Sierra Co-Founder Says AI Agents Could Blow Up Software's Subscription Model

The subscription model that built the modern software industry may not survive contact with AI agents that actually do the work instead of just answering questions.

That's the argument from Clay Bavor, co-founder of AI customer service startup Sierra, in a conversation with CNBC's Arjun Kharpal on the network's "The Tech Download" podcast. Bavor says AI agents are moving out of demo mode and into live business workflows, particularly in customer service, sales and support. The distinction he draws is simple: a chatbot answers a question, an agent finishes a task.

That distinction matters for money. Software has been sold for decades on a per-seat, per-month basis. Companies pay for licenses whether an employee uses the tool once a day or fifty times. Bavor's pitch, according to CNBC, is that agent-based AI opens the door to outcome-based pricing, where companies pay based on results delivered rather than access granted.

Sierra builds AI agents that handle customer service interactions for companies, and Bavor told CNBC the company tests those agents extensively before they go live with real customers. He said businesses are pushing for clearer ways to measure AI's actual return on investment, which is the practical reason outcome-based pricing has gained traction as a talking point in enterprise tech circles. If a company can't tell whether an AI tool is saving money or just adding a line item, the sales pitch gets harder.

Bavor runs a company whose business model depends on convincing enterprise buyers that agentic AI delivers measurable outcomes worth paying a premium for. An outcome-based pricing structure benefits AI vendors who are confident their tools work and want to capture more of the value they create, rather than being capped at whatever a traditional software license would charge. Readers should know a founder pitching outcome-based pricing has skin in the game.

The broader question of whether agentic AI actually reshapes software revenue is still unresolved. A shift like this would ripple through publicly traded software companies that built their valuations on recurring subscription revenue, from Salesforce to ServiceNow to smaller SaaS players. If enterprise buyers start demanding to pay only for completed tasks or resolved tickets, software companies with thin margins on labor-intensive support products could see real pressure on pricing power.

It's also one that's been discussed in tech circles for at least a year without a clean, widely adopted pricing standard emerging. Bavor told CNBC the hardest part of enterprise AI deployment may not be building the agent at all, but what he called the "last mile": getting a working AI system reliably integrated into a company's actual operations, handling edge cases, escalations and the messy reality of live customer interactions rather than a clean demo environment.

The last-mile problem is the practical reason a lot of pricing predictions haven't yet turned into contracts. An agent that handles 80% of customer inquiries correctly still needs humans to catch the other 20%, and figuring out how to price that hybrid reality is harder than a stage pitch suggests.

Bavor also flagged rising AI token costs, according to CNBC, a real operating expense for any company building on top of large language models from providers like OpenAI or Anthropic. As agents get more capable and handle longer, more complex tasks, they consume more compute, and that cost has to land somewhere in a pricing model, whether it's baked into a subscription fee or passed through in an outcome-based rate.

There's no data yet showing outcome-based pricing has become the industry norm, and most enterprise software contracts today are still negotiated on traditional per-seat or per-usage terms. What exists right now is a founder with a direct financial interest in the outcome, making a case that's plausible on the merits but unproven at scale. The next concrete signal to watch is whether major enterprise software vendors, not just AI-native startups like Sierra, start restructuring contracts around measurable task completion rather than access. So far, that shift remains a talking point, not a trend line.