[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-loading-full-expert-prompts-doesnt-make-ai-agents-smarter":10,"sections":35},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},9449,"loading-full-expert-prompts-doesnt-make-ai-agents-smarter","Loading Full Expert Prompts Doesn't Make AI Agents Smarter","A new study finds detailed profession-specific system prompts cost up to 4.5 times more per call without improving accuracy on science benchmarks.","Giving an AI model a detailed job description doesn't make it smarter at the job - it just makes it slower and more expensive.\n\nResearchers tested Scientific Agents, an open-source set of 503 profession-specific AGENTS.md system prompts, against Gemini 3.8 Flash running in the Pi agent harness. They ran 100 matched profiles across nine science benchmarks covering 4,531 questions, comparing each one to four controls, including a bare \"you are a helpful assistant\" baseline. The detailed profiles generated 1.5 to 2.3 times more output tokens and cost 2.2 to 4.5 times more per call, while moving accuracy by an average of just -0.6 percentage points, a difference indistinguishable from noise. On a bioinformatics tool-use benchmark called BioMysteryBench, the profiles did measurably worse, solving 46.7% of problems versus 56.7% for the plain baseline, because the longer prompts triggered more token- and time-limit cutoffs.\n\nThe one place the detailed profiles helped wasn't expertise, it was resilience against flaky infrastructure. On the SuperGPQA benchmark, frequent API drops left the short baseline with a correct first-pass answer only 54.0% of the time, against 71.6% for the long profile, but generic, unrelated prompts of similar length produced the same benefit. In other words, the gain came from padding, not knowledge, which is a thin excuse for that kind of cost increase.\n\nIf a vendor tries to sell you a library of expert AI personas, ask what you're actually buying: better answers, or just more tokens.","[\"ai-agents\",\"prompt-engineering\",\"benchmarks\",\"llm-research\"]","2026-10-02T04:00:00.000Z","2026-10-02T19:30:10.494Z","2026-10-02T19:30:13.805Z","published",null,[],"ai",[26,27,28,29],"ai-agents","prompt-engineering","benchmarks","llm-research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.00084",0,{"sections":36},[37,40,44,49,54,59,64,69,74,79,84,89,94,99],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",5764,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",831,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",437,"2026-10-01T18:10:00.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",317,"2026-10-01T22:00:00.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",198,"2026-10-01T17:38:48.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Science","science",168,"2026-10-01T18:35:55.000Z",{"name":65,"slug":66,"count":67,"latest_published_at":68},"Consumer Tech","consumer-tech",155,"2026-10-01T19:54:10.000Z",{"name":70,"slug":71,"count":72,"latest_published_at":73},"Dev Tools","dev-tools",96,"2026-10-01T16:57:03.000Z",{"name":75,"slug":76,"count":77,"latest_published_at":78},"Software","software",93,"2026-09-30T21:41:11.000Z",{"name":80,"slug":81,"count":82,"latest_published_at":83},"Startups","startups",90,"2026-10-01T21:55:22.000Z",{"name":85,"slug":86,"count":87,"latest_published_at":88},"Gaming","gaming",53,"2026-10-02T02:50:39.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"General","general",50,"2026-09-30T21:37:54.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"How-To","how-to",7,"2026-10-01T09:00:00.000Z"]