2026 16:00:00 GMTA vendor says the model improves itself. This is the research behind that claim, the signal that drives each training loop, and what stops each one. https://macanderson.com/research/governing-an-agent-that-rewrites-itselfhttps://macanderson.com/research/governing-an-agent-that-rewrites-itselfWed, 09 Sep 2026 16:00:00 GMTAn agent that edits its own code needs the same review as any other change. What safety research says about reward hacking, sandboxes, oversight, and typed contracts. https://macanderson.com/research/graph-grounded-retrieval-vs-vector-searchhttps://macanderson.com/research/graph-grounded-retrieval-vs-vector-searchWed, 09 Sep 2026 16:00:00 GMTVector search finds the passage that looks like your question. Graph-grounded retrieval finds the fact that answers it. What the research says about the difference. https://macanderson.com/research/self-evolving-agents-what-the-evidence-showshttps://macanderson.com/research/self-evolving-agents-what-the-evidence-showsWed, 09 Sep 2026 16:00:00 GMTWhat changes when an agent improves itself, what checks the change, and the measured gain, across eight systems from Voyager to AlphaEvolve. https://macanderson.com/research/the-science-of-ai-agents-from-react-to-tool-usehttps://macanderson.com/research/the-science-of-ai-agents-from-react-to-tool-useWed, 09 Sep 2026 16:00:00 GMTPlanning, tool use, memory, and reflection each come from a paper that measured something. This post traces those papers and what agents still cannot do. https://macanderson.com/research/what-an-ontology-buys-an-agenthttps://macanderson.com/research/what-an-ontology-buys-an-agentWed, 09 Sep 2026 16:00:00 GMTAn agent can answer with confidence from the wrong context. This post covers what classes, relations, constraints, and dated facts add to an agent's answers. https://macanderson.com/research/what-swe-bench-measures-and-what-it-misseshttps://macanderson.com/research/what-swe-bench-measures-and-what-it-missesWed, 09 Sep 2026 16:00:00 GMTCoding agents are ranked by their SWE-bench resolve rate. This post covers what that rate shows, where test-based grading goes wrong, and what the rate cannot tell you. https://macanderson.com/research/why-agents-fail-measuring-reliability-and-costhttps://macanderson.com/research/why-agents-fail-measuring-reliability-and-costWed, 09 Sep 2026 16:00:00 GMTAgentBench, WebArena, GAIA, SWE-bench, and tau-bench each measure a different thing. None of them reports what a run costs. This post covers what that hides.