Hey folks, Pratham here.

Welcome to Paradox Weekly, a Masters' Union newsletter, where we break down the ideas, trends, and contradictions shaping business today.

Last week, Infosys announced a strategic partnership with Cognition, the startup behind Devin, an AI that can write, test, and deploy code autonomously.

The press release called it "one of the largest global deployments of agentic software engineering to date."

Infosys, a company with 323,578 employees, just partnered with a startup that has about 49 people.

The company that was supposed to be disrupted by AI coding tools is now deploying them across its engineering organization.

That’s today’s paradox we’re going to discuss. But before that…

First cultural immersion of PGP Bharat concluded at Prem Mandir Vrindavan.

Students learnt about organisational operating framework from Mr. Sumanth Rao, Chief Digital Officer of Prudential Insurance.

We had the MU Alumni Home Coming Party. Meeting them makes me realise how fast time goes by!

Back to the story.

Infosys has been quite loud and honest about how they’re deploying an AI coder across their entire engineering organization.

They could be adapting to survive or they could also be accelerating their own obsolescence by making it easier for clients to see how much work AI can do without human engineers.

Devin starts at $500 per month for engineering teams. An entry-level engineer at Infosys earns roughly ₹3.5 to 4 lakhs per year, about $350 per month.

Devin costs more than a fresher in India.

But Devin doesn't need training, a bench period or three attempts to pass an internal assessment.

And that last part is where it makes a lot more sense.

In February 2025, Infosys terminated 350 trainees at its Mysuru campus for failing internal tests.

Since then, the company has laid off roughly 800 trainees total across four rounds. Many were given three hours' notice.

The company now deploying AI to "create virtual engineers" spent most of 2025 firing trainee engineers.

From Q4 FY2023 through Q1 FY2025, Infosys experienced seven consecutive quarters of net headcount decline.

Peak employment was 343,234.
By March 2024, it dropped to 317,240.

The company has since started hiring again, adding 15,000 to 20,000 freshers this fiscal year. Still a fraction of what it used to be.

If Infosys just partnered with an AI that writes code autonomously, why hire 15,000 to 20,000 freshers this year?

The other answer is more interesting.

Deploying Devin across 323,000 engineers, hundreds of clients, and decades of legacy systems takes time.

Devin doesn't need bench time or training, but Devin does need context: which clients use which tech stacks, which legacy systems have undocumented dependencies, which compliance frameworks apply to which industries.

Infosys has been using Devin internally for six months.

That's enough to see "material productivity gains."

It's not enough to replace tens of thousands of engineers.

So they're doing both.
Hiring freshers to serve existing contracts.
Deploying AI to change what future contracts look like.

The freshers joining today might be the last generation hired at this volume.
Or the first generation that learns to work alongside AI agents from day one.

The answer depends on how fast clients demand productivity gains and how quickly Infosys can deploy Devin beyond pilots.

My question

If AI makes engineers 40% more productive, does that mean you need 40% fewer engineers?

Infosys's own research says enterprises expect 10-40% productivity gains from AI adoption. Those gains have to come from somewhere.

The pyramid model that built Indian IT is breaking.
Hire thousands of freshers, bill by the hour, and promote the survivors.

That model assumed human labor was the product.
Now the product is increasingly the output, not the hours.

Infosys framed this partnership as "accelerating AI value realization for global enterprises."

Another way to frame it: a company that laid off 800 trainees throughout 2025 announced an AI partnership on January 7, 2026, and is now positioning itself as the middleman between enterprises and autonomous coding agents.

The trainees who were let go at Mysuru weren't failing because they couldn't code.

When you raise the bar for humans while deploying AI that doesn't need to clear any bar, freshers get the message: the assessment isn't about quality control anymore.

If Infosys's playbook works, every major IT services firm will follow.

They'll partner with AI coding startups, deploy autonomous agents, and shift their workforce from execution to supervision.

Some engineers will thrive.

They'll become the ones who manage fleets of AI agents, who architect systems that AI implements.

Others won't make the transition.

Especially the ones in the "big fat middle layer." The engineers with 5-15 years of experience who are too expensive to keep and too specialized in legacy systems to easily redeploy.

That pressure could make their careers harder or force them to become a different kind of engineer, one who directs AI rather than competes with it.

My two cents

Infosys is deploying the technology that was supposed to make Infosys obsolete. And they might survive because of it.

The company that was supposed to be disrupted is becoming the one that controls how disruption reaches enterprises. Cognition has the AI. Infosys has the clients. Neither can do enterprise deployments alone.

The threat and the mechanism for survival turned out to be the same technology.

Hit reply: Almost everyone in India knows someone at Infosys, TCS, or Wipro.
If you work in IT services, or have family who does, I want to know: what are the conversations happening inside? Is AI seen as opportunity or threat?

I read every email.

Until next week,
Pratham