{
  "$type": "site.standard.document",
  "content": "---\ntitle: \"On Agentic Tools and Lock-in\"\ndescription: \"A response to Lars Faye's 'Agentic Coding is a Trap': LLMs have the lowest\n  vendor lock-in of any tool I've used in 20 years.\"\ntags: [ai]\n---\n\nI read Lars Faye's\n[Agentic Coding is a Trap](https://larsfaye.com/articles/agentic-coding-is-a-trap)\npost this morning. It engages with what I think is the core question around\nagentic software development, and especially with how we teach others (and\njuniors) to do it effectively. My paraphrase:\n\n> how _necessary_ is the struggle of implementation in delivering good software?\n\nHonestly, I don't think anyone has a precise handle on the answer. My gut feel\nis that it's _somewhat_ necessary, in the same way that I'm still teaching my\nkids arithmetic even though we have ubiquitous calculating devices. But I also\nthink that some of the four-Yorkshiremen-style \"in my day we walked 15 miles in\nthe snow, uphill both ways\" stuff I read is probably a little overblown.\nAlthough my feeling there is no less based on vibes and experience than the\nclaim itself.\n\nI do think it's worth making one other point about the\n[Vendor Lock-In section](https://larsfaye.com/articles/agentic-coding-is-a-trap#vendor-lock-in)\nof the article. Of all the software tools and techniques I've invested\nsignificant time in over my 20-year career as a researcher and developer, LLMs\nhave some of the lowest lock-in I've ever encountered. Seriously, I was more\nlocked in to Emacs[^emacs] than I am to Claude (even though I'm quite happy\nwith Claude at the moment). All the tooling I've built is just md files with\nhuman-readable instructions in them, and switching to a different coding\nharness is, in general, just a file rename away (`mv CLAUDE.md AGENTS.md`).\n\nSo while it's a pain when Claude has an outage, if they were really not working\nfor me I'd let my monthly subscription lapse, and I reckon I'd be just as\nproductive with a new platform within about one hour (partially because the new\nprovider's model would help with the migration).\n\nIt's not to say that I don't read and nod along with Lars's post. I do, if for\nno other reason than that I grieve the loss of the feeling of mastery that\ncomes with being a code-slinging wizard (something I'm now realising was at\nleast a little load-bearing for my work identity). But in the scheme of tech\nand software dev, LLMs are the first thing in perhaps my whole career that I\nthink might actually live up to the hype. They also have the least lock-in of\nanything I've used (cf Google's famous\n[we have no moat and neither does anyone else](https://newsletter.semianalysis.com/p/google-we-have-no-moat-and-neither)\npost). And that makes me a little less nervous about diving in.\n\n[^emacs]:\n    Although regular readers will note that I\n    [broke free of this a year ago](/blog/2026/02/18/ben-s-dev-setup-2026-edition/).\n",
  "createdAt": "2026-05-13T23:14:35.659Z",
  "description": "A response to Lars Faye's 'Agentic Coding is a Trap': LLMs have the lowest vendor lock-in of any tool I've used in 20 years.",
  "path": "/blog/2026/05/04/on-agentic-tools-and-lock-in",
  "publishedAt": "2026-05-04T00:00:00.000Z",
  "site": "at://did:plc:tevykrhi4kibtsipzci76d76/site.standard.publication/self",
  "tags": [
    "ai"
  ],
  "textContent": "A response to Lars Faye's 'Agentic Coding is a Trap': LLMs have the lowest vendor lock-in of any tool I've used in 20 years.",
  "title": "On Agentic Tools and Lock-in"
}