Clara Bennett: I think the industry got this backwards — built the buffer first, then decided the buffer was waste. And now they're removing it faster than anyone has actually tested whether the replacement works.
Clara Bennett: The buffer I'm talking about is bench strength. In practice, bench strength in Indian IT is the share of full-time employees sitting on payroll but not on a live client project at any given moment. Freshers being trained. Engineers between assignments. The firm pays them — nobody bills for them yet.
Clara Bennett: The pre-AI norm across TCS, Infosys, HCLTech, Wipro, and Tech Mahindra was twenty to thirty percent of total headcount. That's not a rounding error. That's a design choice.
Clara Bennett: And the design choice made sense. Indian IT services runs on large, sudden contracts. A client wants five hundred engineers staffed in six weeks — if your bench is empty, you either scramble badly or you lose the deal. The bench was surge capacity. It was how you said yes at scale without breaking delivery.
Clara Bennett: Now — as of August 3rd, 2026 — those same five firms are targeting eight to ten percent bench strength in FY27. TeamLease Digital, which tracks this data, says bench has already compressed to late single digits or around twelve percent over the last three years. EIIRTrend puts it at eight to fifteen percent — the exact figure varies by who's measuring, and that inconsistency is real and worth noting.
Clara Bennett: In aggregate, across TCS, Infosys, HCLTech, Wipro, and Tech Mahindra — roughly seventy-five thousand bench positions have already been eliminated.
Clara Bennett: Seventy-five thousand.
Clara Bennett: The mechanism doing this is what's called demand-driven hiring — you recruit when a client contract is confirmed, not before. AI workforce planning tools have gotten precise enough that firms feel comfortable eliminating the standing reserve. And separately, AI is compressing the actual labor per project — automating coding, testing, maintenance — so fewer human hours are needed per unit of output anyway.
Clara Bennett: The bench timelines reflect it. Redeployment between projects used to take forty-five to sixty days. It's now running around thirty to forty-five. The bench is smaller and it moves faster.
Clara Bennett: And this is not only an Indian IT story. Late July 2026 — Visa announced cuts of approximately twenty-six hundred employees, about seven percent of its workforce. Ryan McInerney, Visa's CEO, wrote in the staff memo that AI is enabling faster product development. That's the explicit stated reason. A global payments company citing AI productivity as grounds for workforce reduction — same logic, entirely different sector.
Clara Bennett: The pattern is consistent enough that it deserves to be taken seriously.
Clara Bennett: But — and I want to be honest about this — the thing I keep pressing on is that AI demand forecasting has never been tested through a genuine demand shock. The bench existed precisely because demand is uncertain. We're removing the buffer on the assumption that the forecasting is now accurate enough to replace it.
Clara Bennett: That assumption is the whole load-bearing wall here.
Clara Bennett: Let's stay with that, because the evidence on the other side of this argument is not nothing — and I think it deserves a real look.
Clara Bennett: The seventy-five thousand — I want to stay with that number a moment longer, because it's easy to let it just... abstract away.
Clara Bennett: Bench strength, in practice, is where freshers live. New graduates on payroll, in training, not yet billable. Engineers between projects being upskilled for the next one. That's who sits on the bench. That's who that number represents.
Clara Bennett: The onboarding runway just collapsed.
Clara Bennett: Kaelyn Lowmaster at Gartner flagged this — surfaced around August 2026 — warning specifically that cutting early-career pipelines risks creating workforce gaps you won't feel for two or three years. Not this quarter. 2028.
Clara Bennett: Firms frame this as leaner operations. Not a layoff wave — just… not replacing attrition. But the line between those two things is semantically slippery, and I think we should say so out loud.
Clara Bennett: The official counter-narrative is reskilling. The idea that benched employees aren't being exited — they're being retrained for AI-adjacent roles, higher-value work. Sumit Pokharna's commentary on AI-driven workforce planning in Indian IT gestures at this. So does the general framing from TCS, Infosys, HCLTech.
Clara Bennett: No source I've found provides concrete data on how many benched employees are actually retrained versus quietly exited. The assertion is doing enormous work to cover what might be a very large number of quiet departures.
Clara Bennett: Enormous work. Unverified.
