{
  "id": "6c8c4e87395b",
  "url": "https://interconnected.org/home/2025/11/28/plumbing",
  "title": "Context plumbing (Interconnected)",
  "note": "As of right now, it seems to me that context management is where AI efforts will go from good to great. Think about it — everyone has access to the same LLMs. We can all write prompts, and regardless of how tricky prompt engineering may be it ultimately can be iterated quickly. Note, LLM's are great at helping you with prompt engineering. So meta. \n\nBut context is a whole different thing.\n\n> So the job of making an agent run really well is to move the context to where it needs to be.\n> \n> Essentially copying data out of one database and putting it into another one -- but as a continuous process.\n> \n> You often don't want your AI agent to have to look up context every single time it answers intent. That's slow. If you want an agent to act quickly then you have to plan ahead: build pipes that flow potential context from where it is created to where it's going to be used.\n\nYou see, you have to first figure out if you have the right context. You then have to make sure it isn't enough. You need to make sure that context is up-to-date. You need to move that context and preserve it. There is a lot of things to do here. And if you do it well, your prompt in the LLM will do amazing things. If you do it poorly, you'll get okay results. \n\nSometimes context is interchanged with data. I don't know that that is right. Some context is data, but not all of it. For example, the windows on your computer screen right now as you read this is important context. Nobody is putting that in a database.\n\nSo, strategically it may be that the benefit you can achieve from LLMs is tuned to how well you can understand and manage context. \n",
  "voice": "mixed",
  "saved": "2025-12-01T23:04:04Z",
  "host": "interconnected.org",
  "kind": "page",
  "tags": [],
  "record": "https://links.thingelstad.xyz/links/6c8c4e87395b/",
  "markdown": "https://links.thingelstad.xyz/links/6c8c4e87395b.md",
  "issues": [
    {
      "number": 335,
      "slug": "335",
      "title": "Weekly Thing 335 / Complexity, Fizzy, Soul",
      "date": "2025-12-06T12:00:00Z",
      "section": "Notable",
      "position": 0,
      "url": "https://weekly.thingelstad.com/archive/335/",
      "page": "https://links.thingelstad.xyz/issues/335/"
    }
  ],
  "connections": {
    "same_host": {
      "count": 9,
      "page": "https://links.thingelstad.xyz/hosts/interconnected.org/",
      "nearest": [
        {
          "id": "79b3d0a492bc",
          "title": "Diane, I wrote a lecture by talking about it (Interconnected)",
          "url": "https://interconnected.org/home/2025/03/20/diane",
          "record": "https://links.thingelstad.xyz/links/79b3d0a492bc/"
        },
        {
          "id": "3cbad391fde4",
          "title": "AI-operated vending machines and business process innovation (sorry) (Interconnected)",
          "url": "https://interconnected.org/home/2025/07/10/vending",
          "record": "https://links.thingelstad.xyz/links/3cbad391fde4/"
        },
        {
          "id": "c0cf7053d958",
          "title": "What I think about when I think about Claude Code",
          "url": "https://interconnected.org/home/2025/09/12/claude",
          "record": "https://links.thingelstad.xyz/links/c0cf7053d958/"
        },
        {
          "id": "8e7e890301d8",
          "title": "mist: Share and edit Markdown together, quickly (new tool) (Interconnected)",
          "url": "https://interconnected.org/home/2026/02/12/mist",
          "record": "https://links.thingelstad.xyz/links/8e7e890301d8/"
        },
        {
          "id": "105f06c6907d",
          "title": "Headless everything for personal AI",
          "url": "https://interconnected.org/home/2026/04/18/headless",
          "record": "https://links.thingelstad.xyz/links/105f06c6907d/"
        }
      ]
    },
    "same_week": {
      "count": 18,
      "nearest": [
        {
          "id": "6bfe04e0bd79",
          "title": "Tiger Style",
          "url": "https://tigerstyle.dev/",
          "record": "https://links.thingelstad.xyz/links/6bfe04e0bd79/"
        },
        {
          "id": "c64bc55e1cbf",
          "title": "Brief thoughts on the recent Cloudflare outage – Surfing Complexity",
          "url": "https://surfingcomplexity.blog/2025/11/26/brief-thoughts-on-the-recent-cloudflare-outage/",
          "record": "https://links.thingelstad.xyz/links/c64bc55e1cbf/"
        },
        {
          "id": "5f096b9b71ce",
          "title": "Seeing like a software company",
          "url": "https://www.seangoedecke.com/seeing-like-a-software-company/",
          "record": "https://links.thingelstad.xyz/links/5f096b9b71ce/"
        },
        {
          "id": "c91153d296ea",
          "title": "Effective harnesses for long-running agents Anthropic",
          "url": "https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents",
          "record": "https://links.thingelstad.xyz/links/c91153d296ea/"
        },
        {
          "id": "ccaa6d1d09b3",
          "title": "Fizzy – Kanban as it should be. Not as it has been.",
          "url": "https://www.fizzy.do",
          "record": "https://links.thingelstad.xyz/links/ccaa6d1d09b3/"
        }
      ]
    },
    "mentions": [],
    "mentioned_by": [],
    "saved_twice": []
  },
  "elsewhere": {
    "wayback": "https://web.archive.org/web/*/https://interconnected.org/home/2025/11/28/plumbing"
  }
}