AI was intended to simplify the economics of business technology. Instead, it has reopened one of the oldest and most fraught questions in enterprise IT: how to pay for productivity without losing control of the bill. Enter the serious discussion around enterprise AI pricing.
After months of experimenting with usage-based and per-conversation charging for AI agents, Salesforce is edging back toward a more familiar model of seat-based licensing. The transition is nuanced, wrapped in credits, caps, and “fair use” language, but its significance is unmistakable. Tech buyers want innovation, yes, but not at the cost of financial unpredictability.
AI agents may well define the next epoch of enterprise software. What they will not do, at least for now, is convince shrewd CIOs and CFOs to sign open-ended checks.
- Salesforce Relaunches Slack as ‘The New Agentic OS For Work’: Key Takeaways
- Why ‘Employee Connection’ Will Be 2026’s Most Important Enterprise KPI
A $450 Billion Market Still Searching for a Business Model
The commercial prize is enormous. Gartner forecasts that agentic AI will account for 30 percent of enterprise application software revenue by 2035, surpassing $450 billion, up from just two percent in 2025. That expectation underpins the strategies of nearly every major software provider, from SAP and Oracle to Workday and Salesforce itself.
Salesforce Chief Executive Marc Benioff has been explicit about the opportunity:
“We’re talking about 3x, 4x the ability to multiply the monetization on customers because, by the way, they’re getting 3 or 4x or 10x more value from our products.”
Yet embedded in that logic is a paradox. If AI delivers the productivity gains vendors promise, and naturally, that’s a big if at this stage, enterprises should eventually need fewer people and fewer licenses. That tension sits uneasily with per-user economics that have defined enterprise software for decades.
Why Enterprises Rejected Usage-Based AI Pricing
Salesforce’s initial approach to Agentforce leaned toward a consumption-based pricing model: per conversation, per transaction, or per action. In theory, this aligned cost with value. In practice, it conflicted with organizational reality.
“When we first started with Agentforce, we were talking about so much per conversation(…), but customers have pushed for more flexibility,” Benioff admitted during Salesforce’s most recent earnings call.
For tech buyers, the problem is operational rather than philosophical. AI usage is unpredictable. Adoption varies wildly across teams. A handful of power users can generate disproportionate costs. For CIOs and CFOs, that makes forecasting and defending AI spend deeply uncomfortable.
Seat-based pricing, however imperfect, restores a degree of certainty that enterprises value more than theoretical efficiency.
Seat-Based Pricing, Rewritten for the AI Era
This form of enterprise AI pricing is not a return to unlimited usage. Today’s “seat-based” AI licenses almost always carry a second meter: credits, AI units, or fair-use thresholds.
For businesses, this hybrid model offers a transitional compromise. IT leaders can explore AI use cases without exposing the organization to runaway costs. Vendors, meanwhile, retain protection against spiraling compute expenses. Control, however, remains asymmetrical and largely remains in the hands of vendors.
AI Is Not Replacing Workers, and That Changes Everything
One assumption underlying early AI pricing strategies was the potential for large-scale workforce reduction. That assumption, so far at least, is proving flawed.
Forrester reports that 55 percent of companies that laid off workers due to AI later regretted the decision. Meanwhile, 57 percent of leaders expect AI investment to increase headcount, compared with just 15 percent who expect it to decrease.
Salesforce executives acknowledge the reality that enterprises often pay for both human labor and AI assistance simultaneously. That dynamic explains why seat counts have not collapsed, and why vendors remain comfortable with per-user pricing, at least for the time being.
The Enterprise Buyer’s Risk: Paying for Capability, Not Outcomes, and Canvasing Where AI Pricing Might Be Heading Next
Predictability comes with a trade-off. Seat-based AI pricing can mask underutilization.
This exposes a more profound truth. AI value does not emerge automatically. It requires process redesign, effective change management, robust governance, and clear accountability. Without that work, AI agents risk becoming the most expensive shelfware enterprises have ever purchased.
Over time, pricing models are likely to become increasingly fragmented. Copilots may remain seat-based, while workflow automation agents migrate toward usage- or outcome-based pricing.
Such models demand transparency, trust, and rigorous measurement, qualities that business software contracts have not traditionally rewarded.
Final Takeaway: Predictability May Be AI’s Most Valuable Feature
AI agents promise intelligence, speed, and scale. But for tech buyers ahead of 2026, their most compelling attribute may be far simpler: financial certainty.
Salesforce’s recalibration signals a recognition that enterprises will not industrialize AI without guardrails in place. The technology may be transformative in the future, but only if buyers can still understand the invoice.
