Co-Sell When Agents Source the Deal

On April 15, 2026, Salesforce committed $50 million in funding and technical assistance to help partners build AI agents and bring them to market, folding AppExchange, Slack, and AgentExchange into a single marketplace so a partner-built agent can sit on the same shelf as a partner-built app. Forward-deployed engineering teams went first to Accenture, Deloitte, PwC, Slalom, and IBM Consulting. What the announcement does not say is what happens to a deal once one of those agents surfaces it: who gets paid, and how much, when the source of a lead is a workflow a partner built and maintains rather than a person who worked the phones.

That question already existed before agents entered the picture. Two sellers touch the same deal, or a partner introduces an opportunity that an internal account executive later closes, and someone has to decide who gets the number. Agents change the timeline of that argument more than they change its substance. A disagreement that used to accumulate slowly across a quarter-long sales cycle now surfaces within hours, because an agent can identify, qualify, and hand off a lead in the time it takes a human rep to read the notification.

The dispute that already existed

Multi-seller crediting has been unsettled for years, long before anyone attached an agent to a pipeline. An Alexander Group survey, cited in WorldatWork's own coverage of the practice, found 58 percent of companies credit two or more sellers on the same sale, 42 percent credit only one, and 5 percent credit five or more, a tail the researchers themselves flagged as worth investigating on its own. Forrester's published guidance on the underlying question is a sequencing rule rather than a percentage: use a split when each seller's individual contribution to closing the deal can be clearly delineated and measured, and use a team quota when it cannot. Most comp teams already know this rule. Few apply it consistently, because measuring a "contribution" has always required judgment calls a spreadsheet can't fully resolve.

The channel runs the same argument

Partner-sourced revenue has its own version of the fight, and it is no more settled. PartnerStack and Wynter's "State of Partnerships in GTM 2026" report surveyed 100 senior revenue, marketing, and partnerships leaders at B2B SaaS companies with at least $50 million in revenue, and found 42 percent attribute partner revenue using multi-touch models, 31 percent use first-touch, and 19 percent use last-touch. A Brevet Group-commissioned Spiff survey of 93 sales professionals, taken in December 2020, found almost a third of companies either split credit informally with no written rule or have no policy on the question at all. That survey predates agentic AI by several years, which is exactly why it's useful here: it shows how unresolved multi-touch credit already was among humans, before an agent could source a deal in the time it takes to read an email.

What an agent adds to the argument

No published survey asks the exact question this episode is built around: when an AI agent surfaces a deal, does the person or team who built and operates that agent get commission alongside the account owner who closes it? The research above is the closest available evidence, applied to a new question rather than measured directly, and that distinction matters enough to say plainly before drawing any conclusion from it. What the proxy data does establish is the shape of the argument: a third party contributed something real to the deal's origin, an account owner did the work of closing it, and most organizations have never fully settled how to split credit between roles like that even when both are human.

Agent-sourced deals add a third role to a fight that used to have two sides. The account owner still closes. The partner or internal team that built the sourcing agent contributed the origination. And increasingly a workflow builder, the person who configured, trained, and maintains the agent doing the sourcing, sits behind that origination without a name most comp plans currently have a line for.

Where the proxy breaks down

An agent's touch is a timestamped, replayable log entry. That should, in principle, resolve exactly the kind of ambiguity that produces human crediting disputes, since there is no longer a dispute over whether the touch happened. It introduces a different problem instead. A logged touch is evidence that an agent acted; whether that action changed the buyer's decision is a separate question the log can't answer on its own. An agent can generate a technically accurate touch record, a qualifying question answered, a document sent, a meeting proposed, with no real influence on why the deal closed, and a comp plan that pays on the log rather than on a verified downstream outcome will reward that noise as readily as it rewards a genuine assist. There is also an incentive problem: paying commission on sourcing events invites partners to build agents optimized to generate touches rather than qualified opportunities, the same volume-over-quality failure mode that already shows up wherever a metric gets compensated directly.

