Box CEO Aaron Levie keeps hearing something unexpected from enterprise customers deploying AI agents. They're using them for work that was never attempted at all. Not because companies chose not to. Because attempting it would lose money at any price point.

His example: a company sitting on 50,000 customer contracts who wants to identify which customers have the highest propensity to buy a new product.

But if an agent could do it for $5,000? "They would do that all day long."

Labor arbitrage misses the point. This is about crossing a threshold from impossible to viable.

Why It Was Actually Impossible

Reading 50,000 contracts sounds like a scaling problem. More contracts, more time, but fundamentally the same task. Except those contracts aren't waiting in a database. They're scattered across procurement systems, email attachments, vendor portals, legacy repositories, third-party platforms. Each with its own authentication requirements, access patterns, data formats.

Building enterprise web agent infrastructure that runs reliable workflows at scale reveals something counterintuitive: the barrier to certain tasks wasn't cost. It was architecture.

82% of enterprises report that data silos disrupt critical workflows, and 68% of enterprise data remains unanalyzed precisely because accessing it requires navigating this fragmentation. The technical reality: application-level fragmentation makes it "difficult or impossible for an application to access and use data that's been stored by another application." When data lives across disconnected systems, the cost of attempting to access it exceeds any value you could extract.

The web wasn't built for programmatic access at enterprise scale. Modern websites deploy behavioral analysis systems that track how browsers actually behave. Mouse movements, scrolling patterns, the unique fingerprint of your device. Making automation look human isn't a prompt engineering problem. It's an architecture problem.

The cost structure compounds the impossibility. Rendering pages with headless browsers through residential proxies requires 2MB per page versus 250kb without. At scale, that's prohibitive enough that attempting the work guarantees you'll lose money. Add maintenance burden: websites change constantly, breaking automations. IT teams spend 19 weeks annually just managing data infrastructure.

For work that wasn't generating value in the first place, that math never closes.

What This Actually Unlocks

Start with a different question. Not "what would you automate with cheaper labor?" but "what work exists in your organization that you literally can't see or access today?"

Analyzing every customer interaction for upsell signals. Reviewing every contract for risk patterns. Monitoring every competitor move across fragmented web surfaces. Work that was always valuable but never viable because humans were never doing it. It was "too expensive to send people to go off and look through."

The unlock here is making invisible work visible and economically viable. That requires infrastructure that can navigate authentication labyrinths, handle anti-bot systems, maintain reliability across thousands of concurrent sessions, and turn unstructured web surfaces into structured, reusable data.

When you build that infrastructure, you're not replacing human labor. You're enabling work that never existed because the web outgrew the tools we had to interact with it programmatically. The vast majority of agent work that Levie describes as "new" isn't replacing jobs. It's filling a void that's been there all along. Work that was always valuable but technically impossible to attempt.

Infrastructure finally catching up to what the web became.