[{"data":1,"prerenderedAt":75},["ShallowReactive",2],{"term-a\u002Fagent-drift":3,"related-a\u002Fagent-drift":59},{"id":4,"title":5,"acronym":6,"body":7,"category":40,"description":41,"difficulty":42,"extension":43,"letter":44,"meta":45,"navigation":46,"path":47,"related":48,"seo":53,"sitemap":54,"stem":57,"subcategory":6,"__hash__":58},"terms\u002Fterms\u002Fa\u002Fagent-drift.md","Agent Drift",null,{"type":8,"value":9,"toc":33},"minimark",[10,15,19,23,26,30],[11,12,14],"h2",{"id":13},"eli5-the-vibe-check","ELI5 - The Vibe Check",[16,17,18],"p",{},"Agent drift is when you give an agent directions to the grocery store and, three detours later, it's at the hardware store buying an entirely different kind of thing you never asked for. Nobody made one big obviously wrong decision, the agent's actions just slowly stopped matching what you actually meant, one small reasonable-seeming step at a time.",[11,20,22],{"id":21},"real-talk","Real Talk",[16,24,25],{},"Agent drift is the gradual divergence of an agent's behavior from the original task intent over a long-running session, often caused by context rot, small errors compounding across steps, or the agent over-optimizing a proxy signal, like making tests pass, instead of the actual underlying goal. Mitigations include periodic goal re-grounding, checkpoints, and scheduled human review at defined intervals.",[11,27,29],{"id":28},"when-youll-hear-this","When You'll Hear This",[16,31,32],{},"\"By hour three the agent had drifted from 'fix the bug' to 'refactor the whole module'.\" \u002F \"We add a checkpoint every 20 steps specifically to catch agent drift.\"",{"title":34,"searchDepth":35,"depth":35,"links":36},"",2,[37,38,39],{"id":13,"depth":35,"text":14},{"id":21,"depth":35,"text":22},{"id":28,"depth":35,"text":29},"vibecoding","Agent drift is when you give an agent directions to the grocery store and, three detours later, it's at the hardware store buying an entirely different kin...","intermediate","md","a",{},true,"\u002Fterms\u002Fa\u002Fagent-drift",[49,50,51,52],"Context Rot","Human-in-the-Loop","Long-Horizon Task","Reflexion",{"title":5,"description":41},{"changefreq":55,"priority":56},"weekly",0.7,"terms\u002Fa\u002Fagent-drift","ljIanZ-_NqZ7V7rqUp6dWpRWTeYZNlm7VxxeZsoM-vM",[60,64,68,71],{"title":49,"path":61,"acronym":6,"category":62,"difficulty":42,"description":63},"\u002Fterms\u002Fc\u002Fcontext-rot","ai","Context rot is when a long AI conversation has been running so long that the model is confusing old instructions with new ones, forgetting what it already...",{"title":50,"path":65,"acronym":6,"category":62,"difficulty":66,"description":67},"\u002Fterms\u002Fh\u002Fhuman-in-the-loop","beginner","Human-in-the-loop is the 'are you sure?' dialog for AI agents.",{"title":51,"path":69,"acronym":6,"category":62,"difficulty":42,"description":70},"\u002Fterms\u002Fl\u002Flong-horizon-task","There's a big difference between 'make this button blue' and 'migrate our entire billing system to a new provider.",{"title":52,"path":72,"acronym":6,"category":62,"difficulty":73,"description":74},"\u002Fterms\u002Fr\u002Freflexion","advanced","Reflexion is an agent re-reading its own exam answers before the next attempt and catching its own mistake.",1792528059411]