Microsoft Says AI Agents Need Their Own Software Seats. The Product That Would Make That Real Has Not Shipped.

Microsoft says AI agents will need their own software seats, just like employees. The product it shipped to make that real tells a different story.

In April, Microsoft executive Rajesh Jha told a conference that autonomous AI agents could become paid software users inside enterprise companies — each with its own login, inbox, and identity. The pitch was not just a product announcement. It was a revenue thesis: as AI automates human work, software vendors do not have to lose seat-based pricing. They can simply charge the agents instead. "All of those embodied agents are seat opportunities," Jha said.

Six weeks later, Microsoft shipped Agent 365. The product is real. The thesis it would prove is not — not yet, anyway.

Agent 365 launched May 1 at $15 per user per month as a standalone governance layer, or bundled at $99 per user per month inside the M365 E7 Frontier Suite. It registers agents, assigns them Microsoft Entra identities, and extends Purview and Defender to cover their activity. It is a genuine governance infrastructure for non-human actors, and it matters.

But the specific capability Jha cited as the basis for new revenue — agents with their own independent identities, operating autonomously rather than on behalf of a licensed human user — remains in Microsoft's Frontier preview program. It is not part of the May 1 general availability release. At launch, agents covered by Agent 365 operate on behalf of, and under the license of, the human users they assist. The human license covers the agent. The agent does not need its own seat.

The distinction matters because it changes who pays. Under the current GA model, if a company with 100 Microsoft 365 users deploys 500 agents acting on their behalf, it pays for 100 seats. Jha's vision implies a different math: 500 agents, each a billable user. That future is architecturally planned — Microsoft's own documentation describes "agentic users" with their own mailboxes and principal names — but it is not yet enforced by the product that shipped.

This is not a minor timing note. It is the gap between the revenue thesis Microsoft is selling to investors and the product it has delivered to customers.

Salesforce is running the same experiment in parallel, with more visible pivots. The company has shipped three distinct pricing models for its Agentforce product in roughly 18 months: $2 per conversation in October 2024, Flex Credits at $0.10 per action in May 2025, and per-user licenses starting at $125 per month in late 2025. All three run simultaneously. The per-user seat model — which Jason Lemkin of SaaStr describes as Salesforce's answer to the "digital workforce" framing Benioff has adopted — is an explicit bet that agents can be wrappers for paid human-equivalent seats. Agentforce reached $540 million in ARR by Q3 fiscal 2026, growing 330% year-over-year. Eighteen months and three pricing pivots later, it has penetrated roughly 8% of Salesforce's 150,000-customer base.

The seat-based pricing model is not holding steady across the industry. Seat-based pricing as the primary wrapper dropped from 21% to 15% of B2B companies in a single year, per Growth Unhinged's 2025 State of B2B Monetization report. Hybrid models — mixing seat, consumption, and outcome-based components — surged from 27% to 41% over the same period. The market is not converging on agent-seats. It is experimenting.

Not all analysts buy the seat-expansion thesis. Nenad Milicevic, a partner at AlixPartners, argues the opposite: AI agents reduce the number of humans who need to interact with software, compressing license counts rather than expanding them. One person overseeing a dozen agents does not need a dozen seats — they need one. "The winners will be open platforms," Milicevic told Business Insider. Vendors that charge premium rates for machine access risk losing customers to competitors that let agents operate freely.

Microsoft's own documentation acknowledges the sequencing issue in places. Agent 365 at GA is a governance layer attached to existing user licenses. The longer-term vision — where agents are themselves the licensed unit — lives in a separate documentation track for the Frontier preview program, not in the product that enterprises are being asked to buy today.

The $99 per user per month E7 bundle is real money. It is being sold to procurement teams as the agent-ready subscription. But the feature that would make agents independently billable — the one Jha cited as the basis for expanding seat counts rather than shrinking them — is not in the box.

What is in the box matters. Agent identity, registry, compliance coverage, and audit trails for non-human actors are genuine enterprise needs as agent deployments scale. IT departments managing shadow AI agents — tools employees deploy without approval — have real governance problems that Agent 365 addresses regardless of whether the seat-expansion thesis proves out.

The honest version of this story is therefore two stories at once. Microsoft is building the infrastructure that would make agents billable seats. That infrastructure is real and needed. The claim that the revenue from that infrastructure is already flowing — that the product shipped May 1 extends Microsoft's TAM by converting agents into users — is a thesis layered on top of a governance tool that has not yet demonstrated its commercial model.

The seat fight is coming. Agent 365 is the opening move, not the settlement.