1. Introduction
In this episode of our AI podcast, we (Gerjon and James) sit down with licensing and FinOps expert Rich Gibbons (https://www.linkedin.com/in/rich-gibbons-microsoft-licensing/) to unpack one of the biggest shifts we’ve seen so far in enterprise AI: Microsoft’s new Copilot CoWork consumption-based licensing model and the arrival of Microsoft Scout as an “autopilot” for AI.
For the last decade, Microsoft’s SaaS story has been simple: pay per user, per month and you’re done; with CoWork and Scout, that world is changing fast, and the financial implications for organizations are huge.
2. Meet the Guest
Rich Gibbons is the founder of “Cloudy with a Chance of Licensing,” a consultancy and content platform focused on making Microsoft licensing, IT asset management (ITAM), and FinOps understandable for normal humans. He’s been working in Microsoft licensing for many years, delivering training and advisory services to organizations on topics ranging from Microsoft 365 and Azure to cloud cost management and AI-driven spend.
Alongside his blog and newsletter “Cloudy with a Chance of Licensing,” Rich also runs a podcast under the same name and is a frequent speaker and trainer in the ITAM/FinOps space, helping customers untangle complex licensing, optimize cloud costs, and understand the impact of new AI pricing models.
3. Setting the Stage
Why did we want to talk about this now? Because with Copilot CoWork and Microsoft Scout, Microsoft has effectively brought usage-based AI billing from the edges of the portfolio right into the heart of Microsoft 365. AI in the enterprise is no longer just a cool feature you get “included” for a flat monthly fee; it’s now something that can quietly turn into hundreds of thousands—or even millions—per year if you’re not paying attention.
In this blogpost, we walk through what CoWork actually is, how the consumption model works, where Scout fits in, and why this is an inflection point for AI governance, financial planning, and even the way we justify AI projects inside our organizations.
4. Episode Highlights
Highlight 1 – “From seat-based to ‘I hope you’ve set a quota’”
Rich explains that CoWork sits on top of Microsoft 365 Copilot as an additional, consumption-based layer: you still need your Copilot/M365 license, but now the real money is in the credits your tasks consume. He draws a straight line from earlier “optional” pay-as-you-go options (Power Platform, Purview, some SharePoint features, Windows and SQL with Azure Arc) to CoWork as the moment where everyone suddenly pays attention.
“CoWork is the big ‘I’m ready for my close‑up’ moment where everyone is paying attention now.”
Highlight 2 – “The bill for 60 users is how much?!”
Using Microsoft’s own calculator, Rich shows how a 60‑user company could end up at around 164,000 dollars per year in CoWork credits alone, with larger organizations jumping into the hundreds of thousands to millions annually. And that’s on top of your existing Microsoft 365 and Copilot licensing.
“For 60 users it came out at about 164,000 dollars a year… and if you scale it up to 1,680 users it was over 400,000 a year—just for CoWork credits.”
5. Deep Dive – Consumption-based AI: why this model changes everything
For roughly the last 10–15 years, the Microsoft licensing story has been comfortably predictable: per-user, per-month subscriptions across Microsoft 365, Dynamics, and most of the SaaS portfolio. With CoWork, Microsoft is pushing AI into a classic cloud consumption model: you pay not just to have the capability, but for every meaningful unit of usage.
CoWork is Microsoft’s take on “closed CoWork”: semi-autonomous agents that you can set off on a task — reviewing contracts, redlining documents, building drafts — while you go and do other work. That’s the promise of AI: it works in the background. But under the hood, every “light” or “heavy” prompt consumes credits, and Microsoft’s calculator uses personas (knowledge worker, customer-facing, technical, leadership) plus estimated prompt types to estimate monthly consumption.
There are two big problems with this model today:
- The numbers are opaque and still evolving. Different versions of Microsoft’s calculators already show slightly different credit assumptions for the same personas, and no one really knows if a “heavy technical prompt” will actually match the estimate of 1,500 credits in real-life.
