The CRM you already pay for grew agents this year. They arrived in a release note, between a new report type and a change to the mobile app, and they can be switched on by anyone with the right admin role. Zoho shipped them, Salesforce shipped them, HubSpot shipped them. What none of them shipped was the job of running them, because that job sits in your department, not theirs.
This is what record-to-act looks like in one department, inside sales at a company of a hundred and fifty people, with the agents on. Not the demo. The week.
A lead comes in from the website at 9:10. In the recording era, it sits until a rep opens it. In the routing era, an assignment rule gives it an owner and a task. In the acting era, the agent reads the form, reads the company's site, fills the industry, size and role fields, scores it, and writes all of that onto the record before the rep sees it.
At 9:14 it does the same for a lead whose company name is a misspelling of an existing account. The agent, acting on the record as submitted, creates a second account. By Friday there are two versions of the customer, and the quote goes to the wrong one.
That is the first job the department does not have: someone who owns what the agent may create and what it must only propose. The rule "agents may enrich a lead but may only propose a new account" is a decision. It is a good one. It has to be made by a person who understands the sales process, written where the system enforces it, and revisited when it is wrong. We argued the reach decision for the whole estate in AI agent governance for the mid-market; in one department it is a Monday morning.
The second job the agent takes is the first reply. A lead that scored above the threshold gets an email within the hour: a short answer to what they asked, a question back, a link to book time. The rep sees it in the thread and takes over from the second message.
The agent's first reply is now the company's first impression, hundreds of times a month, in a voice nobody in the department wrote. Whether it is allowed to quote a price, mention a competitor, promise a timeline or offer a discount is a policy. In the routing era those decisions were made one email at a time by the rep. In the acting era they are made once, in the agent's instructions, by whoever set it up, and they apply to every reply until someone changes them.
Someone has to own those instructions the way someone owns the price list. They have to read a sample of the replies every week, because the instructions were written against the leads the department expected and the leads that arrive are always different. That is reading the trail, and a trail nobody reads is an agent nobody is running.
The third job is follow-up. A deal with no activity in seven days gets a nudge from the agent, referencing the last conversation. It is the most obviously useful agent job in sales, because it is the one reps most reliably skip, and it is where the most trust is lost.
On Wednesday the agent nudges a prospect whose last email said their CFO had died and they would be in touch after the funeral. The last email is in the thread. The agent's instructions said to follow up on silence, and the record said silence.
Nobody writes that case into the rule in advance. Somebody has to own the exceptions: the list of records the agent must not touch, updated by the reps the moment a human situation appears, and checked by the agent before every action. In the routing era the rep was the exception handler by default. In the acting era the exception list is a system object, and it needs an owner and a habit.
The fourth job is the quote. The agent reads the account, the open opportunity, the price list, the stock position in the warehouse system and the last three quotes to similar customers, and drafts one. The rep reviews, adjusts, sends. Quote time drops from two days to an hour, and that is the number that justifies the whole thing.
It is also the job with the sharpest boundary question, because a quote is a commitment. May the agent send it, or only draft it? At what value does a person have to sign? Which discounts may it apply, and from which schedule? We wrote about where this line sits in agentic automation versus workflow automation: the difference is who decides, and the quote is the record where that difference is money.
The answer for most departments is draft only, below a threshold, with a person signing above it, and the threshold rising as the trail shows the drafts were right. That is a boundary that moves, and moving it is a decision someone owns.
By Friday the department has four jobs it did not have a week ago, and none of them is done by the agent. Someone owns the boundary: what each agent may create, change, send and see. Someone owns the instructions, which are now policy. Someone owns the exceptions list. Someone reads the trail and moves the boundary when the trail earns it.
That is one role. It is not IT, because the decisions are about the sales process. It is not the sales manager, who has a department to run, although it reports to them. It is an operations role, part time at first, that runs the agents the way a good sales operations person runs the rules today, and it has to exist before the agents are switched on, not after the first wrong quote.
The vendors will not tell you this, because the job is in your building. The scoring feature a CRM has had for years is the small version of the same thing: a model that scores leads is only as good as the person who checks what it is scoring on, which is why we compared Zia scoring with manual lead scoring on exactly that question. The agents are that question, multiplied by four jobs and given the power to write.
Name the operator. Write the boundary for each agent as rules the system enforces. Create the exceptions list and give the reps the habit of using it. Decide who reads the trail and when. Take the baseline for the one number each agent is meant to move. Then switch on one agent, in draft-only mode, and widen it as the trail earns trust.
If the department does not have someone to do that, the agents should stay off, and that is the honest answer most CRM admins will not give because it is not what the release note promised.
Running the agents is the work we sell as a managed subscription: the boundary, the instructions, the exceptions and the trail, with one accountable party measured on whether the agents' actions were right. Discovery is no-risk: we walk one department's week with you, name the four jobs, and you pay only if you go ahead.
In a no-risk discovery we look at your CRM, finance and operations systems and who owns each part, and show what a connected system would do differently. You pay only if you proceed. Or see how we approach it.
More on the same problem:
"Something isn't working." "We're outgrowing our tools." "We're afraid to make the wrong move." No-Risk Discovery is a short, practical conversation that gets you clarity before you commit to anything big. We'll tell you if we're a fit. If we're not, we'll tell you that too.
Say who you are and ask. A person replies within one business day.
