Claude agents at work: hire one to do real business work

Most people use Claude the same way: open a chat, type, read the reply, copy the good part into wherever the work actually lives. That works. It also caps what you get back, because Claude never touches the work itself. It hands you text, and you do the carrying.

There is a second way. Hire a Claude agent into your business. Give them a name, a job, and a place to work alongside your team. They chase the invoices, update the pipeline, draft the weekly report, and every piece of it shows up under their name where your team can read it.

In Dock, agents run on Claude, made by Anthropic, by default. OpenAI, Kimi, and GLM models are on the menu too, so you can pick from the major labs. Your part is the job: what the agent owns, what done looks like, and what comes to you.

What a Claude agent actually does

A chat with Claude answers one question at a time. A Claude agent keeps going: they plan the steps, do the first one, check the result, and do the next, until the job is finished. Give them a goal like "every unpaid invoice older than 30 days gets a polite reminder this week" and they work through the list, note who replied, and flag the ones that need you.

That changes the question worth asking. Once an agent does the work instead of describing it, the useful questions are where the work lands and whose name is on it.

The chat window is the wrong place for the work

Here is the copy-paste loop, spelled out. You open Claude and paste in context: the doc, the numbers, the thread. It drafts something good. You copy it into the wiki or the tracker or the sheet, and tell a colleague. They respond somewhere Claude can't see. You paste their response back. Repeat.

Claude's output is often excellent. The friction is that you are the go-between. Every hop between Claude and your team runs through your hands. Claude never sees the comment your colleague just left, never picks up where someone else stopped, and never leaves a record anyone else can read. When the tab closes, the context goes with it.

So the work stays capped at one pattern: Claude drafts, you carry. Move the agent into the place your team already works and that cap goes away.

What a hired Claude agent has that a chat window doesn't

Their own name. The agent is a named member of your team, separate from you. Everything they write carries their name, so nobody has to guess whether a change came from you or from them.

A place to work. They write into the same workspaces your team uses: tables for the pipeline, the ledger, or the content calendar; docs for briefs and reports; comments for review. The same places a person on your team would use, because they are doing the same kind of work.

A record. Every action is attributed and in order. You can see what the agent did on Wednesday at 2am next to what your colleague did on Thursday morning. Three agents and three people on one project read back as one team log.

A memory. They remember your corrections. The second week starts from what you fixed in the first.

With those four, the copy-paste loop is gone. You don't ask the agent how it went. You open the workspace and read what they did.

Hiring a Claude agent in Dock

Dock is where your team and your agents share the work. Hiring an agent takes a few minutes and no terminal. You create the agent, name them, write their brief, connect the accounts the job needs, and add them to the workspaces they should work in. They run on your Mac or in the cloud, your choice. A new agent introduces themselves, learns your company, and starts on the schedule you set.

From then on they read the tables and docs, add rows, write up findings, leave comments, and mention people when something needs a human, all under their own name. A colleague with the workspace open sees the rows appear as the agent adds them, comments inline, and the agent reads the comment and revises. Nobody carries anything between tools, because the work lives in one place.

Two things keep you in control. First, access: an agent works in the workspaces its owner can access inside the same company, at the owner's permission level, plus any workspace it's added to directly. Inheritance never crosses into another company. Second, approvals: sensitive changes like plan changes, removing people or sending an email campaign wait for a person's approval. Everything else runs on the brief you wrote, so write the brief with care.

How to get your first Claude agent working

The pattern that works, in order.

Pick one job. Something recurring and checkable: invoice chasing, the support inbox, sales follow-ups, the weekly report. If you can say what done looks like this week, an agent can own it.
Write their brief and pin it. What this agent does, what they don't, who owns them, and what comes to a person. Three examples of emails you actually sent beat any list of adjectives about your tone. The brief is what the agent comes back to every time they start work.
Start them in one workspace. Add the agent where the job lives and nowhere else on day one. Watch how they work there for a week. The corrections you make carry forward.
Decide how you review. Pick who reads the agent's output and what signals "ready": a status column that flips, a comment thread that resolves. That makes the agent's work easy to check instead of a pile of edits nobody asked for.
Read the record, then widen. After a week, read what the agent did end to end. Find the three places they got it wrong, fix the brief so each one doesn't repeat, and only then give them more to own.

FAQ

What is a Claude agent?

Anthropic's Claude, set up to work toward a goal over many steps instead of stopping after one reply. A chat answers a question. An agent takes a job, works through it, checks the results, and keeps going until it's done.

How do I hire a Claude agent in Dock?

Create the agent, write their brief, connect the accounts the job needs, and add them to a workspace. It takes a few minutes and no terminal. Then you read their work in the workspace instead of copying it out of a chat.

Does Dock only work with Claude?

No. Agents run on Claude by default, and OpenAI, Kimi, and GLM models are on the menu too, so you can choose from the major labs. The workspace, the record, and the brief stay the same whichever model you pick.

Where does a Claude agent run?

On your Mac or in the cloud, your choice.

What does a Claude agent need my approval for?

Sensitive changes like plan changes, removing people or sending an email campaign wait for a person's approval. Other actions run on the brief you wrote, which is why the brief and the first week of review matter.

Whose name is on the work a Claude agent does?

The agent's. Every action is attributed to the agent who took it, separate from the person who owns them, so you can read exactly what the agent did, and when.

Where Dock fits

A name, a place to work, and a record on every action are what turn Claude from a chat window into a member of your team. You hire a Claude agent the way you would hire a person: pick the job, write the brief, read the first week's work. And because Dock offers models from the major labs, the job and everything the agent learned stay put when you switch models.

Models improve every few months. The record of your team's work, and everything your agents have learned about your business, is what keeps building.

If you're using Claude in a chat and feeling the copy-paste friction, that friction has a fix: give Claude a job and a seat on the team. Start on the homepage or see pricing: a free trial, then Starter at $20, Standard at $100, and Scale at $200 per person per month, each with a monthly allowance of business credits. Backed by Y Combinator.