If your 2027 workforce plan assumes AI agents will pay for themselves through headcount cuts, a 2,005-executive survey says most companies already running agents have not made those cuts. Genpact's Autonomy in Practice report, published November 12, found that 69% of enterprises with agents in live production grew their agent population over the past year, while 92% held human headcount flat or increased it. The companies furthest ahead stand apart on two other decisions: creating roles built for human-agent work (51% of Leaders versus 22% of everyone else) and naming who owns agent value on the P&L (49% versus 7%). Those are the two decisions to make before budget season closes.
One caveat runs through everything below. Genpact sells agentic operations services, so its research arrives with a preferred conclusion. The numbers are still useful. Read them as a vendor's survey.
What Did Genpact's Survey Actually Measure?
The report surveyed 2,005 senior enterprise decision makers across 17 industries and 14 business functions, at companies with annual revenue from $1 billion to more than $50 billion, according to Genpact's report page, supplemented by interviews with enterprise and technology leaders. The press release notes that the findings are self-reported and rounded.
The headline adoption numbers, all from that release:
- 79% of enterprises have AI agents in production, but only 20% use them broadly and consistently across functions.
- 15% have fully tied AI investment to P&L outcomes.
- On average, 49% of AI initiatives funded in the past 24 months delivered the expected value.
- 9% have completed significant operating model transformation.
- AI now leads or fully runs 40% of decisions, more than double a year earlier, yet humans still decide or audit in at least 87% of cases at every decision stage.
Genpact sorts respondents into four tiers on its AI Maturity Index. Leaders, the top tier, grew from 12% of the sample in 2025 to 17% in 2026, and the combined Advanced and Leaders share rose from 27% to 52%, per the report page. Leaders are defined as companies with a CEO-level mandate that connect AI adoption to how work, technology, governance and financial accountability operate together.
Agentic operations, in Genpact's usage, brings people, AI agents, processes, data, governance and value capture together in one operating model. Keep that definition in mind, because it is also the name of the product line Genpact is selling.
Agent Fleets Grew While Most Headcounts Held Flat
Among respondents with agents in live production, agent populations moved sharply and human headcount barely moved. The split, from the report:
| Change over the past year | Agent population | Human FTE count |
|---|---|---|
| Increased | 69% | 31% |
| Roughly flat | 27% | 61% |
| Decreased | 4% | 8% |
The 92% figure that leads the press release is the sum of the bottom two human rows. With the table beside it, the finding is narrower than it sounds: in most companies running agents, the agent count is the variable that changed, and human headcount was held where it was.
Flat headcount alongside a growing agent fleet means one of three things is happening inside those companies. The agents are absorbing growth that would otherwise have needed hiring. People are being moved to other work. Or the agents are not yet doing enough to change staffing at all, which fits the 49% average delivery rate on funded initiatives. The survey cannot tell you which, and your own data can.
IT Brief Asia's write-up of the report adds that 85% of companies plan to increase AI spending. Spending is rising on initiatives that, on average, deliver about half the time, while the workforce stays the same size. Expect that gap to come up when your CFO reviews 2027 AI budgets.
Where Do the Leaders Differ From Everyone Else?
The widest gaps between Leaders and other enterprises are organisational. Genpact's report gives these Leader-versus-other splits:
- 51% of Leaders create roles specifically for human-agent operations, against 22% of other enterprises.
- 49% of Leaders fully integrate AI value metrics with specific P&L lines, against 7% of others. Across the whole sample the figure is 15%.
- 51% pursue proactive, enterprise-wide operating model transformation for agentic AI, against 17%.
- 87% give agents timely read and write access across the systems they need, against 54%.
- 47% embed governance across the full agent lifecycle, against 11%.
The press release names two of those new roles: exception specialists and agent trainers. An exception specialist handles the cases an agent routes out, the ones that fail a confidence threshold or a policy check. An agent trainer maintains the instructions, examples and evaluation sets an agent runs on. Both sit between an agent's error rate and your customer.
The P&L gap is the bigger one. Seven percent of non-Leaders tie AI value to a specific line. In the other 93%, nobody's bonus moves when the agent saves money or loses it. A finance team cannot plan headcount against a benefit that no one has booked.
Sanjeev Vohra, Genpact's chief technology and innovation officer, summarised the report in the release: "Enterprises have become effective at deploying AI agents. Running them is the hard part." The two Leader gaps above are what "running them" means on an org chart.
Why Flat Headcount Can Still Hide Workforce Cuts
A stable total FTE count can conceal real change in who gets hired, and the strongest independent data says that is where the adjustment is happening. Stanford's Digital Economy Lab, using ADP payroll records for millions of U.S. workers through June 2026, finds no evidence of widespread, economy-wide job displacement. It also finds that employment of workers aged 22 to 25 in AI-exposed occupations now stands 19% below where it would be had it kept pace with less-exposed peers, and that the gap operates mainly through reduced hiring of young workers rather than increased separations. The authors call these early descriptive indicators, not causal estimates.
