{ "id": "9f8b436f5322", "url": "https://www.robert-glaser.de/when-everyone-has-ai-and-the-company-still-learns-nothing/", "title": "When everyone has AI and the company still learns nothing", "note": "Getting AI tools to your company and getting individual use out of them is not the hard thing, it is figuring out how your organization learns and improves with this capability. I'm seeing this right now because people are bringing a typical enterprise mindset to token consumption. Buying AI tools isn't expensive, but tokens are. And if you approach that with a desire to minimize token usage, you are going to destroy your potential. The question should **not** be token consumption, but instead value creation over token consumption. This article suggests an addition that I love — learning acquired!\n\nI'm quoting at length here because I think this is so important.\n\n> I keep coming back to three capabilities companies will need in the messy middle.\n> \n> 1. Agent Operations: which agents and AI tools are running, what systems they can touch, which data they can see, which actions require approval, where identity, audit, permissions, and runtime visibility live. This is the control side, and it matters because agentic work eventually touches real systems.\n> 2. Loop Intelligence: which AI-assisted (or fully agentic) loops actually produce learning, which ones stay open, which ones decay, where agents create leverage, where they sprawl into side quests, which teams are stuck in tight supervision because they lack tests, context, or intuition. Which teams are ready for looser delegation.\n> 3. Agent Capabilities: how useful capabilities get distributed across the organization without pretending that three monolithic agents can do everyone's work. AI is starting to behave more like a fluid base technology than a single application category. It does not fit cleanly into one \"HR agent,\" one \"engineering agent,\" one \"sales agent,\" each sitting somewhere in the enterprise zoo. The better question is how capabilities flow into the places where work happens: employee harnesses, background agents, product teams, platform services, local skills, MCP servers, evaluation suites, runbooks, examples, and domain-specific procedures.\n> \n> This is where the platform question gets interesting. Who owns these capabilities? How does a useful agent skill discovered in one team become available to others without turning into a dead template? How do you enrich a developer's harness differently from a product person's harness, a support team's background agent, or a compliance workflow? Which capabilities belong close to the team, which belong in a platform layer, and which should never be generalized because the local context is the whole point?\n\nHow you operationalize this in a company is not solved at this time. I'm seeing it myself and seeing others wrestle with it. I like the foundational elements that this article puts forward. \n\nAnother model that I've considered is the transition from capital with servers to ephemeral infrastructure in the cloud. In a way capital is a lot like your team. It is hard to move around, hard to acquire, hard to change, but also incredibly powerful for the right things. The cloud made a part of that very flexible and just-in-time. We are already using different LLM models to bring just the right amount of intelligence to a task. And like the cloud we know that required a certain cost in tokens which is easy to know.\n\nThis morning I was doing a weekly review on one of my agents and Claude Code used the logging I have to show me five different skills the agent has and, to the penny, how much each of those cost me over the last week. That visibility applied to other tasks allows us to get much smarter.\n", "voice": "mixed", "saved": "2026-05-07T18:28:53Z", "host": "robert-glaser.de", "kind": "page", "tags": [], "record": "https://links.thingelstad.xyz/links/9f8b436f5322/", "markdown": "https://links.thingelstad.xyz/links/9f8b436f5322.md", "issues": [ { "number": 347, "slug": "347", "title": "Weekly Thing 347 / Scrum, FilamentHound, DO_NOT_TRACK", "date": "2026-05-10T12:00:00Z", "section": "Notable", "position": 1, "url": "https://weekly.thingelstad.com/archive/347/", "page": "https://links.thingelstad.xyz/issues/347/" } ], "connections": { "same_host": { "count": 0, "page": "https://links.thingelstad.xyz/hosts/robert-glaser.de/", "nearest": [] }, "same_week": { "count": 15, "nearest": [ { "id": "b794f0d69866", "title": "Meta lost 20 million users last quarter | The Verge", "url": "https://www.theverge.com/tech/921089/meta-earnings-q1-2026-user-decline-ai-investments", "record": "https://links.thingelstad.xyz/links/b794f0d69866/" }, { "id": "bb5cb88ff04d", "title": "FilamentHound — Best Prices on 3D Printing Filament", "url": "https://filamenthound.com/", "record": "https://links.thingelstad.xyz/links/bb5cb88ff04d/" }, { "id": "f9f7d9cdf592", "title": "Six Years Perfecting Maps on watchOS - David Smith, Independent iOS Developer", "url": "https://david-smith.org/blog/2026/04/29/maps-on-watchos/", "record": "https://links.thingelstad.xyz/links/f9f7d9cdf592/" }, { "id": "ed828c2cc661", "title": "Hand‑drawn QR codes — Seth Larson", "url": "https://sethmlarson.dev/hand-drawn-qr-codes", "record": "https://links.thingelstad.xyz/links/ed828c2cc661/" }, { "id": "2645f23ecf37", "title": "Agentic Coding is a Trap | Lars Faye", "url": "https://larsfaye.com/articles/agentic-coding-is-a-trap", "record": "https://links.thingelstad.xyz/links/2645f23ecf37/" } ] }, "mentions": [], "mentioned_by": [], "saved_twice": [] }, "elsewhere": { "wayback": "https://web.archive.org/web/*/https://www.robert-glaser.de/when-everyone-has-ai-and-the-company-still-learns-nothing/" } }