The comp premium already sitting in the role

Before adding a new commission line for agent-sourced deals, it's worth checking whether the skill involved is already being paid for. BlueThread's own Partnership Roles in AI: 2026 Salary Intelligence Report, published in June 2026 and built from more than sixty compensation sources including Levels.fyi, Glassdoor, PayScale, and PwC's 2025 Global AI Jobs Barometer, found that partner-facing roles with strong AI fluency already command a cash premium of 48 to 55 percent over otherwise comparable roles without it. In dollar terms that runs from about $55,000 for a Partner Manager (base moving from roughly $115,000 to $170,000) up to about $122,000 for a VP of Partnerships (from roughly $248,000 to $370,000), with a Senior Partnerships Manager, a Director, and an AI Alliances Manager landing in between. The market has already repriced the ability to build and operate agentic workflows into base and target compensation, months before any comp plan formally addressed what happens when one of those workflows sources a deal.

That changes the shape of the question. If the skill of building a sourcing agent is already priced into someone's OTE, a separate deal-level commission for the same skill risks paying for it twice, unless the workflow-building work stays clearly separable from the deal-closing work it eventually produces.

The recommendation

Keep the account owner's normal closing credit intact. Build a separate, rules-based sourcing share for documented contributors, sized modestly and paid alongside closing credit rather than carved out of it. Pay that sourcing share to a workflow builder only when operating and improving sourcing agents is an explicit, compensated part of their role. That condition does most of the work: it keeps the sourcing share from becoming a stealth commission for a skill the salary data shows is already compensated through base pay, while still recognizing the cases where an agent's origination work has become a distinct, compensated job function in its own right.

The case against it, and where it holds

Three objections come up whenever this recommendation gets tested against a real comp plan, and each one holds a real mechanism worth answering rather than waving off.

Gaming is the first. A sourcing share that pays on a logged touch creates a direct incentive to build agents that generate high touch volume regardless of quality, since the credit attaches to the log entry rather than to a verified result. The fix is to gate the sourcing share behind a minimum downstream milestone: a lead has to convert to a qualified opportunity, beyond simply registering a contact, before any share is owed. That single gate removes most of the incentive to farm log entries, because volume alone stops earning anything.

Complexity is the second, and it's the more honest limit. Every split policy adds administrative overhead, and the WorldatWork coverage of the Alexander Group data notes this directly: credit disputes are a known cost of any multi-seller policy, and a new agent-sourcing line item is one more rule for a comp team to reconcile every cycle. The practical answer is to scope the sourcing share to accounts or deal sizes where the volume justifies the added complexity, and default to the account owner keeping full credit everywhere else, rather than rolling out a universal policy on day one.

Causation is the third, and the one worth taking most seriously. A logged agent touch is not proof the touch caused the outcome, which is the same limit flagged above. The honest answer here is Forrester's original sequencing rule, applied to a new kind of seller: pay a split only where the contribution can be clearly delineated and measured, which for an agent means tying the sourcing share to a qualified-opportunity conversion rather than to the existence of a log entry. Where that measurement isn't possible yet, the responsible choice is to hold the sourcing share back rather than pay on a proxy for contribution that hasn't been validated.

The Operator Move: audit your comp plan for agent-sourced deals

Twenty minutes. Pull your current partner or channel comp plan and your PRM's crediting rules side by side, and answer four questions honestly.

1. Does either document mention an AI agent as a possible source of a deal, or does "sourced" still assume a person made first contact?
2. If a workflow builder's agent surfaced a lead that closed, is there a written rule for what they receive, or would it get argued out case by case?
3. Does anyone on your team hold operating and improving a sourcing agent as an explicit, compensated part of their role today, or is it unpaid work layered on top of an existing job?
4. Looking at your last five partner-sourced deals, could you say with confidence today who or what actually gets credit for sourcing each one?

Two or more "no" answers means you have a real gap opening in exactly the spot this episode describes, and it's worth closing before your first genuinely agent-sourced deal reaches a comp dispute rather than after one does.