Clara Bennett: And then there's the Forrester data, which I think is the structural crack in this whole story. Fifty-five percent of leaders report regretting AI or tech-driven workforce cuts. Forrester also predicts that half of such layoffs will reverse — and they point to Ford and IBM as companies already rehiring after AI-driven reductions.
Clara Bennett: More than half. Already regretting it.
Clara Bennett: If that reversal pattern holds here — if TCS or Infosys or Wipro needs to rebuild bench capacity in 2028 — the institutional knowledge and the hiring infrastructure lost in this transition becomes the real cost. You can't just turn the pipeline back on.
Clara Bennett: There is a contrast worth noting. American Express is handling AI-related job reductions largely through attrition — not direct cuts. No memo, no explicit link. Compare that to Ryan McInerney at Visa writing explicitly that AI is enabling faster product development as grounds for twenty-six hundred jobs gone. Same AI productivity logic. Very different execution.
Clara Bennett: That range matters. Because the Indian IT framing — leaner bench, demand-driven hiring, AI workforce planning — it borrows the efficiency language without being transparent about which employees absorb the downside.
Clara Bennett: And those employees, in this case, are mostly early-career. That's who the bench was. That's whose trajectory just changed — quietly, without a headline.
Clara Bennett: Now — the number that actually settles this isn't seventy-five thousand. It's eight percent.
Clara Bennett: Because here's the operational reality that Monica Singh's framing in DNA India — 'From 30% to 10%: How AI is forcing TCS, Infosys, HCLTech to run leaner benches' — that headline captures the direction but not the precision problem underneath it. TeamLease Digital is already reading bench in the late single digits. EIIRTrend puts the range at eight to fifteen. Those two ranges don't fully overlap. And that inconsistency means the industry doesn't have a clean baseline to defend — which matters enormously when you're arguing your model is optimized.
Clara Bennett: Optimized against what number, exactly?
Clara Bennett: In practice, if TeamLease is right — if bench is already in the late single digits at TCS, Infosys, Wipro, HCLTech, Tech Mahindra — then the FY27 target of eight to ten percent isn't a planned reduction. It's a description of where you already are. The 'goal' is retroactive framing for a cut that's already happened.
Clara Bennett: That reframe changes the question completely.
Clara Bennett: The question isn't whether the reduction happens. It's whether firms can sustain eight percent bench through a demand uptick — because at eight percent, there is almost no slack. Bench timelines are now thirty to forty-five days. A sudden three-contract surge in Q1 FY28 — that's not hypothetical, Indian IT gets those — and you are staffing from effectively zero reserve, against a thirty-day clock, with no training runway for new hires.
Clara Bennett: That math is uncomfortable.
Clara Bennett: And Kaelyn Lowmaster at Gartner is pointing at the three-to-five-year version of the same problem — that cutting early-career pipelines now means the mid-level talent base of 2029 simply doesn't exist. You can't compress the bench, exit the freshers, and then expect to have experienced engineers in three years. The pipeline IS the bench. Cut one, you're cutting the other.
Clara Bennett: The number I'd watch — and I mean this as the specific thing to hold onto — is the Forrester prediction. Half of AI-driven cuts reverse within eighteen to twenty-four months. Ford, IBM, already rehiring. If that reversal rate applies here, the firms that cut deepest pay the highest rehiring premium. Demand-driven hiring works when demand is predictable. It gets expensive fast when it isn't.
Clara Bennett: So the question I keep landing on — the one that's genuinely still open — is whether eight percent is a durable floor or an overcorrection that looks rational until the next surge makes it suddenly, visibly wrong.
Clara Bennett: The bet these firms are making — TCS, Infosys, HCLTech, Wipro, Tech Mahindra — is that AI-driven demand forecasting is now precise enough to replace a buffer that took decades to build. That when a client contract lands, they can staff it from near-zero reserve, on a thirty-day clock, without a training runway underneath it. That demand-driven hiring at eight percent bench is a permanent operating model, not a fair-weather one.
Forrester says fifty-five percent of leaders who made that kind of bet already regret it.
Clara Bennett: The cost of being wrong here isn't visible in this quarter's margins. It shows up the moment a demand surge hits and the pipeline isn't there — because the freshers weren't hired, the bench was gone, and the firms that cut deepest are now paying a premium to reconstruct something they already had.
