When Marc Benioff announced last week that all of Salesforce and Slack are now exposed to AI agents, meaning you no longer need to buy a seat to access either system, he was not changing a pricing model. Benioff was admitting that the seat-based system of record is no longer where revenue lives. The CRM is becoming a library card. Agents are the workforce. Customers are about to figure out the difference.

The signals are already in the public market. SaaS stocks are down 20% in 2026. Public SaaS companies trade at 4.1x revenue, the lowest multiple of the past decade. None of this is due to a temporary economic correction. It happens because the underlying business model of a per-seat license priced against human labor that sat in a UI, typing in what had already happened, is being structurally repriced.

For most of the last two decades, the CRM has been the system of record for revenue. Sales reps logged calls, updated stages, attached notes, and the company built dashboards on top of that activity. The CRM was the artifact. It was never the thing producing revenue. CRMs were the place where someone documented revenue activity after the fact. That was fine when the marginal cost of producing the documentation was low compared to the value of the deal.

It’s no longer fine. McKinsey's data on seller time has been consistent for a decade: reps spend less than 30% of their time with customers. Skaled put non-selling time at 72%. The CRM is not the cause of that, but it is the visible evidence of a workflow architecture that requires humans to act as data janitors. When agents replace that workflow architecture, the system of record becomes either a passive sink for what the agents produce or entirely irrelevant.

The value of a revenue platform now sits in the data layer, not the UI. An "AI CRM" that bolts an agent onto the same schema, the same activity log, and the same pipeline view that worked for a sales rep in 2014 is not a different category. It is in the same category as a chatbot. The actual leverage is in whether the data is structured so that an agent can reason across the full revenue lifecycle, pipeline, expansion, retention, churn, and whether it is normalized, time-aligned, and governed. Most CRM data is none of those things. The Clari and Salesloft research published before their merger showed that 48% of revenue data is "not AI-ready." That number, from the incumbents themselves, is the most important admission in this market.

The business model breaks. A per-seat license is priced on the assumption that humans are the bottleneck in the revenue process. If agents do most of the work, research, outreach, follow-up, expansion identification, churn prediction, the per-seat model is charging for a unit of work that is no longer the constraint. Salesforce going headless is the first major incumbent to acknowledge this in pricing. Others will follow. The platforms that win on the other side will not be the ones that figure out how to keep charging $200 per seat. They will be the ones who figure out how to counteract the outcome the agents produce.

The strategic asset is the model trained on revenue data, not the application that displays it. There is a reason we are building Gravity on Databricks rather than on top of an existing CRM. The data is the platform. We have built a Signal Data Platform that ingests the entire revenue history of a company, pipeline, expansion, customer health, product usage, billing, and churn into a normalized, governed environment, and we have built a large tabular model purpose-built for that data. The agents are the workforce that operates on top of it. The display layer matters, but it is the least durable part of the system. The data layer creates a defensible position the moment the market accepts that the system of record was always the wrong artifact.

I have spent the last several months in conversations with revenue and customer success leaders at enterprise SaaS companies. The message I hear is consistent. They have spent the last eighteen months buying point solutions for AI in their GTM stack,  an AI SDR, an AI conversation intelligence tool, an AI summarizer, and the revenue numbers have not moved. An MIT study that found 95% of companies that deployed AI saw zero meaningful revenue growth is not a methodological quirk. It is the predictable result of deploying agents on top of a data layer designed for human input.

For builders, the takeaway is that the next durable category is "which AI feature should I add to my CRM?" It is "which company is going to own the revenue data layer for my business in five years, and am I making the revenue software, the data layer, and the model, not the UI. For buyers, the question is not whether to decide now or let it default."

The seat-based CRM is dead. It will take a few more quarters to be obvious in the income statements. The buyers and the operators already know. If you are running revenue at a B2B company and you want to talk about what the agent-first stack actually looks like, reach out.