A month ago we reported that Microsoft had merged its consumer and business Copilot apps and cut five features ahead of a September super app. The super app has now arrived. On 25 September Microsoft unveiled a new Copilot app built around three sections. Home puts Chat and Cowork together, with full versions of Word, Excel and PowerPoint running inside Copilot. Code builds apps, trackers and dashboards from a plain-language description. Autopilot is an always-on agent that Microsoft previously called Scout.
Most consultancies, contractors and client-side PMOs in our sector run on Microsoft 365, so this is the Copilot many readers will be handed. Everyday chat, and Copilot inside Word, Excel, PowerPoint, Outlook and Teams, stays on a per-user licence. Cowork, Code, Autopilot and the newest frontier models run on usage-based billing.
Last week we looked at Futurum's survey finding that 46.9% of enterprises are running over budget on AI. The author’s point is that annual budgets do not sit comfortably with highly variable costs. AI agents add another layer, because they can keep working when nobody is watching, and the bill depends on usage.
Two meters on one app
Microsoft says the user subscription licence covers Copilot in Chat and across the Office apps, plus model selection. Cowork, Code, Autopilot and frontier models, which GeekWire names as OpenAI's Astra and Anthropic's Claude Fable, all sit on usage-based billing.
Our view is that there are two kinds of AI at work. They call for two very different ways of paying, and Copilot offers both.
So the licence count no longer tells you what Copilot costs. The seat price is fixed, while agent spend depends on who builds agents, on which models and for how long.
Project controls teams will reach for Code first. Microsoft says it runs in a sandboxed environment and can be hosted within your tenant. A planner could describe a change register or a cashflow dashboard and have Copilot build it, on the meter.
GeekWire reports that usage-based services are off by default and that administrators can set spending limits for the company, specific groups or individuals through Agent 365. Microsoft says admins can route credit requests into approval workflows and set which model families each group can use. We have not seen a published usage rate in the announcement or the coverage we read.
What an Autopilot could do on a project
Microsoft calls Autopilot "your digital teammate". You give it a name, a role and a goal, and it watches channels, follows up on threads, runs recurring work and picks a project back up days later without waiting for a prompt. Microsoft's own example is a supplier review, with the agent building the schedule and workback plan, then running the follow-ups down to chasing stakeholders for updates.
Swap in a subcontract package tender or a monthly cost report cycle and the fit with delivery work is plain. Much of project management is chasing people for information, and this is the part of the launch closest to how our sector works.
Microsoft says each Autopilot lives in your tenant with its own identity, memory, computer and workspace, with permissions, audit and governance behind it. Jacob Andreou, the executive vice president who leads Copilot, was frank about why that matters.
The hallmark of an amazing, long-running agent is autonomy and flexibility. That is the same thing that makes them absolutely terrifying to an IT admin.
On a project, an agent emailing subcontractors for the firm raises contractual questions too, so someone has to decide what it may send and who checks it first.
Picking the model and paying for the choice
GeekWire reports that a menu shown at Microsoft's event let users pick OpenAI's GPT, Anthropic's Opus or an Auto mode that chooses a model for each request.
My test of true independence going forward is for you to have a private eval that you can hill climb on and substitute a model.
Put simply, keep your own test tasks and swap models when another does better. For a QS practice, that could be a few real bill of quantities checks or variation assessments, run before deciding which model family each team may use. On the meter, that choice is also a cost decision.
Most of this is not in your tenant yet
Microsoft says Home and Code will start rolling out in its Frontier early-access programme in the coming weeks, and Autopilot expands to private preview at the end of the month. Code reaches preview for Microsoft 365 Premium and Pro subscribers later this year. Microsoft plans to say more at Ignite, 17-20 November.
GeekWire frames the launch as a push to boost adoption: Microsoft reported more than 30 million paid Microsoft 365 Copilot seats in July, about 7% of the more than 450 million commercial seats it reported in January. The gap before general release lets firms settle governance before anyone asks for an agent.
Takeaway
For everyday Copilot use, the pricing is still based on a fixed seat. The agent features Microsoft is pushing run on usage-based pricing, which only kicks in when an administrator enables them. That gives IT and finance a window to set the ground rules before the costs start.
Ask IT whether your firm is in the Frontier programme and who will decide when usage-based services are switched on.
Set spending limits by group before anyone builds an Autopilot, and route credit requests to a named budget holder for approval.
Decide which model families each team may use, and test them on a few of your own deliverables first.
Pilot an agent on one recurring, chase-heavy process and record what it costs against the time it saves.
Project Flux will track what Copilot's agents cost and do once they reach construction and property teams. To get that before your next licence review, sign up to the newsletter at projectflux.ai.
Links and Stuff
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
