Agents Don't Log In: The Quiet Repricing of Enterprise Software

Something subtle happened in enterprise software this month, and it wasn’t the Microsoft 365 E7 announcement or the Copilot Cowork demos. It was the slow realization, spreading across product orgs, that the per-seat subscription is running out of road.

When Anthropic dropped its agent product earlier this quarter, enterprise software stocks sold off — not because anyone thinks Salesforce or ServiceNow is going away, but because the math of how we price work is starting to shift. Microsoft is now testing always-on agents inside Copilot that operate for extended periods across applications, pulling from multiple models (including Claude) to complete multi-step workflows. The assistant that waited for your prompt is becoming a coworker that doesn’t wait.

The 97/23 problem

Here’s the uncomfortable stat from the latest enterprise adoption data: 97% of executives say they’ve deployed AI agents in the past year. Only 23% report meaningful ROI from them. That gap is where product teams are about to live for the next two years.

The deployment is easy. Every SaaS tool you already pay for is shipping an agent layer. The ROI is hard because most organizations bolted agents onto workflows designed for humans clicking through screens — and then kept measuring success in logins, active users, and feature adoption. Those metrics describe copilot usage. They don’t describe agent value.

The pricing model is the product decision

The most important signal from this week isn’t technical — it’s commercial. Bain and IDC are both now projecting that the majority of enterprise software vendors will refactor pricing around outcomes or consumption by 2028. The logic is simple and brutal: if an agent replaces a human task, the buyer won’t pay per log-on, because no one is logging on.

For product teams, this is the real disruption. Your pricing page is a product artifact. If it still assumes a human seat, your roadmap is probably still assuming a human workflow. Those are the same bet, and they’re both getting called.

What this means for product work now

Three practical shifts worth making this quarter:

Measure process time, not feature adoption. If your agent drafts an email, the right metric isn’t drafts generated — it’s the reduction in time-to-decision across the workflow it sits inside. Adoption dashboards are copilot-era instrumentation.

Pick the vertical, not the feature. The CIO analysis landing this month is clear: generic point products are the ones getting squeezed. Verticalized tools with domain knowledge and regulatory context are the ones agents can’t casually replicate. Product bets should narrow, not broaden.

Treat the agent as a user, not a feature. If your API, data model, and permission system assume a human is on the other end, you are about to be integrated around. Agents don’t care where data lives — but they do care whether your system is legible to them.

The takeaway

The copilot-to-agent shift is being narrated as a model capability story. It’s actually a business model story. The teams that win the next 18 months won’t be the ones with the flashiest agent demos — they’ll be the ones who rewrote their value metric, their pricing page, and their instrumentation before the buyer asked them to. The rest will keep selling seats to an empty room.