Put that next to Genpact's table. A company can report flat headcount to a survey while it stops backfilling entry-level roles and lets attrition do the work. Executives answering a workforce question will count bodies, and a hiring freeze leaves the body count unchanged for a year.
The visible exceptions are instructive in both directions. Salesforce chief executive Marc Benioff said in an interview reported in September 2025 that the company had cut support headcount from "9,000 to about 5,000" with AI handling about half of customer interactions. Klarna went the other way: after claiming in February 2024 that its assistant did the work of 700 agents, its chief executive told Bloomberg in May 2025 that the company was hiring human support staff again because the cheaper output was lower quality. Our coverage of PayPal's agents resolving 24 million contacts and Apple pausing 5,000 AppleCare layoffs tracks how those staffing decisions were tied to agent performance.
Genpact's 8% who reduced headcount look like Salesforce. For the 61% who held flat, the question to ask is what happened to hiring.
Read It Knowing Genpact Sells the Answer
Genpact's commercial interest lines up exactly with the report's prescription, and that should change how much weight you give the Leader profile. The company's Q3 2026 results, filed with the SEC, show total revenue of $1.385 billion, up 7.3%, with Advanced Technology Solutions up 30.3% to $405 million while Core Business Services was flat at $980 million. Chief executive Balkrishan Kalra called the quarter "another clear proof point that our pivot to Agentic Operations is working." On the earnings call he said Genpact expects agentic solutions bookings to exceed $1 billion in 2026, more than five times the 2025 level.
So a services firm whose larger segment was flat has published research concluding that companies should run people and agents in one operating model, with new roles and redesigned operations. The data may well hold up, and three parts of it still deserve a discount.
First, the maturity index is Genpact's own construct, and the Leader traits are also the components of its offering. Correlation between "has new roles" and "is a Leader" is partly definitional. Second, the data is self-reported by executives, who have every reason to describe their agent programmes favourably. Third, a steady-headcount finding is the easier message for a firm selling agent programmes to clients who worry about labour disruption.
Independent forecasts are harsher. Gartner predicted in June 2025 that over 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. Genpact's own 49% average delivery rate does not contradict that. Clear role design and a named P&L owner speak directly to two of Gartner's three reasons, cost and unclear value, and you can build both without buying them.
What to Do Before 2027 Budgets Lock
The work is to replace an assumed headcount saving with named roles, named owners and a measured baseline. Here is the order to do it in.
This Week:
- Pull agent count and human FTE change for every function running agents in production, over the last twelve months. Build your own version of Genpact's table. If you cannot count your agents, start there; our piece on Anthropic monitoring 30,000 agents covers the oversight metrics to track.
- Pull entry-level requisitions opened and filled in the same functions over the same period. Flat headcount with closed junior requisitions is a workforce decision someone made without writing it down.
This Month:
- For each production agent, name the person who owns its P&L line. If the honest answer is "the AI team", the benefit is not booked anywhere, and your CFO will treat it as unproven.
- Write job descriptions for exception handling and agent training in the two functions with the most agent volume, and price them. Our coverage of bank listings for agent orchestration and Goldman's new hires managing agents shows what these roles are paying and where they sit.
- Check your review gates. Humans decide or audit at least 87% of the time at every stage, per Genpact. If yours approve nearly everything, read our human-in-the-loop buyer's guide before you count those reviewers as the control.
Before Budget Sign-Off:
- Remove any line in the 2027 plan that books an agent-driven headcount reduction without a measured baseline and a named owner. Replace it with a funded role line and a review date.
- If a services vendor is pitching an agentic operations engagement, ask for the roles they would create on your side, the P&L line they would tie value to, and what happens to their fee if the 49% delivery average applies to you.
The Bottom Line
Agent programmes are in the same place the last wave of workflow automation reached: the technology is deployed, the fleet is growing, and only 15% of companies in Genpact's sample have tied its value to a P&L line. The workforce effect so far shows up in hiring and in role mix, where the Stanford payroll data finds it, more than in total headcount. ServiceNow's disclosure that AI shrank sales ops while quota sellers grew shows what a deliberate version looks like: a shift in which roles exist, owned by someone, with a number attached.
Name the owner and fund the exception desk before you write down the saving.
Continue Reading
- ServiceNow Says AI Cut Sales Ops as Quota Sellers Grew Double Digits
- Goldman Says New Hires Manage AI Agents Before Learning the Job
- Bank AI-Related Listings Rise 49% as Agent Orchestration Grows
- Apple Pauses 5,000 AppleCare Layoffs Tied to Its AI Support Agent
- AI Coding Agents Lift Pull Requests 35.7%, Harvard Study Finds
- Gainsight Cuts About 270 Jobs as CS Pricing Moves Off Seats