Microsoft made Copilot Cowork generally available worldwide on 16 June, ending a roughly twelve-week run through its Frontier early access programme.
Charles Lamanna, Microsoft's executive vice president for Copilot, Agents and Platform, called it "the fastest growing feature in the history of our Frontier program," adding that Cowork "has among the highest user satisfaction of any Copilot or agent experience we have shipped."
The headline feature is familiar from the preview: an agentic system that executes long, multi-tool tasks end to end across Outlook, Teams, Word and Excel, producing finished outputs instead of drafts a human still has to assemble. The headline change is the pricing model underneath it, and that is where the real implications sit for any organisation already budgeting for AI-driven productivity tools.
What actually changed at GA
Cowork now layers usage-based billing on top of the existing Microsoft 365 Copilot seat licence. The mechanics:
Every task consumes Copilot Credits, priced at $0.01 each
Cost is calculated from model use, context retrieval, tool calls and runtime
Tasks are classified as light, medium or heavy depending on reasoning depth and the number of knowledge sources involved
Organisations choose between pay-as-you-go pricing or a prepaid P3 plan for those able to forecast usage volume in exchange for a discount
Lamanna described the shift bluntly in comments to AFP: usage-based billing is "the only way to make the model work," since some users now run hundreds of tasks a week, a volume that would drive costs sharply higher under flat-rate billing. He compared the new metering to a familiar everyday transaction: "like you're filling up your gas tank at the pump."
There is also a notable technical change bundled into GA: Cowork is no longer exclusively built on Anthropic's models. Users can now select between Anthropic's Claude Opus 4.8 and Sonnet 4.6, with Frontier programme participants gaining access to OpenAI's GPT-5.5 and a forthcoming Microsoft-built model, internally referred to as Cowork 1.
Lamanna framed this as a deliberate move away from vendor lock-in: "You are not locked into one model with Copilot, you can use the most efficient model or frontier models."
He also claimed Cowork's average cost per prompt now runs "30 to 40% cheaper" than Anthropic's own Claude Cowork agent, thanks to that multi-model routing design. A Microsoft spokesperson separately confirmed the company is evaluating DeepSeek's open source models as a further option, though no firm commitment has been made.
For organisations that participated in the Frontier preview between 30 March and 16 June, there is a grace window. Usage during that period will not be billed until 1 July, but admins have until 30 June to configure usage-based billing in the Microsoft 365 admin centre or lose access entirely once the deadline passes.
Why the pricing shift matters more than the feature list
Per-seat software pricing has been the default model in construction and engineering technology for two decades, and it is predictable for exactly that reason. Usage-based billing breaks that predictability deliberately. Microsoft's own logic, echoed across its Partner Center guidance, is that consumption pricing lets organisations pay in proportion to the agentic work actually being completed, instead of for static seat access regardless of use.
There are two practical consequences worth separating out:
Governance overhead rises. IT and finance teams now need spending limits, usage alerts and reporting dashboards configured before Cowork becomes part of daily workflow, not after. SoftwareOne's guidance to its clients put it directly: this is more than a billing change, it is "a prompt to review who's using Cowork, why, and whether your licensing, spend controls, and governance approach still make sense as AI-driven work scales."
Vendor lock-in loosens. Cowork's earlier exclusivity to Anthropic's models meant buyers were effectively making a single-vendor AI bet inside their Microsoft stack. Opening the model picker to GPT-5.5 and an in-house Microsoft model changes that calculus, particularly once Microsoft's own model becomes the cheaper default option.
Directions on Microsoft's analysis flagged the second point as a strategic move to position Microsoft's own models as the lower-cost option across the wider Copilot family, a pattern worth watching given how aggressively the company has historically used pricing to steer platform behaviour.
What it means for project delivery teams
Cowork's task categories map reasonably well onto repetitive AEC documentation work:
Compiling submittal logs from scattered correspondence
Drafting RFI responses pulled from project documentation
Building first-pass schedules from a meeting transcript
Batch-editing large spreadsheets or comparing document versions across revisions, two use cases Lamanna highlighted directly in the GA announcement
The agentic framing matters because it produces a usable output instead of a half-finished draft, which is the gap that has limited earlier generations of AI copilots in this sector. New integrations with Miro and Monday.com, with Adobe, Box and Canva reportedly coming, extend that reach further into the collaboration tools many AEC teams already use alongside Microsoft 365.
The caution is cost predictability. A firm running Cowork lightly across a handful of admin tasks will likely see negligible credit spend. A firm that scales agentic workflows across project controls, where tasks routinely pull in large document sets and multiple tool calls, could see costs climb in ways that are harder to forecast than a flat per-user fee.
Microsoft's own Cost Management dashboard and Customer Cowork Estimator are explicitly positioned to help with this, which is itself a signal that the company expects cost surprises to be common enough to need dedicated tooling.
The competitive backdrop
Cowork's GA lands in the same fortnight as SpaceX's $60 billion move to acquire Cursor, a reminder that the agentic AI market is consolidating fast and that pricing models are becoming a genuine point of competitive differentiation, well beyond a billing footnote. The companies building the underlying models and the companies building agentic products on top of them are consolidating at very different speeds, and buyers increasingly need to track both layers separately.
Anthropic's own Claude Cowork occupies similar territory to Microsoft's product, and the two are now positioned as direct alternatives for organisations weighing where agentic, multi-tool AI work should live, inside the Microsoft stack they already pay for or as a standalone platform.
For an AEC firm already running Microsoft 365 across its project teams, the practical advantage of Cowork is integration depth, not raw model quality, since it sits natively inside Outlook, Teams, Word and Excel without requiring a separate login or workflow. That integration advantage narrows somewhat as competing agentic tools build their own Microsoft 365 connectors, so it is not a permanent moat, but it is a real one today.
The Microsoft Scout question
Buried in Microsoft's June announcements is a second product worth flagging for anyone tracking this space closely: Microsoft Scout, an always-on personal agent currently available only to a limited set of Frontier customers. Scout is designed to operate continuously across Teams, Outlook, OneDrive and SharePoint, working with everyday data such as chats, emails, calendar entries and contacts, and it can be expanded to act on browser activity and local resources through the desktop app.
It is a different product from Cowork, oriented towards individual productivity instead of discrete agentic tasks, but its existence signals that Microsoft sees agentic AI as a layered product strategy, not a single feature.
Firms evaluating Cowork now should expect Scout, or something like it, to arrive in their own roadmap discussions before long, and the governance questions raised by usage-based billing today will resurface in a more personal, harder-to-audit form when that happens.
Takeaway
Model your likely Cowork credit spend before the 30 June Frontier deadline, not after. Treat this the way you would any new consumption-based vendor contract, with a forecast and a spending ceiling agreed in advance.
Assign clear ownership of Copilot Credit governance internally. This is now a recurring operational cost line, not a one-off licence renewal, and it needs the same scrutiny as cloud compute spend.
Watch for Microsoft's in-house Cowork 1 model to undercut Anthropic and OpenAI options on price once it ships. If cost becomes the deciding factor for task routing, output quality on complex AEC documentation tasks deserves a deliberate quality check, not an assumption that cheaper means equivalent.
Map which of your current manual documentation workflows, RFIs, submittal tracking, meeting-to-schedule conversion are good early candidates for Cowork given its strength in long, multi-tool tasks over single-step queries.
Ready to build smarter?
Project Flux covers the platform shifts reshaping how AEC teams buy and deploy AI tools, going well beyond the feature announcements. Subscribe for the weekly newsletter built for people who have to justify the spend.
Links and Stuff
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
/
1