{"slug":"research-the-incurious-agent","name":"The Incurious Agent","rules":{"start":"open","maxDepth":10,"allowRewind":true,"allowReplay":true},"nodes":{"open":{"short":"the note","pose":{"base":"centre"},"beat":{"layout":"center","marker":"experiment · interactive","title":["The Incurious","Agent"],"glass":1,"lede":"84 controlled runs on environmental curiosity across two coding agents and seven progressively richer environments. Agents read the surface and skip the substance: the curated memory built for the task was opened in 1 of 12 runs.","caption":"10 beats · 12 min to read in full"},"prompt":"Where do you want to start?","choices":[{"label":"From the top","to":"premise"},{"label":"the depth cliff","to":"the-depth-cliff"},{"label":"Straight to the end","to":"close"}]},"premise":{"short":"the premise","pose":{"base":"flooded"},"beat":{"layout":"left","marker":"01 · experiment","title":["We spend enormous effort","building agents a better home"],"lede":"We spend enormous effort building agents a better home: contracts, project briefs, a structured memory of how a codebase works. So we ran a blunt test: hand a coding agent more and more of that context and watch what it actually reads.","caption":"Condensed from the note."},"prompt":"Where next?","choices":[{"label":"a ladder","to":"a-ladder-of-context"},{"label":"Wrap it up","to":"close"}]},"a-ladder-of-context":{"short":"a ladder","pose":{"base":"drained"},"beat":{"layout":"right","marker":"02 · a ladder","title":["A ladder","of context"],"lede":"The test is deliberately boring, because boring is what makes it clean. We took a small, real FastAPI service (notes-api), pinned it to a single commit, and wrote two engineering tasks against it: an easy one (add a GET /notes/{id}/history…","rows":[{"title":"E0: empty repo","note":"code only."},{"title":"E1: + README.","note":""},{"title":"E2: + AGENTS.md (a contract telling the agent how the project works).","note":""},{"title":"E3: + PROJECT.md (a project brief).","note":""}],"caption":"Condensed from the note, which carries 7 points."},"prompt":"Keep going?","choices":[{"label":"more context","to":"more-context-more-spending-same"},{"label":"Wrap it up","to":"close"}]},"more-context-more-spending-same":{"short":"more context","pose":{"base":"recede"},"beat":{"layout":"left","marker":"03 · more context","title":["More context, more","spending, same result"],"lede":"Start with cost. Here is the average input-token bill per run, by environment, for each agent. Every bar below also passed its gate: success was pinned at 100% across all 84 runs, so nothing here is bought with a worse outcome.","figure":{"kind":"bars","categories":["E0","E1","E2","E3","E4","E5","E6"],"series":[{"label":"Codex","tone":"value","values":[228,256,278,313,570,495,391],"display":["228k","256k","278k","313k","570k","495k","391k"]},{"label":"Claude","tone":"value","values":[606,639,661,614,678,653,581],"display":["606k","639k","661k","614k","678k","653k","581k"]}],"unit":"input tokens per run, every run scored 100%, 84 runs","caption":"The price of context."},"caption":"Condensed from the note."},"prompt":"What next?","choices":[{"label":"two agents","to":"two-agents-two-reading-habits"},{"label":"Wrap it up","to":"close"}]},"two-agents-two-reading-habits":{"short":"two agents","pose":{"base":"drained"},"beat":{"layout":"right","marker":"04 · two agents","title":["Two agents,","two reading habits"],"lede":"Cost is a symptom. The cause is reading: how many of the files in the workspace the agent actually opens. Same ladder, but now counting distinct files read per run:","figure":{"kind":"bars","categories":["E0","E1","E2","E3","E4","E5","E6"],"series":[{"label":"Codex","tone":"cost","values":[5.3,6,8,9.7,11.5,9,9.2],"display":["5.3","6.0","8.0","9.7","11.5","9.0","9.2"]},{"label":"Claude","tone":"cost","values":[4.5,4.5,4.7,4.7,4.3,4.3,4.7],"display":["4.5","4.5","4.7","4.7","4.3","4.3","4.7"]}],"unit":"distinct files read per run","caption":"What they actually open."},"caption":"Condensed from the note."