The interns that do real work.

Most AI work on the market right now is framed around agents. You ask for a task, an agent completes the task, and the interaction ends. Agents are measured on benchmarks that look a lot like multiple choice questions with tool use bolted on. That framing is fine for what it is, and we have shipped plenty of it. It is also the reason so many of these systems feel impressive for a minute and useless by the end of the week.

An intern is not an agent. An intern is someone you hand a role to. Sales. Growth. Dev. GTM. The role comes with a rhythm, a set of tools, a set of people it reports to, and a memory of what happened yesterday. Measured on a single task, an intern might look slower and dumber than a finely tuned agent. Measured over a week, the intern outperforms, because it is carrying context the agent has to rediscover every morning.

Training for a role changes almost everything about what you do. Evaluations stop being about a single task and start being about a week of work that ends in an outcome. Memory becomes the main product, not a feature. The model stops being the most interesting system on the screen, and the coordinator between the models starts being the thing worth writing about. That is the part of the research we spend the most of our time on, and the part that takes the longest to get right.

Role based training also changes what an intern is allowed to do. The interns we train have small hands on purpose. They do not touch production code without review. They do not email a customer without a human eye on it. They do not move money. They do a lot of small things well, and they know which decisions belong to a human instead of to them. The point is not autonomy for its own sake. The point is reliability over a long time horizon, which is the only reason anyone would trust a system to own a role at all.

A lot of what looks like intelligence in a real week of work is actually continuity. Remembering what we tried last Monday. Remembering who said what on the Thursday call. Remembering which kinds of leads convert and which ones never do. A big model does not give you this by default. A good intern does, because we trained for it specifically, and because the coordinator between the interns holds the shared spine that no single intern would have on its own.

The handoff is the other part people underestimate. Work in a company does not stay inside one role. The campaign started by the Growth intern ends up needing a list scrub from the Sales intern. The bug report filed by the Dev intern ends up needing a release note from the GTM intern. Agents are poor at this because they are trained to finish, not to pass. Interns are specifically trained to pass, and the coordinator is specifically trained to route. When we talk about an intern doing real work, the handoffs are the part that makes the phrase true.

We are not the first people to notice any of this. The useful contribution of the lab is practical, not theoretical. We train narrow, reliable, role based systems. We build the coordinator that binds them. We ship them into real companies so we find out where the cracks are, before the cracks become products. Calling them interns is a little bit of marketing, but mostly it is a frame that keeps us honest. Interns are measured by the work they do over a quarter. That is the bar.

The next few posts here will be more technical. How the coordinator holds context. How memory is structured so two interns can agree on what happened. How we decide when a role is ready to leave the lab and go live inside Enter, or whichever product the next intern ships into. For now, this is the thesis in one place. If you want to see what it looks like in practice, Enter is the version of this thinking that you can sign into today.