{
"$type": "site.standard.document",
"bskyPostRef": {
"cid": "bafyreibqlc4ihjrx4db73i425mpaopsbj6je5d5h6xdnu6gsmieyzea5oe",
"uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mjrkezwcnek2"
},
"path": "/t/why-llm-agents-keep-failing-and-it-s-not-the-prompt/175361#post_1",
"publishedAt": "2026-04-18T12:01:56.000Z",
"site": "https://discuss.huggingface.co",
"tags": [
"Zenodo",
"Beyond Prompting: Decoupling Cognition from Execution in LLM-based Agents...",
"https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6600840"
],
"textContent": "Most LLM agent failures I’ve seen share the same pattern:\n\nThey don’t break because of the model.\nThey don’t break because of the prompt.\n\nThey break because we force the system to “figure everything out” from scratch on every interaction.\n\n* * *\n\nIn traditional software, we don’t rebuild logic every time we run a function.\nWe define structure, reuse components, and control execution.\n\nWith LLM agents, we’re doing the opposite.\n\n* * *\n\nThis led me to explore a different approach:\n\n-> What if reasoning wasn’t embedded in prompts,\n→ but structured and executed as reusable components?\n\nThat’s the idea behind ORCA — a cognitive runtime for LLM agents.\n\n* * *\n\nI’ve put the full concept in a paper in zenodo and now also on SSRN:\n\nZenodo\n\n### Beyond Prompting: Decoupling Cognition from Execution in LLM-based Agents...\n\nRecent advances in large language model (LLM) agents have largely relied on prompt-centricdesigns, where complex tasks are executed through monolithic, single-shot or loosely structuredprompting strategies. While effective in some settings, this...\n\nhttps://papers.ssrn.com/sol3/papers.cfm?abstract_id=6600840\n\nCurious if others are hitting the same limits with prompt-based systems.",
"title": "Why LLM agents keep failing (and it’s not the prompt)"
}