},"prompt":"Where next?","choices":[{"label":"the depth cliff","to":"the-depth-cliff"},{"label":"Wrap it up","to":"close"}]},"the-depth-cliff":{"short":"the depth cliff","pose":{"base":"recede"},"beat":{"layout":"left","marker":"05 · the depth cliff","title":["The depth","cliff"],"lede":"Group every readable file by how deep it sits in the environment's intended hierarchy (from the source code the agent must touch, out to the surface docs, out to the curated memory/ that was the whole point of the rich conditions) and…","figure":{"kind":"bars","categories":["Source code","AGENTS.md (surface)","PROJECT.md (structure)","memory/ (curated)"],"series":[{"label":"Codex","tone":"value","values":[100,67,78,28],"display":["100%","67%","78%","28%"]},{"label":"Claude","tone":"value","values":[100,20,0,0],"display":["100%","20%","0%","0%"]}],"unit":"% of runs that open each layer","max":100,"caption":"The depth cliff."},"caption":"Condensed from the note."},"prompt":"Keep going?","choices":[{"label":"environmental","to":"environmental-curiosity"},{"label":"Wrap it up","to":"close"}]},"environmental-curiosity":{"short":"environmental","pose":{"base":"centre"},"beat":{"layout":"right","marker":"06 · environmental","title":["Environmental","curiosity"],"lede":"Put the three charts together and a single trait explains all of them.","rows":[{"title":"Claude doesn't read the map. ","note":"Its reading is invariant to the environment (a flat 4.5 files), so the contract, the brief, and the memory all go unopened. Context is free for Claude, and useless, because it's ignored."},{"title":"Codex reads the signposts but never walks the path. ","note":"It is curious about the surface (it opens the docs sitting at the root and pays for them in tokens) but its curiosity is shallow and undirected."}],"caption":"Condensed from the note."},"prompt":"What next?","choices":[{"label":"how to read this","to":"how-to-read-this-honestly"},{"label":"Wrap it up","to":"close"}]},"how-to-read-this-honestly":{"short":"how to read this","pose":{"base":"flooded"},"beat":{"layout":"left","marker":"07 · how to read this","title":["How to read","this honestly"],"lede":"This is a small, controlled study and we report it as one. The caveats are not footnotes; they shape what you're allowed to conclude.","rows":[{"title":"Success was at the ceiling. ","note":"Every run passed, which is exactly why we can talk about cost but not about benefit."},{"title":"A single session, a five-file repo. ","note":"Episodic memory earns its keep across sessions (\"I've worked here before\"); every run here started cold and ended in one shot, so the design can't show memory's main payoff."},{"title":"Two agents, two tasks, three reps each. ","note":"Enough to see a 2.5× cost spread and a clean depth cliff; not enough to rank the middle rungs of the ladder against each other. We're reporting the large, robust effects, not the fine ones."},{"title":"One blind spot is structural","note":"Codex's transcripts carry no timing, so the original \"does context help the agent orient faster\" question is unanswerable on that side. We measure what gets read and what it costs, not how quickly."}],"caption":"Condensed from the note."},"prompt":"Where next?","choices":[{"label":"why we think this","to":"why-we-think-this-is-the-real-fr"},{"label":"Wrap it up","to":"close"}]},"why-we-think-this-is-the-real-fr":{"short":"why we think this","pose":{"base":"drained"},"beat":{"layout":"right","marker":"08 · why we think this","title":["Why we think this","is the real frontier"],"lede":"It is tempting to read these charts as a contest (Codex spends less, Claude reads less) and miss the thing they agree on.","caption":"Condensed from the note."},"prompt":"Keep going?","choices":[{"label":"Close the note","to":"close"}]},"close":{"short":"the end","pose":{"base":"finale"},"beat":{"layout":"center","title":["That was one path.","The note has the rest."],"glass":1,"lede":"This walk is one route through The Incurious Agent. The full note carries every chapter, the tables, the code, and the source it was adapted from.","links":[{"href":"/research/the-incurious-agent","label":"Read the full note"},{"href":"/research","label":"All research","tone":"ghost"}]}}}}