The Lenient Project Is Dead: AI and IT Services Margins
Agentic AI isn't killing IT services, it's killing the slack. Why margin compression is reshaping TCS, Infosys and Wipro, and the pivot that survives.
Note: Statistics and figures reflect data available as of July 2026. Verify for latest figures.
A client I'll keep anonymous walked into a renewal conversation last quarter with a single slide. On it: our original proposal for a modernization program, staffed at 40 people over 18 months, and next to it their counter-offer. Twelve people. Six months. Same scope.
Their reasoning fit in one sentence: "You have AI now."
I've replayed that meeting a dozen times. The uncomfortable part isn't that they were being unreasonable. The uncomfortable part is that, on the specific workstreams they pointed at, they were roughly right. And every delivery lead I compare notes with is having some version of the same conversation.
That anecdote is a composite, stitched from a few real negotiations rather than one deal I can name. But the pattern is not composite at all. It is showing up in the numbers.
Here is my thesis, stated plainly: agentic AI is not killing IT services. It is killing the slack in IT services. The padding that quietly funded our margins for two decades is evaporating, and no amount of press-release optimism changes that math.
The old margin machine
To understand what's breaking, you have to be honest about how the money actually got made.
The classic Indian IT services engine ran on three fuels. First, time-and-materials billing, where you get paid per person per month regardless of how efficiently that person works. Second, bench economics, where a pool of paid-but-unassigned engineers lets you staff deals fast and bill them soon. Third, effort-based estimation, where a project is priced by how many person-hours it "should" take, anchored to historical productivity that everyone silently knew was conservative.
None of this was fraud. It was the operating model. But notice what all three have in common: they reward volume of human effort, not outcomes.
The most profitable projects in that world were what I privately called the lenient ones. Generous scope. Forgiving timelines. A client who valued predictability over speed and didn't scrutinize whether a 40-person team truly needed 40 people. Those engagements had room to breathe, and that room was the margin.
Agentic AI walks straight into that room and turns on the lights.
What agentic AI actually changes
The shift isn't that AI writes some code. We've had autocomplete for years. The shift is that clients have recalibrated what they believe is possible, and that recalibration resets their willingness to pay.
TCS CEO K. Krithivasan told analysts on the Q1 FY27 earnings call that a 10-15% productivity improvement from AI is "achievable from day one on an annual basis," versus the 3-5% the industry historically booked. Sit with that. Even the conservative, blended, whole-portfolio number from the largest Indian IT firm is a step-change, not a rounding error.
You'll hear louder claims elsewhere. Vendors and some client-side benchmarks float 2-3x output for a single engineer paired with agents on scoped tasks. I'd treat those with caution; they describe narrow, favorable workloads, not a full delivery lifecycle with its meetings, integration, and rework. But you don't need 3x to break the old model. You need clients who believe a smaller team can do the job, and TCS's own 10-15% is enough to seed that belief across every procurement team on earth.
That belief lands in three places:
- Productivity expectations reset. Buyers now open negotiations assuming AI-level efficiency and price accordingly. The burden of proof has flipped: you have to justify the headcount, not them.
- Per-seat pricing gets squeezed. When the client assumes one engineer plus agents covers what three used to, they resist paying for three seats. The unit you were selling is losing its meaning.
- RFPs demand AI-native delivery. "How will you use AI to reduce cost and timeline?" is now a scored line item, not a nice-to-have. Wipro CEO Srini Pallia framed it directly in the FY26 results, saying advancements in AI "are reshaping client priorities."
The compression evidence
This is where I have to be careful, because the honest picture is more nuanced than the doom headlines, and the nuance matters.
Start with the clearest signal: margins. TCS reported a Q1 FY27 operating margin of 24.0%, down 130 basis points quarter-on-quarter, per its July results. Management attributed most of that to annual wage increments and guided to recover toward 25%+ by year-end. So this specific dip is partly seasonal, not a pure AI story, and I won't pretend otherwise.
But look at the growth alongside it. TCS grew revenue just 3.2% year-on-year in constant currency (13.9% in reported rupee terms, flattered by currency), per the same results. Infosys, meanwhile, set FY27 guidance of only 1.5-3.5% constant-currency growth with margins of 20-22%, per its FY26 earnings. Wipro went further and guided Q1 FY27 IT services revenue to -2.0% to 0% sequentially in constant currency. That is a market leader openly signaling a possibly shrinking quarter.
Now here's the twist I want to flag honestly, because the brief I was working from assumed "deal sizes shrinking, volumes rising," and the data doesn't cleanly support that. Order books are actually robust. TCS booked $9.5 billion in total contract value in Q1 FY27, including a single AI-led transformation deal with SKF worth over $800 million. Infosys landed $14.9 billion in large-deal TCV across FY26. Deals are not drying up.
