AI pricing shifts to 'priced work'

AI vendors are starting to charge for jobs completed, not just employees logged in. That is pushing software pricing closer to labor pricing. (businessinsider.com) Goldman Sachs said in a March 10, 2026 report that agentic AI tools are forcing software companies to rethink the seat-based model that dominated software for more than a decade. The firm said vendors are increasingly selling “units of labor” or measurable output instead of access for named users. (goldmansachs.com) That shift is already visible in product pages. Zendesk says its AI agent usage is billed in “automated resolutions,” replacing its older monthly-active-user model for bots, and Intercom says Fin charges $0.99 per conversation outcome when the bot resolves an issue or completes a procedure handoff. (support.zendesk.com) (fin.ai) The basic idea is simple: if software now drafts contracts, answers support chats, or reviews documents on its own, vendors can argue they should be paid per task finished. Goldman Sachs said that reframes the sales pitch from software budgets tied to headcount to budgets tied to output and productivity. (businessinsider.com) (goldmansachs.com) That matters in 2026 because AI tools are moving from copilots, which assist workers, to agents, which complete sequences of actions with limited human input. Goldman Sachs wrote in July 2025 that AI agents could unlock business productivity by performing multi-step work, especially in customer service and software. (goldmansachs.com) The pricing change also lines up with how cloud and model costs are incurred. Google Cloud’s Document AI, for example, charges by pages processed, with list prices such as $1.50 per 1,000 pages for Enterprise Document OCR and $30 per 1,000 pages for Form Parser at lower volume tiers. (cloud.google.com) Salesforce is offering the same mix. Its Agentforce pricing page says customers can buy AI “digital labor” through consumption-based pricing with Flex Credits or Conversations, alongside per-user licensing for other use cases. (salesforce.com) Support software has become the clearest early test case because the output is easy to count. Klarna said in February 2024 that its OpenAI-powered assistant handled 2.3 million conversations in its first month, did work equivalent to 700 full-time agents, and cut average resolution time to under 2 minutes from 11 minutes. (klarna.com) The old seat model has not disappeared. Zendesk still sells core plans from $19 to $169 per agent per month billed annually, and Intercom still lists agent-based subscriptions starting at $29 per seat per month, but both now layer outcome pricing on top. (zendesk.com) (intercom.com) Investors are watching because pricing determines how much of AI’s productivity gains software companies can keep. Goldman Sachs said fears that AI could compress software revenue are tied not just to better automation, but to whether vendors can convert that automation into new billing models before customers demand lower prices. (goldmansachs.com) The next fight is likely to be over measurement. Charging per seat is easy to audit; charging per resolution, conversation, page, or action depends on whose definition of “done” a customer accepts. (support.zendesk.com)