For most of the software industry’s history, companies bought software and their employees used it to do their jobs. Software has always automated work, from calculating payroll to routing customer calls. What’s changing with AI agents is how much of the work some software companies can take responsibility for delivering themselves. In sales, customer service and customer success, that can include work previously assigned to employees or outside service providers. Customers now have reason to reconsider what they’re buying, how they measure its value and what they expect vendors to stand behind.
Salesforce offers a particularly useful example because few companies are more closely associated with the SaaS model. The company that once built its brand around “No Software” is now showing what happens when customers don’t necessarily need to interact with the software’s interface either. The company reported that annual recurring revenue for Agentforce, its platform for building and deploying AI agents, had exceeded $1.5 billion, up more than 240% year over year. (It’s worth remembering the company has alsorecently changed how it identifies and reportslines of revenue.) Salesforce also expanded its partnership with Anthropic so sellers can use Salesforce data and workflows through Claude without opening Salesforce itself.
Patrick Stokes, Salesforce’s president of applications and marketing, described Salesforce in Claude as a knowledge-worker agent, while Agentforce is designed for “autonomous work or work that touches the end customer.” That distinction has economic implications beyond Salesforce. Software that helps an employee do a job creates a different commercial relationship from a vendor offering to perform defined work on the customer’s behalf. What customers buy, how they measure its value and how vendors charge for it can change accordingly.
No Jitter has already examined how AI is changingconsumption and outcome-based pricing. The next question is what happens when vendors begin selling more of the work their technology performs.
From selling software to delivering work
Post-sale activities managed by customer success teams show another side of the same problem. Companies typically reserve dedicated customer success managers for larger or more valuable accounts because the economics don’t support giving every customer the same level of human attention. Smaller customers often receive pooled or digital support. AI could make more active engagement practical across accounts that haven’t traditionally received dedicated human coverage, including monitoring adoption, identifying risk and supporting renewals.
Forrester analyst Shari Srebnick has written that while CS leaders commonly ask what they can automate, more advanced organizations are starting to rethink how they’d design customer success with AI available from the start.
Gainsight, whose software helps companies manage customer adoption, retention and expansion, is taking one version of that idea further with Atlas, an AI-native service for long-tail renewals. CEO Chuck Ganapathi describes the opportunity as a “Goldilocks zone”: work that matters and is repetitive enough to systematize but has enough variation to require an agent rather than a script. Customers can build AI themselves, buy agents or effectively hire Gainsight to perform some of the work, and combine those approaches.
When software and services start to overlap
After hearing Ganapathi discuss Atlas at the recent AI-Native Services (AINS) Summit, Akash Bhatia, who leads Boston Consulting Group’s global technology sector, saw the potential for AI-native services to reshape other services businesses. These companies, he argued, need the domain expertise of a services business combined with technology that lets them deliver the work efficiently at scale.
Emergence Capital, the venture firm behind the event, argues that companies built this way can achieve gross margins above 50% while delivering five to ten times greater speed or throughput than traditional services businesses. As an investor in the category, the firm has an interest in that thesis, but the markets it identifies extend well beyond customer success, including insurance claims, compliance and supply chain operations.
The economics are central to the idea. Traditional services businesses typically require revenue to grow alongside headcount, limiting the margins and scalability associated with software. The premise behind AI-native services is that AI can change that relationship by allowing companies to deliver substantially more work without adding people at the same rate.
Getting paid for the work
Those different ways of delivering the work also create a pricing problem. Salesforce’s Claude integration makes the shift concrete: a seller can use Salesforce data and workflows without opening Salesforce, while Salesforce can still charge for that activity through consumption. The underlying data, permissions and workflows remain valuable even when the application itself isn’t the interface.
Salesforce also acquired m3ter this year, whose technology supports usage and outcome-based pricing. Consumption offers one way to charge when software rather than a person is using the application, but it still measures usage rather than value. Tokens, API calls and agent actions don’t tell a customer if useful work was completed, if it was done well or what it was worth.
That helps explain the interest in pricing tied to completed work or outcomes. Outcome-based pricing isn’t necessarily the successor to seat-based SaaS, and the idea isn’t new. But once software performs more of the work, customers have more reason to ask if some of what they pay should reflect the work completed, its quality or the result. Gartner expects outcome-based pricing to account for 40% of AI-agent vendor customer-service revenue by 2029. At the same time, Gartner points to the difficulty of defining and measuring an outcome. A resolved service request isn’t necessarily a satisfied customer, and a renewal can reflect years of product experience and human relationships.
Performing more of the work also doesn’t mean the vendor assumes responsibility for everything that follows. Customers still control the policies, data and permissions the system operates under. In a recent analyst briefing I was part of, discussion about AI in production quickly moved to cost, quality, accountability, knowledge maintenance, escalation and scale. Buyers need to know what the system will do, how they’ll know if it did the job well and what happens when it doesn’t.
For years, companies have bought software, hired employees and engaged service providers through largely separate decisions and budgets. AI gives them more ways to accomplish the same work. Increasingly, customers can buy the work itself, along with expectations for how well it will be performed, rather than simply access to the software used to do it. That changes what customers have reason to measure and what vendors need to stand behind.
