When Your Company Blocks Claude Cowork

Last updated: August 2026

If your company approved Claude Code but not Cowork, you have lost the friendly door and kept almost all of the capability, so learn the terminal and move on. If they gave you API access through a corporate gateway instead of a subscription, you have also silently lost the desktop app, the browser interface, and voice input, and most tutorials you read will not apply to you. If your internal systems are self-hosted, the official connectors will not reach them at all and you need a different approach entirely. Here is how to tell which of those you are, and what to do about each.

Why trust this comparison

This comes from teaching four working professionals through exactly these walls in July 2026, three of whom had an employer constraint that made standard advice wrong for them. I ran the API-only path alongside a student for four sessions, including the part where the homework I set turned out to be impossible on her account. I have not tested every enterprise configuration, and I have no visibility into how any specific company's security team will rule. What I can tell you is which paths worked, which failed, and where the failure showed up, and I will name the point where I got it wrong. Anthropic's own Claude Code documentation is the reference for the tool itself.

If your company approved Claude Code but not Cowork

This is the most common version, and it is better news than it feels like on day one.

Cowork is the friendlier surface. It works on your local files, runs multi-step tasks, and does not ask you to open a terminal. Losing it genuinely hurts if you are not a developer, and anyone telling you the two are basically identical is doing you a disservice. They are not. Claude Code does the same jobs and it is meaningfully more technical to sit in. If you have not seen what you are being asked to give up, the chat vs Cowork vs Claude Code breakdown lays out which surface does what.

But almost everything a non-developer actually wants from Cowork is available in Claude Code. Reading a folder of documents and producing a summary. Generating content in your house voice from a context file. Multi-step work across a project. Connecting to your other tools. That last one matters more than people realize, because tool connections run through the terminal side anyway, so the connection work was always going to happen there.

The student I taught through this actually preferred the terminal once she got past the first hour, because a technical program manager on her team had trained her on it and told her to use it. Her company's approved path and her own preference happened to line up. The person who unblocked her was a manager who had bothered to get technical, which is the practical case for how much code a product or program manager should learn.

Verdict: accept the swap, budget one session for the terminal, and do not treat it as a downgrade.

If your company gave you API access instead of a subscription

This is the one that catches people, because nobody tells you what you lost.

An API key through a corporate gateway gets you Claude Code and third-party tools. It does not get you the desktop app, it does not get you claude.ai in the browser, and it does not get you voice input. Those are subscription products. If you have only ever been handed an API key by your IT department, you do not have an account on the consumer surfaces at all.

I learned this by getting it wrong. I sent a student home with a technique that needed the chat interface: talk freely into chat, have it turn your rambling into a clean prompt, then paste that into Claude Code. It is a genuinely good technique for someone who freezes at an empty prompt box. She spent a week unable to do it, improvising by dictating into a document and copying it across, which was resourceful and not what I asked for. Neither of us realized until the next session.

There is a second consequence: API billing has no guardrails. Subscription products carry usage limits and automatic compaction because the provider does not want you consuming far more than you paid for. Take that away and nothing stops you. Check what your company has capped, because on this path the meter simply runs.

Verdict: workable, but audit every technique against your actual account before you rely on it, and find out who is watching the bill.

If your internal tools are self-hosted

Then the official connectors are not going to reach them, and no amount of prompting fixes that.

Hosted connectors work by calling a vendor's cloud service. If your company runs its own instance behind a firewall, there is no public endpoint to call. This is not a limitation you can prompt your way around, and it is worth establishing early rather than three sessions in.

The path that worked for my student was a community connector run locally through Claude Code, authenticated with a personal access token that she generated herself. It took three sessions to get live, and the token turned out to be the easy part. What made it slow was the folder it lived in, a package update that broke the connection, and a display fault that ate the first fifteen minutes of every call. It does work. It is just not the smooth path.

Before you promise anyone an automatic connection, get one specific answer from IT: will security permit a local connector to reach the internal server, with a personal access token. That single question decides whether the whole approach is even possible. The Model Context Protocol documentation is what to send them if they ask what you are proposing.

