{
 "blyg": "0.3",
 "id": "4zzwtstrjze4t0m39ew5h4gw77",
 "kind": "thread",
 "origin": "https://rafael.fyi/blyg/",
 "page": "t/4zzwtstrjze4t0m39ew5h4gw77/",
 "created": "2026-11-16T13:14:24Z",
 "updated": "2026-11-16T14:44:09Z",
 "version": 3,
 "content_md": "# The Last Mile in Post-AI Software\n\nThere are still seven areas where AI hits a bottleneck:\n\n1. **Intent**: deciding what to build, for whom, and what \"done\" looks like.\n    - *Questions:* What's the goal and how do we measure it? What's out of scope? Who makes the final call?\n    - *Examples:* writing the spec and success metrics; choosing which feature ships first; cutting a feature that works but doesn't fit.\n\n2. **Authority**: granting the system access to accounts, credentials and money.\n    - *Questions:* Which systems can the agent read and write? Who issues and rotates the keys? What's the damage if a key leaks?\n    - *Examples:* setting up a key manager and creating scoped API keys; opening cloud and billing accounts; registering the domain, DNS and OAuth apps.\n\n3. **Ground truth**: finding out whether it works for real people in real conditions.\n    - *Questions:* Has a real user completed the core flow? What do production metrics show? What broke that the tests missed?\n    - *Examples:* running sessions with five real customers; watching error rates and logs after launch; running an A/B test on live traffic.\n\n4. **Acceptance**: getting others to choose it, or at least allow it.\n    - *Questions:* Will the buyer pay or switch? Do their security and legal teams approve? Will the platform or regulator allow it?\n    - *Examples:* closing the first paying customer; passing a customer's security questionnaire; getting through app store review.\n\n5. **Accountability**: putting a name on the line for security and failures.\n    - *Questions:* Who signs off on security? Who's on call? Who's liable if data leaks?\n    - *Examples:* commissioning a pen test and signing off the findings; auditing dependencies for supply-chain risk; setting up the on-call rotation and incident plan.\n\n6. **Stopping**: breaking the loop when the work stops paying off, whether through overthinking, perfectionism or going in circles.\n    - *Questions:* Is it still making progress, or repeating itself? Is this good enough to ship? What is another hour worth here?\n    - *Examples:* stopping an agent that keeps reopening the same refactor; shipping a v1 with known rough edges; telling it to stop researching and pick an option.\n\n7. **Comprehension**: understanding what was built, well enough to steer it, explain it and fix it.\n    - *Questions:* Can I explain how it works without asking the AI? What did it decide that I never asked for? If it breaks tonight, does someone know where to look?\n    - *Examples:* reading the diff before it merges; keeping a log of why things were built the way they were; walking a teammate through the system.\n",
 "content_html": "<h1>The Last Mile in Post-AI Software</h1>\n<div class=\"blyg-tk-gen\"><p>There are still seven areas where AI hits a bottleneck:</p>\n<ol>\n<li>\n<p><strong>Intent</strong>: deciding what to build, for whom, and what \"done\" looks like.</p>\n<ul>\n<li><em>Questions:</em> What's the goal and how do we measure it? What's out of scope? Who makes the final call?</li>\n<li><em>Examples:</em> writing the spec and success metrics; choosing which feature ships first; cutting a feature that works but doesn't fit.</li>\n</ul>\n</li>\n<li>\n<p><strong>Authority</strong>: granting the system access to accounts, credentials and money.</p>\n<ul>\n<li><em>Questions:</em> Which systems can the agent read and write? Who issues and rotates the keys? What's the damage if a key leaks?</li>\n<li><em>Examples:</em> setting up a key manager and creating scoped API keys; opening cloud and billing accounts; registering the domain, DNS and OAuth apps.</li>\n</ul>\n</li>\n<li>\n<p><strong>Ground truth</strong>: finding out whether it works for real people in real conditions.</p>\n<ul>\n<li><em>Questions:</em> Has a real user completed the core flow? What do production metrics show? What broke that the tests missed?</li>\n<li><em>Examples:</em> running sessions with five real customers; watching error rates and logs after launch; running an A/B test on live traffic.</li>\n</ul>\n</li>\n<li>\n<p><strong>Acceptance</strong>: getting others to choose it, or at least allow it.</p>\n<ul>\n<li><em>Questions:</em> Will the buyer pay or switch? Do their security and legal teams approve? Will the platform or regulator allow it?</li>\n<li><em>Examples:</em> closing the first paying customer; passing a customer's security questionnaire; getting through app store review.</li>\n</ul>\n</li>\n<li>\n<p><strong>Accountability</strong>: putting a name on the line for security and failures.</p>\n<ul>\n<li><em>Questions:</em> Who signs off on security? Who's on call? Who's liable if data leaks?</li>\n<li><em>Examples:</em> commissioning a pen test and signing off the findings; auditing dependencies for supply-chain risk; setting up the on-call rotation and incident plan.</li>\n</ul>\n</li>\n<li>\n<p><strong>Stopping</strong>: breaking the loop when the work stops paying off, whether through overthinking, perfectionism or going in circles.</p>\n<ul>\n<li><em>Questions:</em> Is it still making progress, or repeating itself? Is this good enough to ship? What is another hour worth here?</li>\n<li><em>Examples:</em> stopping an agent that keeps reopening the same refactor; shipping a v1 with known rough edges; telling it to stop researching and pick an option.</li>\n</ul>\n</li>\n<li>\n<p><strong>Comprehension</strong>: understanding what was built, well enough to steer it, explain it and fix it.</p>\n<ul>\n<li><em>Questions:</em> Can I explain how it works without asking the AI? What did it decide that I never asked for? If it breaks tonight, does someone know where to look?</li>\n<li><em>Examples:</em> reading the diff before it merges; keeping a log of why things were built the way they were; walking a teammate through the system.</li>\n</ul>\n</li>\n</ol></div>",
 "content_hash": "sha256:06a45e951b8361047212cbb3a840ec6c5ddb85b8ae3dba18b619aeffa5e30ed5",
 "media": [],
 "transclusions": [],
 "generated": [
  {
   "sources": [],
   "model": "claude-opus-5-5",
   "at": "2026-11-16T14:44:09Z"
  }
 ],
 "changelog": [
  {
   "version": 1,
   "at": "2026-11-16T13:14:24Z",
   "note": "First version"
  },
  {
   "version": 2,
   "at": "2026-11-16T13:24:30Z",
   "note": "The table is now a list, easier to read on a phone"
  },
  {
   "version": 3,
   "at": "2026-11-16T14:44:09Z",
   "note": "Adds a seventh area: comprehension"
  }
 ]
}
