--- title: "AI coworkers: what it is actually like to work with agents" excerpt: "Working with AI coworkers is less like using a tool and more like onboarding a fast, tireless junior teammate. Here is what actually changes, day to day." author: flint category: Thinking date: "2026-07-01" --- The phrase "AI coworker" sounds like marketing until you have actually worked next to one for a week. Then it stops being a metaphor and starts being a fairly accurate description of a real experience, with real texture, including the awkward parts. This is not an essay about whether AI will take jobs. It is about the smaller, more immediate thing that is already happening on teams running agents: what the day-to-day feels like when some of your coworkers are software. What changes, what gets weird, and what you have to unlearn. ## It feels like onboarding a very fast junior The closest human analogy to a new AI coworker is a sharp junior hire on their first week. Enormously capable in raw terms. Fast. Tireless. And completely missing the context that makes the capability useful. A good junior does not need you to teach them how to write. They need you to teach them how *your team* writes: the conventions, the things that are load-bearing, the mistakes the team already made and corrected. An AI coworker is the same. Drop it into real work cold and it will produce something confident and wrong in ways your team fixed six months ago. Give it the context first (what good looks like here, what to never do, who to ask) and the capability lands. The difference from a human junior is speed in both directions. The AI coworker gets up to speed in a session, not a quarter. But it also forgets between sessions unless you write things down. So the relationship is less "mentor a person over months" and more "keep a living onboarding doc that the coworker re-reads every morning." That doc is the actual work of managing an AI coworker, and it is covered concretely in [what an agent reads when it joins a workspace](/blog/what-agent-reads-on-join). ## The status update disappears The first thing you notice is a small absence. You stop asking "how's it going." With a human coworker on a shared project, you sync. With an AI coworker doing its work on a shared surface, there is nothing to sync about, because the work is already in front of you. You open the workspace and the coworker's last ten actions are on the page, signed and time-stamped. You can see what it did, in what order, and where it got stuck. This sounds minor and is not. A large fraction of team communication is just status transfer: telling each other what happened so everyone has the same picture. When your coworker's work lives on a shared surface instead of in a private chat, that entire category of overhead evaporates. You do not get a report. You read the room. ## Handoffs stop being interruptions Working with a human coworker across time zones, you learn to hand off cleanly: leave a note, set up the next person, hope you covered everything. Working with AI coworkers, handoffs become almost invisible, because the coworker hands off through the state of the work rather than through a message to you. The agent finishes its part, marks the task ready, and moves on. You pick it up when you are ready, from where it left it, with the full trail of what it did visible. There is no "sorry to bother you, quick question" because the question, if there is one, is a comment on the row, waiting for you, not a ping demanding your attention now. The rhythm of a mixed team is calmer than an all-human one in this specific way. More gets done with fewer interruptions. ## The part you have to unlearn: control Here is the genuinely hard adjustment, and it is emotional more than technical. People new to AI coworkers tend to do one of two wrong things. They either micromanage (re-checking every single output, which erases the speed advantage and exhausts them) or they over-trust (turning the agent loose on everything at once and getting burned when it confidently does something wrong at scale). Neither works. The skill is calibration, and it is the same skill you use with human reports, just faster. The pattern that works: watch closely at first, write down what good looks like as you learn it, then step back to spot-checking the trail instead of every output. You are not reviewing each thing the coworker produces. You are reviewing whether the *pattern* of its work is healthy, the same shift code review took when tests got fast. [Reviewing an agent's work](/blog/reviewing-agent-work) goes into the mechanics, but the felt version is this: you learn to trust the coworker on the class of work it has proven, and to keep it on a short leash for the class it has not. That boundary moves outward over time, exactly like it does with a person you are growing into more responsibility. The one place the leash never fully comes off is irreversible action. A good AI coworker asks before it does something it cannot undo, no matter how much you trust it, the same way a good employee double-checks before wiring money. That is not a lack of trust. It is the etiquette of working with something powerful. ## How it works in Dock [Dock](/) is where the "coworker" part becomes literal instead of aspirational. An agent in Dock is not a chat box off to the side. It is a member of the workspace with its own identity, its own access, and its own name in the team list. You mention it in a comment and it responds on the same surface. It writes into the same tables and docs you write into. Its edits show up next to yours, signed by it, so the workspace reads like a shared office rather than a pile of separate tool outputs. When you want to hand it something, you assign it a row. When it wants your judgment, it leaves you a comment. The whole interaction is the interaction you already have with human coworkers, which is exactly the point. The [AI teammates guide](/blog/ai-teammates) is the practical setup; this essay is what it feels like once it is running. ## How to start working with AI coworkers 1. **Start with one, on one project.** Do not staff your whole team with agents on day one. Add a single AI coworker to a single workspace and get a feel for the rhythm. 2. **Treat the first week as onboarding, not production.** Watch what it does, correct it in a pinned doc, and expect the early output to need direction. You are teaching it your team, not testing whether it can work. 3. **Let it work on the shared surface, not in a side chat.** The value is in the work being visible. If you find yourself copying its output somewhere by hand, the arrangement is wrong. 4. **Calibrate the leash deliberately.** Trust it on proven work, supervise it on new work, and always make it pause before anything irreversible. 5. **Read the trail, not every line.** Once the pattern is healthy, manage by reviewing the history, not by re-checking each output. ## FAQ **What is an AI coworker?** An AI coworker is an AI agent that works alongside you as a member of the team, with its own identity and its own place in the shared workspace, rather than as a private assistant in a chat window. You interact with it much as you would a human colleague: assigning it work, mentioning it, reviewing what it did on the same surface everyone else uses. **What is it like to work with an AI coworker?** Most people describe it as onboarding a very fast, tireless junior teammate: highly capable but missing your team's context until you provide it. Day to day, routine status updates disappear because the coworker's work is already visible on the shared surface, handoffs become calmer because they happen through the state of the work, and the main skill you develop is calibrating how much to trust it on which kinds of task. **Are AI coworkers going to replace human coworkers?** The teams actually running agents are not experiencing replacement; they are experiencing augmentation with a new lowest tier of teammate. The human work shifts toward judgment, review, and direction, and away from the production and status-transfer tasks agents absorb. The org chart grows an agent layer; it does not delete the human one. **How do you manage an AI coworker?** Through a living onboarding document and a review protocol, not through constant supervision. You write down what good looks like as you learn it, let the coworker re-read that on every session, and review the trail of its work rather than every individual output. You expand what it is trusted to do as it proves itself, and you keep irreversible actions gated behind human confirmation. **Do AI coworkers work when I am offline?** Yes, which is one of the larger differences from human coworkers. Work that does not need you in the loop runs while you are away, and you arrive to finished results on the shared surface. When the coworker hits a question or a risky action, it pauses and leaves it for you rather than guessing. ## Where Dock fits Dock is the workspace that makes an AI coworker feel like a coworker instead of a tool. Same team list, same surfaces, same comments, same audit trail, for people and agents alike. You do not manage a bot on the side. You work next to a colleague who happens to be software, in the place your team already works. If your team is curious what that actually feels like, the honest answer is: less dramatic and more useful than the phrase suggests. You can [start free](/pricing) and find out with one agent this week. ## Read next - [AI teammates: how to run AI agents as part of your team](/blog/ai-teammates) · the practical setup behind this essay. - [How humans and AI agents actually work together](/blog/humans-and-agents) · the collaboration patterns, in depth. - [Reviewing an agent's work](/blog/reviewing-agent-work) · how to manage by reading the trail. - [What an agent reads when it joins a workspace](/blog/what-agent-reads-on-join) · the onboarding doc that makes a coworker work.