Verdict: possible through a local connector and a personal access token, but confirm with IT before you commit to a timeline.

If you cannot install anything without approval

Then be honest about the ceiling before you build a plan on top of it.

One of my students cannot install downloaded skills at work without going through a full committee, and suspects the download sites are blocked outright. Technically the files are local, they are not a deployment to anything, and nothing is being pushed to company infrastructure. Practically, that argument does not survive contact with a security committee, and it is not a fight worth having in your first month.

What still works: everything you write yourself. Context files, your own prompts, your own procedures documented as files in your own project folder. None of that requires an install or an approval. The gap between a downloaded capability and one you wrote yourself is smaller than it looks, and the version you wrote fits your actual work better.

Verdict: build your own rather than waiting on approval, and keep the approval request running in the background.

The numbers side by side

	Subscription	API through a corporate gateway
Claude Code	Yes	Yes
Desktop app and Cowork	Yes	No
claude.ai in the browser	Yes	No
Voice input	Yes	No
Cost model	Flat monthly, published pricing	Per token, uncapped unless your company caps it
Usage guardrails	Built in	None
Best for	Individuals and small teams	Companies that need central control and audit

Accurate as of August 2026. Both sides of this table move quickly, so check before you quote it to anyone.

What I tell my students

Establish your constraints in the first hour, not the fourth. Every one of the engagements above lost real time to a limitation that was discoverable on day one. In one case the constraint had already been sent to me in a message before the session, and I still got surprised by it live, because I had not re-read what the client sent. That is on me, and it is the single most preventable failure in this whole category.

The second thing I tell them is to stop reading generic tutorials. If you are on API-only access behind a corporate gateway with self-hosted internal tools, you are not the reader that most of this content was written for, and following it will send you down paths that end in a wall. Work from what your account actually has. The same discipline applies to whichever chat tool your employer does allow, because the tier you are sitting in decides what happens to anything you paste into it, and that is the ground covered by the rules for keeping a chat assistant safe on company data.

And the honest one: sometimes the answer is that your company has made a decision that costs you capability, and the useful move is to name that plainly rather than to pretend a workaround is equivalent. My student advocated internally for better access, which is a slower fix than a workaround and a much better one. She is still working on it, and she is right to.

One more thing worth planning for, because it caught me by surprise. Partway through the engagement her employer started running its own internal AI training programme, teaching a structured approach to organizing context that was, underneath the different vocabulary, the same idea as the context file we had already built together. My first instinct was to treat it as a competing track. That was the wrong read. The stronger move is to map their framework onto what you are building, so you are operationalizing your company's own methodology on your real work rather than running a parallel system nobody sanctioned.

If your organization is standing up its own AI programme, get the material and align to it deliberately. Approvals get easier when what you are doing is visibly the thing your employer already told everyone to do.

Frequently Asked Questions

Can I just use my personal Claude account for work instead?

Almost certainly not, and it is worth taking seriously. One of my students flagged that she had been using a personal account for work content, which is a data governance problem in most companies and a contractual one if you are under an NDA. Ask before you assume.

Does Claude Code do everything Cowork does?

Not everything, but close enough for most professional work: reading files, generating content, multi-step tasks, and connecting to other tools. What you lose is friendliness, not capability. If you have never opened a terminal, budget an hour for the adjustment and read whether a non-coder can use Claude Code first.

How do I find out what my account actually includes?

Try to sign in to each surface: the desktop app, the browser, and the command line tool. The ones that reject you are the ones you do not have. It takes two minutes and it saves a week.

Is a community connector safe to use at work?

That is your security team's call, not mine, and you should ask them rather than me. What I will say is that running one locally with a token you generated is a materially different proposition from sending company data to a third-party cloud service, and framing it that way tends to make the conversation go better.

Most of this is thirty minutes of conversation once someone has seen the same walls a few times, and it saves the three sessions I spent discovering them the slow way. If you are stuck behind one of these, bring it to a free Discovery Call and we will work out which path your company will actually approve.