When we're talking to enterprises evaluating web agent platforms, the first questions aren't about reasoning capabilities. They're operational: What happens when sites change structure overnight? How do I audit what agents actually did? Can I measure whether this is working?

Salesforce's 2025 acquisition spree—eight AI companies including an $8 billion bet on Informatica—shows where defensible value actually lives in the agent market. Look at what they're buying: operational scaffolding.

Consider Spindle AI's observability platform. In production, agents don't just succeed or fail. They succeed differently each time. One run navigates a site in 12 steps, the next takes 15 because a modal appeared. Without observability infrastructure, you can't tell if that variance is acceptable adaptation or drift toward failure. Spindle makes agent behavior measurable, which is what you need before you can make it reliable enough to build business processes around.

Salesforce's shopping list tells the story:

• Observability (Spindle AI): Infrastructure to measure agent behavior variance and detect drift toward failure
• Data governance (Informatica, $8B): Systems to manage data quality and compliance at scale
• Dynamic navigation (Convergence.ai): Capabilities to handle interface changes and unpredictable UI patterns

None of these make good demos. All of them answer operational questions enterprises actually ask.

At TinyFish, where we operate infrastructure running web agents across thousands of sites, we see why these capabilities create defensibility. When a hotel booking site A/B tests its search interface, suddenly half your monitoring runs fail. Not because your agent got dumber. Because the DOM structure changed. When a regional site adds CAPTCHA to one geography but not others, you need infrastructure that detects this and routes around it. When authentication flows break because a site's mobile and desktop versions diverge, you need systems that handle OAuth tokens expiring mid-workflow and 2FA appearing unpredictably.

Operational problems, all of them. They require infrastructure depth—the kind enterprises evaluate when deciding whether to depend on your platform.

ServiceNow's $2.85 billion Moveworks acquisition targets enterprise integration across 100+ systems. CB Insights reports the three largest of 85 AI acquisitions in Q1 2025 went to enterprise agent companies. None for reasoning capabilities. All for operational infrastructure.

Watch what this means for market structure. If reasoning commoditizes—and the trajectory of foundation models suggests it will—then defensibility lives in the layer that makes agents reliable at scale. The web resists automation through bot detection, rate limits, and constant interface changes. Making agents work despite this adversarial environment requires infrastructure that handles authentication labyrinths, monitors for drift, provides audit trails, and measures performance reliably enough that enterprises trust it.

From our perspective operating agents that monitor pricing across fragmented web surfaces, Salesforce's $8 billion answer tracks: In a world where reasoning capabilities become increasingly commoditized, lasting value lives in the infrastructure that makes agents trustworthy enough to automate workflows that matter.

If this pattern holds, we'll see bifurcation. Companies that answer operational questions—Can you handle authentication across 50 regional variations? Can you adapt when sites change? Can you provide audit trails?—build defensible businesses. Companies focused purely on reasoning become feature providers absorbed into someone else's infrastructure.

The competitive moat lives in having infrastructure reliable enough that enterprises confidently answer "yes" to: "Can I build my business process around this?"