Key takeaways
- An agent seat is a line in the capacity model, not a person with a quota.
- Capacity plans start from tasks; headcount follows from what the tasks need.
- Human review time is part of the agent seat's cost and needs its own hours.
- No hire should be cancelled on agent output nobody has measured yet.
- Measure tasks and teams, not individual reps, unless you are ready for the EU and California rules on AI that monitors staff.
What is an agent seat in a capacity plan
An agent seat is the line in a sales capacity plan that records work done by AI assistants or agents: the tasks they cover, the person accountable for their output, the cost basis and the review time they need. It sits beside the human roles rather than replacing one, and it carries no quota of its own.
Writing it down matters because agent work is otherwise invisible in the model. Without the line, the plan either ignores support the team relies on or quietly assumes savings nobody has checked. This page covers headcount and the agent seat; quotas, territories and the sales calendar sit in the 2027 sales plan.
Why does sales capacity planning change in 2027
It changes because part of the work no longer scales with people. The classic model multiplies ramped reps by their productivity; once research, list building, drafting and record upkeep run on assistants, the limit moves towards what only people do: conversations, judgement and approving what goes out.
Costs change shape too. Many AI tools bill by usage rather than by seat, so the agent line needs a cost basis that grows with work done. Treat any gain as a hypothesis until a quarter of real results backs it, and say so in the plan.
Which sales tasks stay with people
Keep with people the work where trust, judgement or accountability decide the outcome. Agents can prepare for most of it, but a person should own it.
- Discovery conversations and reading what a buyer does not say.
- Negotiation, pricing exceptions and commercial terms.
- Deciding which accounts to pursue and when to walk away.
- Approving messages and record changes before they reach a customer or the CRM.
- Relationships with champions, partners and existing customers.
How do you measure the pilot without monitoring reps
Measure tasks and team totals rather than scoring individual people. Time per task, drafts accepted, errors caught in review and pipeline touched, summed for the pilot group, answer the capacity question without building a tool that watches each rep.
The distinction matters in 2027. Under the EU AI Act, AI that monitors or evaluates workers becomes high-risk from 2 December 2027, with duties for the company using it, and California's CCPA regulations ask for a risk assessment before automated processing infers an employee's performance from systematic observation. This is not legal advice.
How do you turn the task split into a 2027 hiring plan
Turn it into hires role by role: the hours left on human-only and review work, set against the hours a ramped person gives, show how many people each role needs. Add ramp time for new starters and cover for leavers, and link each new hire to pilot evidence rather than assumed savings.
- Roles and the weekly tasks each one performs today.
- Task split: human, agent-assisted or agent-run, with an owner for each.
- Agent seat: tools, cost basis, review hours and the accountable manager.
- Pilot results: what was measured, over which period and with which group.
- Hiring plan by role and quarter, with ramp time and the evidence behind each hire.
Size 2027 headcount and the agent seat
Measure the work first; the model comes after the evidence.
- Break each sales role into its weekly tasks and note how long each takes today, from calendars and team-level CRM activity rather than estimates.
- Mark each task as human, agent-assisted or agent-run, and name the person who checks agent output.
- Run the agent-assisted tasks with a small group for a set period, measuring tasks and team totals rather than scoring individuals.
- Add the agent seat to the model with its cost basis, such as seats, usage or credits, and the review hours it needs.
- Work out the headcount each role needs from the remaining human hours, with ramp time for new starters.
- Decide whether freed time goes into more accounts per rep, deeper coverage of key accounts or new segments, then revisit the split each quarter.
Where Pebbles Ai fits
- CapabilityAI workspacePebbles Ai is an AI workspace whose assistants research accounts with cited sources, find fresh leads with work emails and write outreach and social posts from your company knowledge. The agent line's work is done by its Marketing and Strategy assistants in chat, under the execution mode the user picks: in Manual outside actions wait for approval, in Plan they are queued and in Auto they run without asking, which makes review time a setting you can plan for.
- CapabilityProspecting & customer recordsFind fresh leads works on up to ten target companies per run and uses one lead from the Pebbles wallet for each lead found, which gives list building a clear unit for the agent seat's cost basis.
Common questions
Does an agent seat replace a sales hire in 2027?
Not by default. The agent seat records work assistants take on; whether that changes hiring depends on what a measured pilot shows. A capacity plan should never assume a hire is saved before that evidence exists.
How do you budget for AI agents that bill by usage in 2027?
Estimate usage from a pilot, set a monthly ceiling, and review actual spend against the work done each quarter. Keep the agent line separate from seat licences so finance can see what drives the cost.
Who owns the agent seat in a 2027 capacity plan?
A named sales or RevOps manager, accountable for its output in the same way as for a team member's. They own the task list, the review rules and the quarterly check on cost and quality.
About this page. Written by the Pebbles Ai team for people who run sales, marketing and growth. Last updated .