--- title: Why I Think Small Teams Will Get More Out of AI Agents description: The things that slowed down machine learning inside big companies are slowing down AI agents now, and small teams mostly don't have those problems. Some thoughts on where agents can help a small business first. pubDate: 2026-09-29 image: /images/articles/small-team-ai-infographic.png --- I spent a lot of years doing machine learning inside big companies (pricing, data science, some AI strategy work), and lately I keep noticing that the things that slowed us down back then are slowing down AI agents now. That got me thinking that small teams might have an edge here, at least for a while. On paper, the big companies had what you'd want to make this work. They had money and plenty of technical people. Even so, getting a useful model into production took a long time, and a lot of projects stalled before anyone wrote any code. ## The technical part was the easy part This surprised me early on. Building the model was usually the easiest piece of the project. An agent today could help with a lot of that work, writing code, training a first model, drafting the documentation. What slowed things down was everything around it. There were a lot of stakeholders, each in a different part of the business with their own expectations, so picking which problem to work on was political before it was technical. The infrastructure existed, but every platform belonged to some team. One team wanted you on theirs, another didn't want theirs to change, and buying anything new meant a procurement process that could take months. Getting data in a regulated business meant working out who's allowed to see what and who has to approve the request. And finding someone who knew both AI and the specifics of the business was hard. (There were also people who pushed back on anything new because it made their job harder, which is understandable.) Buying an agent doesn't change any of that. You can hand it the technical problem and it will probably do fine, but getting permission to hand it over is the slow part. ## What I hear from people at larger companies A few friends at larger companies tell me similar things. They have access to Claude or Copilot, but credits are tight, so the tool gets partway through a task and stops. Sometimes running code is turned off, so the agent can't execute anything. In some coding setups it's closer to autocomplete in an editor than an agent with access to the environment. So people end up with output they don't fully trust. My guess is that in a lot of these places the pressure is to show progress on AI more than to make money from it, since the company is doing fine either way. (The big tech companies seem to be the exception... they're much less locked down.) ## Why small teams have it easier A small team looks pretty different. Expertise sits with a few people, and there's a lot less process to get through. The problems are at a size where a team or pro plan goes a long way, so you don't need an enterprise contract. The owner can say yes in the same conversation where the idea came up, and often they're the person you were already talking to. And a small business usually needs the help, since it's competing with companies that have more people. I think two things agents are good at matter most here. They can take on a lot of documentation work, and they can build tools around the workflows you already have. Both come up constantly in a small business. ## A hypothetical example: quoting in a small shop Say you run a small machine shop that makes custom and short-run parts. Quote requests come in as drawings and specs, by email or through a portal, and someone has to estimate material, machine time, and labor, then send a price back. The estimating is probably slow and manual, built up over years in spreadsheets and in one person's head. And the tools the shop runs on probably never quite fit how you work, with no easy way to change them. Both of those are places an agent could help. For the documentation side, you could set up something that reads the incoming spec and drafts the quote and job paperwork from your own pricing rules, and your estimator reviews it before anything goes out. That should save time on the busywork while the estimator still makes the call. For the workflow side, a pricing tool built around how your shop already estimates is something you could probably afford now, whether someone in-house sets it up or you bring in outside help, since the people who build these tools are using agents too. I'd expect faster quotes to help, since customers often go with whoever gets back to them first. ## Where to start If you can make the decision yourself, I'd start by looking at your own business for those two things. Where are people spending hours each week writing documents, and which workflows would run better with a tool built around them, if anyone could justify the cost? My guess is that's where an agent would help first. What's the thing in your business that eats the most hours in documents or spreadsheets? I'd be curious to hear. Thanks for your time. Julio