RC
A prospecting agent finds the account, drafts the first note, books the meeting. An AE runs the call and closes it four months later. At commission time, three questions land on RevOps at once: Is this sourced or inbound? Does the SDR team get the meeting credit? And what happens to the quota that was set assuming humans did the sourcing?
Comp plans encode a company's theory of who creates revenue. That theory is now wrong in a lot of places, quietly. Salesforce's State of Sales (February 2026) found 54% of sellers have already used AI agents, and nearly nine in ten plan to by 2027. Most comp plans were written before any of that was true.
Credit was always a proxy for work
Comp plans rarely pay for revenue alone. They pay for the work that produced it, routed through proxies: meeting set, opportunity created, stage advanced, deal closed. Sourced versus inbound exists because someone decided outbound effort deserved a higher rate. SDR meeting credit exists because setting a meeting was hard, measurable work.
An agent collapses those proxies. The meeting still gets set. The opportunity still gets created. But the work the proxy was measuring was done by software, grounded in data the company already owned, and the rate attached to it was priced for a person.
This is not an argument for paying people less. It is an argument that the proxies have come loose from the work, and a plan built on loose proxies produces disputes, not motivation.
Four places the plan breaks first
Not everything breaks at once. In our conversations with GTM leaders, the pressure shows up in the same four places.
Sourced versus inbound. If an agent found the account from a signal and a rep sent the agent's note after a light edit, the deal does not fit either bucket. Most plans default it to sourced, which pays the outbound rate for inbound effort.
SDR meeting credit. When an agent books a meeting, who was the SDR? If the answer is "the person who approved the agent's draft," you have turned a prospecting role into a review role without repricing it.
Stage advancement and splits. Agents that run follow-up, update the CRM, and surface next steps advance deals. Nobody writes a split for software, so the credit flows to whoever owns the record, which is often an accident of routing.
Quota baselines. Quotas assume a certain capacity. Salesforce's same report says top-performing teams are 1.7x more likely to use agents for prospecting. If capacity rose because of agents and quota did not move, the plan pays for leverage it did not price. If quota moved and the agents underdeliver, the plan punishes reps for a tooling decision they did not make.
Each of these is fixable. What makes them hard is that the plan is a legal and emotional document, and reps read every change as a pay cut until proven otherwise.
The real problem is attribution you cannot defend
Comp disputes are attribution disputes with money attached. A rep contests credit when they believe the record is wrong about what happened. An agent makes the record harder to defend in two ways.
First, an agent's work is often invisible in the CRM. It enriched the account, scored it, drafted the note. None of that is a stage change, so the CRM shows a rep-created opportunity and nothing before it. The agent's contribution is undercounted, and the first human touch gets full credit.
Second, where agents do write to the CRM, they can overcount. A follow-up agent that logs an activity every time it nudges a contact makes the opportunity look worked when it was not. If any plan element rewards activity, the agent inflates it.
A plan can survive either error if the record is clear about which actions were taken by a person, which by an agent, and which by a person approving an agent. Most records are not. We wrote about this from the pipeline side in Your AI Dashboard Can't Tell You If a Deal Moved. Comp is where the same gap becomes a payroll question.
What RevOps can decide before the next plan year
You do not need a new philosophy of selling. You need a few decisions written down before Q1 plans go out.
Make agent actions visible and labelled. Every write an agent makes to the CRM should carry an agent identity, not a shared service account or the approving rep's name. If you cannot tell agent work from human work in the record, no comp rule can.
Decide what agent-sourced means. Pick one definition and apply it: agent found the account and a human approved the first touch is agent-sourced. Then set a rate for it. It can equal the sourced rate for now. The point is that the category exists.
Reprice the review role on purpose. If SDRs are approving agent drafts, say so in the role description and the plan. Pay for conversion quality and judgment, not meetings set, and retire activity metrics the agent can inflate.
Reset quota with the leverage in view. When agent capacity enters a territory, document the assumed uplift in the quota memo. Reps will accept a higher number they can see the reasoning for far more readily than a surprise.
Keep a human on the credit decision. Agents can propose attribution. A person should approve it, and the approval should be auditable. This is the same control you want on any agent write that moves money.
None of this is the plan for 2030. It is the minimum that keeps the current plan defensible while the ratio of agent work to human work keeps shifting.
Where the layer underneath comes in
Every decision above depends on one thing the comp team cannot build alone: a record of what every agent did, under which identity, with which approval, tied to which deal stage. That is not a comp system feature. It is governance and attribution working together underneath the agents.
That is the layer wysdym is built to be. Every agent action runs through per-agent identity and approval rules, and outcomes are attributed back to deal stages rather than to activity counts, so credit questions can be answered from the record rather than argued from memory. If your comp committee is already fielding the sourced-versus-inbound question, see how it works or talk to the founders.