For thirty years, buying enterprise software meant answering one question: how many people need a login? Salesforce, Workday, ServiceNow, SAP — the entire architecture of modern IT spending was built on the seat. Count your headcount, multiply by the per-user price, sign the renewal. CFOs budgeted for it, procurement teams negotiated discounts against it, and vendors grew by selling more logins into bigger organisations.
That arithmetic is coming apart, and not for the reason most executives assume. It isn’t that AI is making software cheaper. It’s that AI agents are making the seat itself the wrong unit of account — and the industry that spent three decades perfecting seat-based pricing is now scrambling to reprice itself around something much harder to define: outcomes.
What actually broke
The mechanism is simple enough. When a human sales rep works a lead, she needs a CRM seat. When an AI agent qualifies that same lead, updates the record, drafts the follow-up and escalates only the deals that need a person, nobody needs a login for most of that workflow. Salesforce’s own numbers make the shift concrete: in its most recent quarter, the company reported 3.2 billion “Agentforce” work units processed, up 97% quarter-on-quarter, alongside roughly $1.5 billion in Agentforce annualised revenue — activity that has almost nothing to do with counting human users. Marc Benioff has been unusually candid about why: “We’re trapped in old per-user pricing models,” he told investors, arguing the company needs pricing “enormous” enough to capture value that no longer flows through a seat at all.
Salesforce isn’t alone. Workday, ServiceNow and SAP are all migrating, in varying degrees, toward consumption, transaction and outcome-based tiers — a customer might now pay per resolved support ticket, per completed workflow, or per unit of “commercial activity” an agent generates, rather than per named user. One case cited in recent industry coverage involved a $40 million enterprise contract structured around an activation agent handling 1,500 interactions a day — priced not by who could log in, but by what got done.
For a CEO or CFO, this sounds like good news: pay for value, not access. It isn’t that simple, and the fine print is where the real strategic risk lives.
Who defines the outcome, owns the leverage
Outcome-based pricing sounds like it shifts risk onto the vendor — you only pay if it works. In practice, it shifts power to whoever gets to define, measure and attribute the outcome, and that is rarely the customer. What counts as a “resolved” ticket? What’s the baseline against which improvement is measured? What happens when the agent partially solves a problem a human then finishes? Enterprise buyers who spent decades getting good at negotiating discounts off list price are now being asked to negotiate the definition of value itself — a far harder skill, and one most procurement functions haven’t built yet. Vendors that control the metrics can quietly capture more economic value than they ever did under per-seat pricing, not less.
This is the first assumption CEOs should discard: that the move away from seats is automatically a cost-saving one. It’s a renegotiation of where value gets captured across the software stack, and right now the vendors are setting the terms.
The cautionary tale conventional wisdom keeps skipping
The clean version of this story — agents replace people, costs collapse, done — runs into an inconvenient case study. In 2024, Klarna announced its AI assistant was doing the work of 700 customer service agents, handling 2.3 million conversations in its first month. It was held up globally as proof that generative AI could simply substitute for headcount. By 2025, Klarna was reversing course, rehiring humans for the roles it had automated away. CEO Sebastian Siemiatkowski’s own explanation was notably direct: the company had “underestimated the tradeoff” between cost and quality, and concluded that “investing in the quality of human support is the way of the future.” Klarna’s AI still handles roughly two-thirds of inquiries — this isn’t a retreat from automation — but the pure substitution narrative didn’t survive contact with customers who, in independent surveys, rated empathy above speed by a wide margin.
The lesson isn’t that agentic AI doesn’t work. It’s that the industry keeps discovering, case by case, where the quality floor sits — and that floor moves the economics back toward hybrid models faster than the breathless deployment announcements suggest. Any executive treating “we deployed agents” as a strategy rather than the start of a longer calibration process is reading the story a year behind where it actually is.
Where this lands hardest: India’s services economy
If seat-based software pricing is unwinding, the labour-hour pricing model underneath the $315 billion Indian IT services industry is exposed to the identical logic, and arguably more acutely. TCS, Infosys, HCLTech and their peers built their scale on billing effort — engineers, hours, delivery teams — for work that agentic coding and operations tools now perform in a fraction of the time. TCS’s decision in 2025 to cut roughly 12,200 roles, concentrated in middle and senior management, was explained by leadership partly through AI-driven efficiency gains reshaping how work gets delivered. Nasscom’s own FY26 review makes the structural shift visible in aggregate: industry revenue is projected to reach $315 billion, growing 6.1%, while net headcount additions have slowed to a crawl relative to that growth — what the report itself describes as a “decoupling of revenue expansion from hiring.”
The companies best positioned aren’t necessarily the ones cutting fastest; they’re the ones already selling AI-augmented outcomes rather than augmented headcount. TCS now discloses roughly $1.8 billion in annualised AI-linked revenue; Infosys and HCLTech report AI contributing single-digit percentages of topline and growing. That’s the real race — not who adopts agents internally first, but who moves from billing time to billing outcomes before their clients force the renegotiation on worse terms. Indian IT leadership has navigated model transitions before, from onshore bodies to offshore delivery to cloud-native platforms; this transition is faster and structurally deeper, because it attacks the billing unit itself rather than the delivery geography.
What this means at the top of the organisation
The practical implication isn’t “adopt AI faster.” It’s that procurement, legal and finance functions need to develop real fluency in outcome-contract mechanics — baseline definitions, attribution methodology, audit rights — with the same seriousness they once applied to seat-count audits. Boards evaluating vendor relationships, and Indian IT clients evaluating their own service contracts, should treat every renewal as a chance to ask who actually defines success, not just what the headline price looks like.
The more durable question for any CEO watching this unfold isn’t how many jobs agentic AI will eventually take. It’s who ends up owning the definition of value once software and services both stop charging for access and start charging for results — because whoever writes that definition, on either side of the negotiating table, is about to hold most of the leverage in enterprise technology for the next decade.
