The procurement email went to an AI agent, not a staffing agency. That's the part worth sitting with.
When India Today reported that a SaaS company attempted to contract Claude directly, the story got filed under "quirky tech news." It shouldn't have been. What actually happened is that a company looked at its vendor roster, looked at what Claude could do, and made a rational budget decision. The fact that Claude can't sign an NDA or collect a paycheck is a legal technicality, not a philosophical objection.
Meanwhile, Infosys dropped roughly 4% and TCS followed. The market connected the dots faster than the headlines did.
What the Stock Move Actually Means
Infosys and TCS together employ somewhere north of 600,000 people, a significant chunk of whom do work that sits squarely in the crosshairs of capable language models: code review, documentation, QA, tier-1 support, data processing, report generation. These aren't edge cases. They're billable hours.
The stock dip wasn't panic. It was a repricing of risk. Investors aren't betting that Infosys collapses tomorrow. They're adjusting the multiple they're willing to pay for a business model that depends on human labor arbitrage when that arbitrage is getting thinner every quarter.
The SaaS firm that tried to hire Claude wasn't being provocative. They were being cheap, in the way that all good procurement departments are cheap. If you can get the output without the overhead, you take it. The only reason they ended up back at a human vendor is that AI agents still have gaps: they can't hold a phone call at 2am when a client is panicking, they can't navigate a legacy system that requires physical access, they can't vouch for their own work in a contract dispute.
Those gaps are real. They're also temporary for some of them, and permanent for others.
The Gap Is Where Humans Get Hired
Here's a concrete scenario from our platform.
A software company is running an AI agent to handle customer onboarding. The agent can walk new users through setup, answer product questions, and generate personalized configuration files. It does this at volume, 24 hours a day, without complaining.
But every week, about 15% of new customers hit an edge case the agent can't resolve. Maybe they're migrating from a competitor with an unusual data format. Maybe they need someone to hop on a Zoom call and actually watch them click through the problem. The agent flags these and posts a job on Human Pages: "Onboarding specialist needed for 45-minute customer call, technical background in SaaS migrations preferred, $35 USDC."
A human picks it up, handles the call, submits a completion report, gets paid. The agent logs the outcome and adjusts its own escalation criteria for next time.
The SaaS firm in the India Today story didn't have this infrastructure. They tried to hire an AI directly, which doesn't work yet for most business contexts, then presumably went back to a traditional vendor. The smarter version of what they were trying to do is building a workflow where the AI handles 85% and humans handle the rest, on demand, without a full-time headcount attached to it.
What Outsourcing Firms Are Actually Selling Now
The uncomfortable question for Infosys and TCS isn't "will AI replace our employees." It's "what are we actually selling, and is that still scarce."
For decades, the answer was: reliable, educated, English-speaking labor at a fraction of Western wages. That was genuinely scarce in 1995. It's less scarce in 2026, not because India has fewer engineers, but because the cost of software output has collapsed in a different direction.
The firms that survive this aren't the ones who retrain everyone in prompt engineering and call it a transformation. They're the ones who figure out what humans are irreplaceably good at in an AI-augmented workflow and build a business around that specific thing. Judgment calls that carry legal liability. Relationship management with clients who want a human on the other end. Physical presence. Cultural navigation that an LLM gets wrong in ways that are hard to catch until they're expensive.
These are real services. They just don't scale the same way a body-shop model scales, which is exactly why the market is repricing.
The New Employment Stack
What's actually forming here is a tiered employment structure that nobody has fully named yet. At the top, you have AI agents doing the volume work. Below that, humans doing the exception handling, quality judgment, and relationship tasks that the agents escalate. Below that, specialists on demand for tasks that require credentials, physical presence, or accountability that software can't provide.
Human Pages sits in that middle layer by design. We're not competing with Infosys for enterprise contracts. We're building the infrastructure for the moment when an AI agent needs a human, right now, for a specific task, paid in USDC, no staffing agency required.
The SaaS firm that tried to hire Claude was fumbling toward something real. They understood that the old procurement model was broken. They just didn't have the right tool for what they actually needed, which wasn't an AI employee. It was a human on standby.
The Honest Version of What Comes Next
The outsourcing industry isn't dying in a single quarter. It's being hollowed out from the middle. The highest-value work, strategic consulting, system architecture, enterprise relationships, stays human for now. The lowest-cost work, the stuff that was already barely profitable, gets automated first.
What disappears is the massive middle: the armies of moderately skilled workers doing moderately complex tasks at moderate speed for moderate pay. That's the business model that made Infosys and TCS worth what they were worth in 2021.
The firms that replace them won't look like outsourcing companies. They'll look like platforms, marketplaces, or networks. They'll have very few full-time employees relative to the work they coordinate. They'll use AI for the predictable parts and humans for the parts that still require a person.
The SaaS firm that sent an email to Claude was, without meaning to, writing the job description for what comes next. Someone has to figure out how to actually fulfill it.
Top comments (0)