So where's the compression? It's in the conversion. Strong bookings and AI-led deals are not translating into the revenue growth or margin expansion the same TCV would have produced five years ago. TCS's incremental AI revenue was $75 million this quarter, down from $125 million the prior one, which management called "lumpy." You're winning the work and getting paid less per unit of it. That is exactly what killing the slack looks like: the machine still runs, it just produces thinner margins per rupee of contract.
Zoom out and the industry is still growing. NASSCOM projects India's tech sector to reach $315 billion in FY26, up 6.1%, with exports around $246 billion, per its Strategic Review 2026. Gartner sees worldwide IT services spend surpassing $1.87 trillion in 2026, the largest single spending category, per its April 2026 forecast. The pie is expanding. The problem is that the profitable shape of the old slice is deforming.
The counterargument I owe you
Every few years someone declares Indian IT dead. Y2K remediation was supposed to be a one-off sugar high that would end us. Cloud was going to disintermediate the system integrators. RPA was going to automate away the BPO tier. Each time, the services industry adapted, absorbed the disruption, and grew.
There's a strong version of this optimism, and I hold part of it. The very AI wave compressing our margins is also creating a genuine new revenue pool: enterprises need someone to migrate off bloated SaaS stacks, rebuild workflows around agents, and integrate models into legacy estates that will not modernize themselves. I wrote about that opportunity in detail in the SaaSpocalypse piece, and I stand by it. Indian IT is arguably better positioned than anyone to win the migration and integration wars.
But adaptation is not the same as continuity. The firms that survived Y2K and cloud didn't do it by protecting the old model. They did it by cannibalizing it before someone else did. Optimism about the new pool is warranted. Complacency about the old margins is not. Both things are true.
The pivot playbook
If the slack is gone, the answer isn't to defend it. It's to sell things where AI helps you more than it helps the client's imagination. Four moves matter.
1. Outcome-based pricing
Stop selling seats and start selling results. If agents make your delivery cheaper, capture that value by pricing to the business outcome, tickets resolved, claims processed, migration completed, rather than to hours burned. Wipro is publicly pushing a "services-as-a-software" model through its new AI Native Business and Platforms unit, per its FY26 commentary, and that phrase is the whole strategy in three words: charge like software, not like a staffing agency.
2. AI governance and audit as a service
Someone has to make enterprise AI safe, compliant, and explainable, and clients cannot do it alone. The EU AI Act's high-risk system rules are still nominally due to bite on August 2, 2026, with non-compliance penalties up to EUR 15 million or 3% of global turnover, though a provisional Digital Omnibus agreement reached in May 2026 could push that deadline to December 2027 for some categories if formally adopted, which I covered in the EU AI Act guide. Either way, governance, documentation, model auditing, and compliance are high-trust, hard-to-automate work. That is exactly where a services firm's institutional credibility becomes a moat instead of a cost.
3. Agent orchestration as managed service
The demos are easy; keeping fleets of agents reliable in production is not. TCS described a retailer running 70 AI agents across 60+ IT-ops workflows, cutting incidents by 80%, per its Q1 FY27 call. Standing that up and keeping it running is a recurring, sticky, ownership-based engagement. Managed orchestration is the new managed services, and it doesn't depend on billing bodies.
4. Proprietary accelerators and IP
Pure labor arbitrage is the thing agentic AI attacks most directly. Reusable accelerators, domain platforms, and pre-built agent frameworks let you deliver outcomes without linear headcount, which means AI improves your economics instead of only the client's leverage. IP is how you stop competing on price per person.
What this means for you
If you're an IT professional
The skills that survive compression are the ones agents make more valuable, not the ones they replace. Orchestration over execution: designing, supervising, and debugging agent workflows beats hand-cranking the code they now generate. Domain depth: an engineer who understands insurance claims or trade settlement is pricing-power; a generic coder is a line item. And judgment work, architecture, governance, client trust, ambiguous problem framing, is precisely what outcome-based deals reward.
TCS said more than 50% of its Q1 lateral hires had next-generation skills and set a target to convert at least 1% of its base into "forward-deployed engineers," per its results. Read that as the direction of travel. Reskilling isn't optional insurance anymore; it's the job.
If you're a leader
The harder discipline is deciding what to stop selling. Stop defending T&M engagements whose only logic is billable headcount, because a competitor will underprice them with agents and you'll lose the account and the margin. Stop treating the bench as a strategic asset when utilization is the metric under attack. And stop pitching AI as a cost the client should fund, when they now expect it as the baseline you already operate on.
The lenient project, the one with generous scope and forgiving timelines that quietly funded the quarter, is dead. Mourning it is a waste of a good quarter. The firms that win the next five years will be the ones that killed it themselves, on their own terms, and rebuilt around outcomes, orchestration, governance, and IP before the client's single-slide renegotiation did it for them.
Related Reading
Enjoying this article?
Get posts like this in your inbox. No spam, unsubscribe anytime.