- The user is in control of spend. Even if the CFO sets a policy—“we only use CoWork for X, Y, Z tasks”—individual users can still decide to throw massively complex tasks at it, or chain prompts in unexpected ways, and the organization just has to pay the resulting bill.
This creates a dynamic that’s very similar to the early days of cloud: organizations forecast 100,000 per month and end up spending 1 million, often without a clear understanding of which workloads or behaviors drove the overrun. The difference now is that we’re not just talking about compute and storage; we’re talking about AI tasks that feel “one click away” and are driven by knowledge workers, business users, and autonomous agents.
6. Real-Life Stories & Examples
Rich shares multiple tangible examples that make these abstract numbers feel real. In one scenario, he models a 60‑user organization where CoWork credits alone reach roughly 164,000 dollars per year, and a 1,680‑user scenario that tips over 400,000 dollars annually—again, just for CoWork, before you even factor in E3/E5 or Copilot seat licenses.
He also points to the pattern Microsoft has followed for years: gradually shifting metrics from per server → per processor → per core → per user and now into per usage with cloud and AI. Initially there’s uproar—like when SQL Server went per core—but give it a version or two and everyone forgets there was ever a different way.
We discuss “inference whales”: customers paying 20 dollars a month for an AI subscription but quietly consuming 200,000 dollars worth of underlying infrastructure in the background, a situation AI providers are now confronting as they push toward profitability. Rich expects Microsoft’s move will set the pattern for the rest of the industry, with other SaaS and AI vendors following with similar hybrid seat-and-consumption models.
There are also the classic cloud gotchas: forgotten dev environments that rack up 50,000-pound bills, or autonomous agents chained together so that one seemingly innocent action fires off 64 different agents, each burning through credits in different services. In the AI agent world, Rich warns, we’ll likely see agents that keep running long after their creator has left the company, quietly consuming budget until someone finds them.
7. Key Takeaways
- AI costs are shifting from fixed to variable. CoWork, Scout, GitHub AI features, Copilot Studio agents, and even Agent 365 are all moving toward usage-based billing, making AI spend more volatile and harder to forecast.
- Seat licenses are just the entry ticket. Your Copilot/M365 license gives you the right to use CoWork, but the real cost is in the credits you burn per task—potentially hundreds of thousands per year even for mid-sized companies.
- Forecasting AI spend is as hard (or harder) than cloud forecasting. We still struggle to predict cloud costs years after public cloud launched; trying to predict how humans and autonomous agents will use AI is at least as difficult.
- User behavior now directly drives the bill. Even with quotas, caps, and guidelines, users can chain tasks, use heavier prompts, and experiment in ways that dramatically shift consumption.
- You must define ‘why’ you’re using AI. If you don’t know whether your goal is efficiency, headcount reduction, new revenue, or better employee experience, you can’t judge whether the spend is justified.
- Ownership and FinOps for AI are essential. Someone in the organization—CFO, Head of FinOps, Head of AI—needs to own the question: “What value are we getting for this AI spend?” and connect cloud, licensing, and business outcomes.
- Expect every major vendor to follow. Rich expects most SaaS products to adopt “per user + per consumption” models; Microsoft just has the scale and confidence to go first.
8. Closing Thoughts
From our perspective, CoWork and Scout mark a turning point: AI in the enterprise is no longer just a “cool feature” experiment—it’s a line item big enough to worry CFOs, FinOps teams, and architects alike. We’re excited about what CoWork-style agents and autonomous tools can do, but we’re equally convinced that organizations will need stronger governance, cost controls, and a clear definition of value before they roll these capabilities out at scale.
In upcoming episodes and blogposts, we’ll keep exploring this intersection of AI, licensing, FinOps, and governance: how to build AI agents responsibly, how to design quotas and guardrails that don’t kill innovation, and how to measure ROI on real-world AI use cases. We’d love to hear how you’re approaching CoWork, Scout, and other consumption-based AI tools in your organization—drop us a comment, reach out to us through your preferred channel, or share your own “surprise bill” stories so we can all learn from them.
