Microsoft introduced the “Frontier Firm” in its 2025 Work Trend Index, and the phrase has since travelled well beyond the report. The idea is simple enough. A Frontier Firm is an organisation built around intelligence on tap, where people work alongside AI agents and, eventually, direct whole teams of them. Microsoft described a progression in three phases: AI as an assistant for each employee, then agents joining teams as digital colleagues, then humans setting direction while agents run entire processes and check in when they need a decision. It even floated a new metric for managers, the human-agent ratio.
It’s a neat framework, and I don’t doubt that parts of it will come true. But vendor frameworks tend to describe the vendor’s roadmap more accurately than they describe the customer’s future, and this one is no exception. Strip away the language about digital colleagues and the Frontier Firm reads as a set of instructions for how Microsoft would like businesses to buy AI: in larger, more ambitious commitments, shaped by consultants, and paid for through a mix of per-user licences and metered consumption.
The reason isn’t hard to find. Selling Microsoft 365 Copilot seat by seat has been slower going than the launch buzz suggested. Plenty of organisations bought a pilot batch of licences, handed them to early adopters, then struggled to justify the next order. A per-seat assistant is easy to trial and just as easy to shelve. Reframing the conversation around how the whole business is organised moves it away from the IT budget and towards the board, where the numbers are larger and the decisions stickier. It also moves it into the partner channel, which is where the frontier language has landed next. The logic behind singling out partners Microsoft has recognised for AI transformation work is clear enough: advisory work and process redesign come first, with licences and Azure consumption following close behind.
Handing this to partners makes commercial sense for Microsoft. Agentic projects are messy. They involve mapping processes, cleaning up permissions, wiring agents into line-of-business systems and persuading staff that the new arrangement won’t quietly remove their jobs. That’s slow, bespoke work, and Microsoft has never wanted to do much of it itself. Partners absorb the cost of change management, while Microsoft collects the recurring revenue from Copilot licences, Copilot Studio usage and the Azure AI services underneath. Everyone’s incentives point in the same direction, which is fine until you’re the customer.
That’s where I’d push back on the enthusiasm. A partner rewarded for driving Microsoft adoption has a structural reason to recommend Microsoft-shaped answers. Sometimes those are the right answers. If your identity, documents and email already live in Entra ID and Microsoft 365, building agents on the same platform has real advantages, not least that the permissions model is already in place. But agents built deep inside Copilot Studio, Dataverse and Microsoft Graph are hard to lift out later. Microsoft has opened the door a little on models, adding Anthropic’s Claude as an option in parts of Microsoft 365 Copilot alongside OpenAI’s models, yet the orchestration, the data layer and the governance tooling remain firmly its own. The Frontier Firm, as drawn, is a Microsoft firm.
There’s also a quieter problem with the framework itself. It assumes the main constraint on most businesses is a shortage of intelligence, and that adding agents relieves it. Watch enough of these projects and the real constraint is usually duller. Nobody owns the process from end to end, the data describing it is scattered across three systems and a shared spreadsheet, and the exceptions outnumber the rules. An agent can’t fix that. It will automate the mess faithfully and at speed. The organisations that get somewhere with agents tend to be the ones that did the boring process work first, and that work never needed a frontier label.
None of this means buyers should ignore the idea. The underlying point, that AI pays off when you change how work is organised rather than when you hand out a chatbot, is correct. It’s also a welcome correction to two years of seat-count thinking, when success meant licences assigned rather than work changed. The mistake would be to adopt the framework wholesale, along with the vendor’s preferred sequence of purchases. A sensible business can take the process-redesign lesson, choose a partner on the quality of its delivery rather than its badges, and keep asking which parts of the resulting system it could move if it ever had to.
The bigger unresolved question is commercial. If partners are going to sell outcomes, such as a faster month-end close or fewer service tickets, instead of hours and licences, someone has to carry the risk when an agent-run process gets things wrong. So far, the frontier talk covers what agents can do and says little about who pays when they misfire. The first partner willing to price that risk into a fixed-outcome contract will tell us more about how real the Frontier Firm is than any report Microsoft publishes.
