{
  "blyg": "0.3",
  "id": "4j61xh32f3bkthvhp0mgakxmc9",
  "kind": "fragment",
  "origin": "https://www.msweet.net/notes/",
  "page": "862-amber-lynxes",
  "author": {
    "name": "Matthew McDowell-Sweet",
    "url": "https://www.msweet.net/"
  },
  "created": "2026-11-15T09:19:36Z",
  "updated": "2026-11-15T09:19:36Z",
  "version": 1,
  "content_md": "Perhaps we see a natural divide emerge between machine agents and human ones; machine agents become managers whilst humans remain individual contributors, with all the tensions and opportunities that that brings.\n\n> No model is a clear explorer. Most runs pick an idea early and stay with it for the full 24 hours. This matches Fork and Flush, where single runs mostly stay inside one basin for their whole budget, and Scaling Discovery Requires Forgetting, where agents become anchored to their own past attempts. Both reduce this by periodically resetting the agent's context and continuing from the best result so far.\n\n[How models explore ideas · Bespoke Labs ↗](https://bespokelabs.ai/blog/how-models-explore-ideas)",
  "content_html": "<p>Perhaps we see a natural divide emerge between machine agents and human ones; machine agents become managers whilst humans remain individual contributors, with all the tensions and opportunities that that brings.</p>\n<blockquote><p>No model is a clear explorer. Most runs pick an idea early and stay with it for the full 24 hours. This matches Fork and Flush, where single runs mostly stay inside one basin for their whole budget, and Scaling Discovery Requires Forgetting, where agents become anchored to their own past attempts. Both reduce this by periodically resetting the agent's context and continuing from the best result so far.</p></blockquote>\n<p class=\"source\"><a href=\"https://bespokelabs.ai/blog/how-models-explore-ideas\">How models explore ideas · Bespoke Labs ↗</a></p>",
  "content_hash": "sha256:1265780851ac4bcab183163c90ec92f56d12abb1f2e42b601f6cb3663fefd0f7",
  "media": [],
  "changelog": [
    {
      "version": 1,
      "at": "2026-11-15T09:19:36Z",
      "note": null
    }
  ]
}
