The spark this week wasn’t a new benchmark chart. It was a product move.
Anthropic rolled out new plug‑ins for its Claude “Cowork” agent, pitched at automating chunks of white‑collar workflow across areas like legal work, sales, marketing and data analysis. Investors didn’t treat that as a neat feature. They treated it as a warning label stuck on the old services model: “billable hours may be editable”.
This sits on top of a broader direction Anthropic has been advertising for a while: Claude as an agent that can take actions inside software, not just chat about it. Its “computer use” capability, for example, is explicitly about teaching a model to interpret what’s on a screen and choose clicks/typing steps to complete tasks, with plenty of caveats about brittleness and safe implementation.
Markets don’t need the tool to be perfect today. They just need it to look good enough soon to shift bargaining power.
The outsourcing map that suddenly looks… editable
The headlines swept the globe, and so the underlying fear is global: if agentic AI can chew through repeatable digital work, the big cross‑border delivery pyramid takes a hit everywhere it exists.
That includes:
- Offshore delivery giants (India’s IT services majors).
- BPO and “knowledge work” hubs (South‑East Asia, especially the Philippines).
- Nearshore regions that sell time-zone overlap and language fit (parts of Eastern Europe for EU clients, Latin America for North American clients).
- Data and workflow incumbents in Europe and the US that monetise “expert work wrapped in software”.
The sell‑off that followed made that clear. It wasn’t contained to one market or one type of company.
South Asia: India wore the first punch, because it’s the poster child for labour leverage
India’s listed IT services names copped the most immediate market reaction because they’re the most visible expression of the model investors are questioning: large teams, repeatable delivery, and margins defended through scale.
Reuters linked the move directly to concerns that Claude’s new plug‑ins could reduce the need for large staffing models, with the Nifty IT index dropping about 6.3% and major firms (including Infosys, TCS and HCLTech) down sharply on the day.
The market logic is blunt:
- If buyers think agents can do first‑pass tickets, test runs, documentation, and routine workflows, they’ll ask for productivity givebacks.
- If pricing compresses faster than delivery models can be rebuilt, margins wobble.
- If the pyramid compresses at the bottom, entry‑level hiring becomes the pressure valve.
None of that means “India is finished”. It means the easy story (cheap labour + scale = predictable earnings) looks less frictionless than it did.
Europe: the data and “expert workflow” businesses suddenly look contestable
If India is the labour‑arbitrage heartland, Europe is where a different kind of services premium lives: highly curated content, legal and regulatory workflow, and “information-as-a-must-have”.
That’s why the jolt wasn’t confined to offshore IT. Reuters reported a sharp sell‑off in European and US data/software/professional services stocks after Anthropic’s move, with firms like Thomson Reuters, RELX and Wolters Kluwer hit hard as investors worried AI could eat into their workflow and information businesses.
This is the important nuance: the market wasn’t just pricing “outsourcing gets cheaper”. It was also pricing “the tools that sell expertise may get unbundled”.
South‑East Asia: the Philippines is the other big “services engine” everyone watches
The Philippines is a different flavour of the same story: huge scale in outsourced services, but a mix that includes both contact-centre work and more complex BPM/IT roles.
A Reuters report on the sector (based on IBPAP’s figures) described the industry hitting about 1.82 million jobs and US$38 billion in revenue in 2024, while industry leaders argued generative AI would be more augmentation than apocalypse, paired with upskilling into higher-value work.
So the near‑term question in Manila isn’t “does AI arrive?” It’s “what gets automated first?”
Agentic AI is most likely to bite hardest where:
- processes are standardised,
- outputs are easy to check,
- and the work is already sitting inside software systems that can be driven by tools.
That’s a lot of back-office work. It’s also a lot of what BPO has historically been paid to do.
Americas and nearshore: Latin America and Eastern Europe don’t escape, but the pitch may change
Nearshore regions have been selling a simple bundle for years: closer time zones, easier communication, solid talent, and lower costs than onshore. Agentic AI doesn’t remove that demand overnight, but it changes what clients want to buy.
Instead of “send me a team”, the request starts to look like:
- “send me operators who can supervise agents”,
- “send me domain specialists who can validate outputs”,
- “send me platform engineers who can wire this into our systems safely.”
And importantly: some of the biggest global services firms are not denying this shift. They’re trying to become the channel through which it happens. Accenture, for example, expanded its partnership with Anthropic and said it would train a large cohort of staff on Claude as part of a push to drive AI adoption in enterprise delivery.
Cognizant’s own outlook this week leaned into the “AI demand boosts services” narrative, pointing to strong demand as enterprises integrate AI and move to cloud, and noting partnerships including Anthropic.
So while markets were doing the fear trade, parts of the services industry were simultaneously saying: “Yes, and we intend to sell the shovels.”
The stock move is really about pricing power and who owns the workflow
Strip out the theatre and you’re left with two investor questions:
1) Who captures the productivity dividend?
If clients believe agents cut effort, they’ll want a share of the savings. That’s a direct threat to labour‑indexed revenue.
2) Who owns the workflow and the data?
If agent tooling becomes broadly available, the value may migrate away from “who supplies the headcount” toward “who controls the process, permissions, audit trail and data access”.
That dynamic isn’t unique to India. It plays out in every region where services have been priced like labour.
Reality check: agents still can’t sign an SLA
The market narrative can run ahead of the engineering. Even Anthropic’s own documentation and research writing around agentic/computer‑use capability stresses limitations and the need for safeguards and oversight.
In enterprise reality, a few constraints still keep humans and services firms in the loop:
- Accountability: AI doesn’t take contractual responsibility for outages, compliance failures or customer harm.
- Permissions and governance: agents need tight access controls, audit logs and “least privilege” by design.
- Legacy sprawl: automating a modern SaaS workflow is one thing; automating 17 stitched-together systems from 1998 is another.
- QA bottlenecks: the work often shifts from doing to reviewing, and review still needs skilled people.
So the nearer‑term outcome is less “services replaced” and more “services re-priced and re-shaped”.
What to watch next
If you want signals that matter more than demos, watch for:
- Client language in earnings calls about “productivity pass‑through” and rate resets.
- Hiring mix changes (especially entry‑level intake) at large services firms after a few quarters of agent rollout.
- Platform moves: vendors bundling governance, auditability and safe tool-use into “enterprise agent stacks”, because that’s where the friction lives.
- Whether the sell‑off broadens or fades as executives (like Nvidia’s Jensen Huang) keep pushing the “augment, don’t replace” counter‑narrative.
Anthropic’s move gave markets a concrete excuse to reprice the global services economy. From Bengaluru to Amsterdam to Manila, the same question is now on the table: if agents can do more of the grind, who still gets paid for the grind, and who gets stuck holding the risk?