Microsoft Copilot Chief Predicts AI Agents Will Squeeze Business Software Pricing

"If my agent interacts with your app exclusively, the agent's your customer, not the end user," Lamanna said Wednesday morning, speaking with Madrona's Matt McIlwain in the opening keynote at the venture capital firm's annual IA40 Summit in Seattle. "And agents are very harsh customers."

Lamanna is Microsoft's executive vice president for Copilot, Agents and Platform, so this isn't a startup founder taking shots at incumbents. Madrona's agenda lists the session as "Keynote: Agency in the Agentic Era — Building the Next Generation of Apps and Platforms" at 9:10 a.m. on September 30. The summit ran September 29 – 30, 2026 · Four Seasons, Seattle.

What Lamanna actually argued​

His case is simple. When software is used mostly by an agent working for a person, the vendor loses direct contact with that person. According to GeekWire, agents will pick whatever service is cheapest, fastest and most reliable, he said, and they'll move people to another one quickly. Apps that sit behind agents will have a hard time holding on to pricing power or customer loyalty.

He sorted business software into two groups:

• "Thick" apps are where people spend most of their workweek. His examples were CAD programs, contact-center software and Excel for someone who builds financial models all day.
• "Thin" apps are the ones people open, use briefly and close. Lamanna said this covers the vast majority of business software. He expects those apps to end up running "headless," meaning agents such as Copilot and ChatGPT use them through the back end and the user never sees their interface.

He compared a headless app to a company that sells only on Amazon. It can still make money, but prices get squeezed when buyers are simply choosing between "your pencils or those pencils." A brand like Nike has more room to set its own terms.

Section summary: Lamanna expects software people live in to keep its customers. Software people just pass through becomes a back-end service, and agents shopping for back-end services compare price, speed and reliability without any brand loyalty.

Microsoft is on the list too​

Lamanna didn't exempt his own products. Word, Excel and PowerPoint won't go away, Lamanna said, but most people will use them inside Copilot, which can now run Office without switching apps. In his view, the standard setup for an office worker ends up as Copilot for AI and Teams for communication, "and that's maybe it."

Microsoft has already built the product behind that forecast. On September 25, five days before the keynote, Microsoft chief marketing officer for AI at Work Jared Spataro announced a redesigned Copilot app with three new pieces:

Two points from Spataro's post are useful here. Microsoft says work in Copilot stays synchronized with the Office apps, which fits Lamanna's statement that Word, Excel and PowerPoint aren't going away. And Microsoft describes the Office apps as sharing the Copilot front door, not being replaced by it.

That leaves a question for Microsoft: if Word becomes something people mostly reach through Copilot, has Microsoft turned its best-known products into thin apps? Microsoft's answer seems to be that it owns the agent those apps sit behind, so any leverage lost by Word ends up with Copilot.

The irony in the "headless" argument​

GeekWire's August 6 profile of Lamanna adds context. Lamanna said then that he isn't a fan of the word "headless" because it suggests the software is dumb. He also said you can't simply connect an AI model to an API built ten years ago without working out cost, performance and retrieval.

In the same interview he described Work IQ as becoming the headless version of Microsoft 365, so users could get to email, documents and files without opening Outlook or SharePoint. He also said Microsoft wants to be "the Copilot company," with Copilot as the front door to Dynamics, GitHub, Exchange, SharePoint and OneDrive.

So Wednesday's warning also describes Microsoft's own plan: become the agent layer other vendors have to serve, and keep Microsoft 365 as the back end that agents call constantly. In August Lamanna predicted a large surge of demand on Microsoft 365 back-end services because agents never stop working.

Section summary: Lamanna isn't only describing a trend from the outside. Microsoft is building the agent layer that would take pricing power away from thin apps, and it is converting its own services to work behind that layer.

Microsoft needs other vendors' cooperation​

Microsoft's position has a weak spot, and GeekWire pointed it out: Copilot works only if other companies let their software run behind it. In the August profile Lamanna explained that Copilot uses connectors to pull data from services such as Salesforce and ServiceNow so people don't have to leave the app.

