It finally happened. On June 16, Microsoft made Copilot Cowork generally available, and I will say this up front: I loved it during the preview. It is good. More than half the Fortune 500 was running it during the Frontier preview, and you can feel why. So this is not a hit piece. It is a heads-up, followed by the part nobody sends you in the launch email — how to actually keep the cost under control.
Here is the headline that surprised me. Cowork is not just bundled into your Copilot license anymore. It runs on pay-as-you-go, metered pricing on top of the license you already pay for. And once I saw the numbers for how I personally use it, I had to sit down for a minute.
What the new pricing actually is
The model is consumption-based. You pay $0.01 per “Copilot Credit,” and every task you run burns a variable number of credits depending on four things: how much the model works, how much it has to retrieve for context, how many tool calls it makes, and how long it runs. Microsoft buckets tasks into three rough sizes. A light task runs about 100 to 300 credits, so a dollar to three dollars. A medium task is 400 to 700 credits, call it four to seven dollars. A heavy task is 700-plus credits and climbs from there.
That sounds tiny until you remember the whole point of Cowork is to run a lot of tasks. And this sits on top of the Microsoft 365 Copilot license, which is $30 per user per month for enterprise. There is also a prepaid option where you commit to a volume up front for a discount, plus a grace period for Frontier preview users through July 1. But the core idea is simple: the better pieces just moved to a meter.
I ran my own usage. It was $500 to $900 a month.
I do not guess at this stuff, so I took Microsoft’s calculator, and then I took a screenshot of my actual Cowork activity and ran it through AI to estimate what my real-world usage would translate into. The answer came back somewhere between $500 and $900 a month in extra fees. For one person. On top of the license.
I am a heavy user, so that is the high end — most people on your team will not touch it that much. But do the math on an organization. If you have a thousand people and even a fraction of them lean on Cowork the way I do, you are not signing up to spend a thousand dollars a month. The value that made Microsoft the best AI bundle out there — all that integrated Copilot inside Microsoft 365, plus Cowork — is exactly what is now getting carved out and metered.
The better pieces just moved to a meter. The trick is no longer “do I use it” — it’s “who uses it, for what, and how well.”
Now the optimistic part: don’t panic, optimize
Here is what I keep reminding myself. A meter is not a wall. It is information — or at least it will be, once the tooling catches up (more on that gap in a second). The people who get burned will be the ones who run Cowork on autopilot for everybody. The people who do fine will be the ones who run it on purpose, for the right people, on the right work. So here is how I am approaching it.
Be incredibly strategic about who gets it. This is the first rule, ahead of everything else. Cowork is not a tool you hand to the whole org and walk away. The people using it need to be chosen deliberately and specially trained — it behaves differently than regular Copilot chat, and an untrained user burning heavy tasks is exactly how a bill gets away from you. I am going to training on it next week; Microsoft has some open sessions, and my team is poking around right now to learn what they can. Build that learning curve into your plan, not as an afterthought.
Estimate before you commit. Run the calculator, but do not trust the demo numbers — feed it your real activity the way I did. A screenshot of a typical week through an AI estimate will get you far closer to the truth than the marketing math. Know your number before the bill knows it for you.
Separate the thinking from the doing. This is the big one. You do not need a metered agent to answer general questions. Use regular Copilot chat for the back-and-forth — the brainstorming, the “what’s the next step,” the cleanup. Do all of that prep cheaply, get everything lined up, and then hand Cowork one clean, well-defined run instead of ten sloppy ones. Most of the waste I have seen comes from making the expensive tool do the cheap work.
Scope it to projects, not people. You probably do not need Cowork humming for every user all day. Turn it loose on specific, high-value projects where an automated assistant genuinely earns its keep, and keep the light users on the flat parts of the bundle. Match the tool to the job.
Watch the heavy tasks. The 700-plus-credit runs are where the money goes. A handful of those a day is what turns a reasonable bill into a four-figure one. Be deliberate about which jobs deserve the heavy treatment and which can run lighter.
The honest gap: governance and reporting aren’t all the way there yet
Here is where I want to be straight with you, because I do not want to leave the impression that managing this is easy today. It is not. The governance toolset is not all the way there yet.
Compare it to the other side of my desk. With Claude Cowork, I can go into the settings and actually see how much of my allotment I have used — within a five-hour session and across the week. It is right there. With Microsoft Copilot Cowork, I do not see a clean, user-facing tool like that. There is an admin panel that supposedly breaks the numbers down, but figuring out what an individual person or project actually cost is genuinely difficult right now, not the easy click-and-see you would want before you commit real budget to it.
I am confident that changes. The moment finance teams start asking “who spent what,” Microsoft is going to get a lot of requests for exactly this, and they move fast when the whole market is asking. I fully expect the governing, measuring, reporting, and planning tools to catch up quickly. But I am writing this in June 2026, and today the honest answer is that the meter is running ahead of the dashboard. Plan around that gap until it closes.
And know your alternatives, because they’re real
This is where I land, and it is meant as encouragement, not a complaint: you have genuinely good options at every price point now, and that is a great position to be in. Claude Cowork has a flat $100-a-month package, and I have not even come close to using half of it. Before that I was on the $20 Pro plan and got a heck of a lot of work out of it — sometimes I had to wait five hours for my tokens to reset, but once you get set up and learn to optimize, it works great. Twenty dollars. Google’s Gemini-powered tier runs around $100 a month, which felt steep when I first saw it and now looks like a bargain by comparison.
For a small business, the math basically makes itself. A flat hundred dollars a month with a capable assistant beats an open-ended meter you cannot predict. For a large enterprise with mostly light users and a few power users on scoped projects, the Microsoft bundle plus careful Cowork metering can still pencil out — especially with the prepaid discount. There is no single right answer, and that is the good news. The trade-offs are finally visible.
My verdict, right now
So where do I land today? At work, I am not going to push for a large rollout of Copilot Cowork — not yet. What I will do is test it deliberately on a few scoped projects to see if we can still get real bang for the buck out of it, with the right people trained and watching the meter. If the value is there, great. If the governance tooling matures the way I expect, even better. We’ll see. That is an honest “prove it to me,” not a no.
At home, I am sticking with Claude Cowork and Claude Code. I switched to the Max plan, which is $100 a month — expensive for personal use, I will admit — but I have not hit any limits with it. And here is the funny part: I may actually be able to downgrade in the coming months, because most of the things I built it for are now built. Once the automation is standing on its own, I do not need the big account to keep it running. That is its own kind of win.
The bottom line
Yes, I was disappointed. Watching the best pieces of the best bundle move to pay-as-you-go stings, and I think some heavy users are going to be unhappy when the first bill lands. But the tool is good, the meter is manageable, and the alternatives are strong. Be strategic about who uses it, train them, do your thinking cheap and your doing clean, scope it to the work that matters, and plan around the governance gap until Microsoft closes it. Panic is expensive. A plan is not.
Written by Brad Rowland — IT Infrastructure and Operations leader, automation builder, and AI implementer.
