Enterprise software has been priced for a world where humans were the bottleneck. Seats worked because headcount was a reliable proxy for work: hire more support reps, buy more Zendesk licenses; expand sales, add Salesforce users; scale finance, add ERP access. The model was simple, predictable, and incredibly profitable.
AI agents change the underlying unit of production. When software can do more work with fewer humans, the customer’s productivity gain becomes the vendor’s revenue leak … at least under per-seat pricing.
Seats were never the value
Gokul Rajaram’s core point is not “add AI features”. It’s that AI forces software companies to rethink what they charge for. The most exposed vendors are those whose pricing is tightly coupled to the number of people performing tasks inside the product. Customer support is the cleanest example. If a company can reduce its human agent team because AI resolves more tickets, the “right” outcome for the customer is fewer seats for the vendor.
That’s why Rajaram flags a scenario like Zendesk. A buyer doesn’t need to abandon the platform to cut spend. They can simply wrap it, keep Zendesk as the system, but let AI agents absorb volume. Over time, the customer might go from 50 seats to 20, not because Zendesk is worse, but because the workload has shifted from people to machines.
Outcome pricing is the logical end state … and a painful transition
If seats stop working, vendors have to charge for something else. The most intuitive replacement is outcomes: tickets resolved, issues deflected, invoices processed, leads qualified, claims adjudicated. This is a redefinition of the commercial relationship.
Outcome pricing forces hard operational questions vendors could previously ignore: What counts as “resolved”? What’s the quality threshold? Who owns risk when the AI is wrong? How do you prevent gaming? And, critically, how do you manage the shift from $30 per seat per month to a few cents or a few dollars per completed action?
The short-term problem is financial. Even if outcomes-based pricing is more aligned with value, the transition can compress revenue and make it less predictable. Public markets tend to punish “messy middle” transformations. The provocative implication is that some companies may need the freedom of going private for a period, one or two years, so they can rebuild product, contracts, and pricing without being measured quarter-to-quarter on a model that’s becoming obsolete. It will be painful for many companies to rebuilt an AI product from an a SaaS business.
Coming back to Zendesk, the company is a useful real-world illustration precisely because it’s already moved. It has pushed AI heavily and begun introducing pricing that charges when an AI agent actually resolves an issue, shifting the billing unit closer to customer value rather than human usage.
The competitive moat shifts from UI to control points
AI agents also change where power sits in the software stack. Historically, a great UI and user adoption created leverage: more end users meant more licenses and more lock-in. But AI doesn’t care about dashboards. It cares about access: APIs, permissions, workflow execution rights, audit trails, compliance boundaries, and clean data.
That shifts the advantage toward “systems of record” (ie. products that hold durable, high-stakes data and sit at the center of business processes). Salesforce’s CRM database, SAP/Oracle ERP, core finance systems, and clinical record platforms are harder to displace because they’re embedded in governance, regulation, and institutional memory. AI can automate around them, but it still needs to read and write into them.
For workflow tools that sit on top (e.g. support desks, scheduling, lightweight collaboration, certain analytics layers) the threat is different. They can be hollowed out without being replaced. Customers keep the tool, but use fewer seats, fewer features, and rely more on automation outside the product. The vendor’s churn stays low while revenue retention deteriorates.
That’s why “API strategy” becomes more than developer relations. It becomes border control. If an external agent can extract value while commoditizing your UI, incumbents will try to capture the agent layer (by building their own) or tax the agent layer (through usage pricing, API fees, or bundled AI tiers).
What management teams should do now?
-
Stress-test your exposure to seat compression – if customer value increases when human usage decreases, your pricing is misaligned with your product’s future.
-
Experiment with outcome units early – start with one measurable outcome (e.g., ticket resolution, case deflection, lead qualification) and build billing, reporting, and customer trust around it.
-
Prepare for a revenue narrative shift – outcome/usage models can be better businesses but uglier in transition; management credibility depends on explaining new metrics clearly.
-
Invest in control points – durable data, permissions, workflows, compliance, and execution rights matter more when agents are the primary users.
The biggest mistake is to treat AI as a feature roadmap item. AI is a business model event. The question is no longer “How do we add AI?” It’s “What are we selling when humans are no longer the unit of work and how do we price it before someone else does?”. It will take 1-2 years for companies to do it. It will be hard, painful, but also the only way forward.
[NB] Last week I wrote a (paywalled) article on “Is AI just the death of software?” – I recommend subscribing and reading it.
Selective bibliography:
[1] Andreessen Horowitz (a16z) (2024) ‘Death of a Salesforce’. Andreessen Horowitz. Available at: https://a16z.com/death-of-a-salesforce/ (Accessed: 5 February 2026).
[2] Eggemeier, T. (2025) ‘Zendesk’s new pricing for AI agents (charged only when issues are resolved)’. Zendesk (Press release / Blog). Available at: https://www.zendesk.com/blog/ (Accessed: 5 February 2026).
[3] Miller, R. (2025) ‘Investing in software when AI agents are coming — notes on Gokul Rajaram’. Compound with René. Available at: https://www.compoundwithrene.com/p/investing-in-software-when-ai-agents (Accessed: 5 February 2026).
[5] O’Shaughnessy, P. (Host) (2025) ‘Gokul Rajaram: Lessons from investing in 700 companies’. Invest Like the Best (Podcast episode; YouTube version). Available at: https://www.youtube.com/watch?v=JUsb1FYOstA (Accessed: 5 February 2026).
[6] Salesforce (2025) ‘Salesforce introduces Flex Credits for Agentforce / AI consumption-based pricing’ (Press release). Salesforce Newsroom. Available at: https://www.salesforce.com/news/ (Accessed: 5 February 2026).
[7] Frontier Enterprise (2025) ‘Zendesk CRO on AI, pricing, and the shift from seats to usage/outcomes’ (Interview). Frontier Enterprise. Available at: https://www.frontierenterprise.com/ (Accessed: 5 February 2026).
[8] Bloomberg (2025) ‘Private equity bets on AI: Zendesk’s go-private and AI growth’ (Syndicated via SWI swissinfo.ch). Bloomberg / SWI swissinfo.ch. Available at: https://www.swissinfo.ch/ (Accessed: 5 February 2026).