That raises an obvious problem. If Salesforce's leaders heard Lamanna say that apps behind agents lose pricing power and loyalty, why would they want to sit behind Copilot? GeekWire's profile noted that Salesforce and SAP are pursuing their own "headless" strategies. Every major vendor wants to be the agent, and none of them wants to be one of the interchangeable pencils.

Microsoft's September 25 announcement shows how it plans to handle this. It introduced a plugin registry that puts Microsoft, partner and custom plugins in one catalog, where IT approves and manages them centrally and developers publish once to reach supported Copilot experiences. That is a marketplace, and Microsoft runs it. Whoever controls the marketplace usually has the leverage, which is the same position Lamanna attributed to Amazon in his pencil analogy.

Amazon shows platforms can refuse​

Amazon's recent actions show both how Lamanna's idea could play out and what limits it. It took two separate steps within a few days.

First, it blocked a consumer agent. GeekWire reported on September 20 that Amazon cut off Meta's Muse agent from shopping on Amazon.com on behalf of customers. Amazon said Meta never asked permission, that Muse doesn't identify itself while browsing, and that it appears to capture and store customer credentials. Meta disputes the credentials claim. It has said Muse has no visibility into people's passwords or payment methods, and that credentials go into secure storage the agent uses without seeing them. Amazon is also fighting Perplexity in court over the Comet browser agent. According to GeekWire, the Ninth Circuit ruled on August 4 that under federal anti-hacking law it was the user, not the AI company, who was accessing Amazon's computers. Amazon's rehearing petition was denied September 10, which leaves contract and terms-of-service claims as its remaining legal route.

Second, it let a business agent in on its own terms. On September 23 Amazon announced a Selling Partner plugin that lets merchants manage listings, inventory and prices from Anthropic's Claude or Amazon's own Quick assistant without logging into Seller Central. Sellers choose which data the plugin can reach and approve each action. The plugin is in beta in Amazon's U.S. stores.

Lamanna mentioned Muse himself during the keynote and predicted it will reach far more people than work-focused agent tools such as Copilot and Claude Cowork.

Amazon's two moves show that a platform with enough market power doesn't have to become a commodity back end. It can decide which agents get access, require them to identify themselves, and require permission first. Lamanna's forecast fits thin apps that can't afford to refuse agent traffic much better than it fits platforms that can.

What this means for IT and developers​

Lamanna gave a strategy talk, not a product briefing, and he offered no numbers on how far prices might fall. Still, some practical points follow for WindowsForum readers:

1. Check which of your apps are "thin." If staff only open a SaaS tool to look something up or file a ticket, that tool is the kind Lamanna expects agents to take over. Your renewal leverage may improve once an agent could replace it.
2. Budget for AI spend as well as seats. Microsoft's September 25 post splits Copilot pricing in two. A user subscription license covers everyday use at a fixed cost. Usage-based billing covers agent-heavy capabilities such as Cowork, Code and Autopilot. If agents become the main users of your software, your costs will follow agent activity, not headcount.
3. Use the cost controls. Microsoft says Agent 365 cost management now covers Code and Copilot Managed Runtime, with Copilot Studio agents planned for October. Admins can set spending policies, send credit requests through approval workflows and decide which model families each group of users can access.
4. Govern plugins deliberately. Central approval in the new plugin registry is where IT decides which third-party services Copilot can reach. In practice that makes you the gatekeeper for which vendors get traffic from your agents.
5. Developers should design for agents. Lamanna's August point about old APIs matters here. If your product is thin, an agent-ready interface with good cost, speed and retrieval performance is what will get it chosen. Agents won't be won over by a nicer UI.

The bottom line​

Lamanna's warning is a forecast, and a convenient one for Microsoft. By his own reasoning, the most valuable place to be is the agent at the front, and Microsoft is working hard to put Copilot there. His thick-versus-thin distinction is still a useful test for any software portfolio: if people don't spend real time in an app, an agent may soon choose it for them based mostly on price.

It's reasonable to doubt how fast this will happen. Amazon's split approach to agents, the legal fights over agent access and Microsoft's own need for Salesforce and ServiceNow to cooperate all suggest the change will be negotiated, not automatic. For more on this topic, see WindowsForum's coverage of Copilot licensing and E7 tiers, Copilot Studio agent governance, and the rollout of Microsoft 365 Frontier program features.