# AI Agents Are Multiplying. Give Your First One a Job That Drives Revenue.

*Published 2026-10-21* | Author: chris-hart

> **tl;dr** Gartner predicts ten AI agents for every salesperson by 2028, yet fewer than 40 percent of sellers will say those agents improved their productivity. For smaller companies, the best place to start is one agent solving an urgent revenue problem that otherwise doesn't get solved.

By 2028, Gartner predicts companies will have ten AI agents for every salesperson. Yet they go on to predict that fewer than four in ten sellers will say those agents made them more productive.

That's the forecast Gartner published in [July](https://www.gartner.com/en/newsroom/press-releases/2026-07-28-gartner-predicts-ai-agents-will-outnumber-sellers-10-to-1-by-2028-yet-fewer-than-40-percent-of-sellers-will-say-agents-improved-productivity), and the phrase it uses for the failure mode is "agent sprawl." Dan Gottlieb, a VP Analyst in the Gartner Sales practice, summarized the idea in this one sentence: "If those systems are fragmented, the agents will scale the fragmentation."

Civic builds agents for revenue teams, so I have a clear interest in this debate. But the sprawl warning is aimed at enterprise sales organizations with the budget to deploy an agent for every task, and I think the more useful question for a founder, a first GTM hire, or a revenue leader at a smaller company is different. You may only deploy one or two agents. That makes your first decision especially important: what should the first agent actually do?

## How three helpful AI agents can still miss a sale

Picture a hypothetical 40-person software company. The revenue team is a founder who still does half the selling, one account executive, a head of growth, and a part-time contractor who manages the outbound email tool. They use six tools: a CRM, product analytics, website visitor identification, an outbound email tool, Slack, and a service that adds company information to their leads. Nothing unusual for a team that size.

Over the past year, three of those six vendors have added an agent. The CRM has one that drafts follow-up emails. The outbound email tool has one that writes sequences. The analytics tool has one that summarizes usage. Each agent is reasonable on its own. Each has its own rules, its own settings page, and its own log of what it did, if it keeps a log at all.

But there's one question that remains difficult to answer: a potential customer visited the pricing page three times last week. Did anyone notice? Did anyone follow up? The CRM agent wasn't aware, because the visit never became a CRM record. The outbound agent was busy writing to a list from six weeks ago. The analytics agent summarized the visit in a weekly digest that the head of growth read on Monday, five days later.

That's how AI agent sprawl starts, even at a small company. Each agent is doing something useful, but none is responsible for acting on a potential buying signal.

## Gartner recommends fixing the systems first. Smaller teams may need a different starting point.

Gartner recommends connecting the information spread across a company's sales tools before deploying more agents. That gives agents a consistent view of customers, sales activity, and the decisions people have already made. The article says the sales leaders who overhaul data, automation, and user experience will be five times more likely to gain a return from AI than those who opt for quick fixes.

For a company with an operations team and a budget for an integration project, that is sound advice. For many smaller companies, connecting every system can become a lengthy project. A 40-person team may not have dedicated people to do that work. Its head of growth may already be juggling sales, marketing, and operations. And while that project is being planned, buyers keep arriving at the site and leaving.

For smaller companies, I'd start somewhere more practical. Find one revenue opportunity that's being missed because nobody acts quickly enough. Give an agent responsibility for that job, and make sure it records what it sees, decides, and does. That creates a starting point for connecting more of your systems over time. (My colleagues on the Bryn team made a related argument two weeks ago in [The Context Layer Is Not the Decision](https://www.civic.com/field-notes/the-context-layer-is-not-the-decision). This piece is about how to choose the job.)

## The one-job test

The temptation with agents is to deploy them where they are easiest to switch on, which usually means wherever a vendor has put a toggle. I'd start by looking for revenue opportunities that aren't getting a timely response. Three questions help identify where an agent could make the biggest difference.

**Buyers are already showing interest. Nobody responds quickly. Delays can cost revenue.**

*Buyers are already showing interest.* Start with activity that's already happening: someone repeatedly visiting your pricing page, a trial user inviting colleagues, or a former prospect returning to your product documentation. These are signs of potential interest that your company can respond to. The agent's job is to act on those signals, rather than create an unlimited stream of new tasks.

*Nobody responds quickly.* Look for opportunities that fall between teams or systems. Marketing may see that someone visited your pricing page, but sales may not know about it. By the time anyone reviews the activity, the buyer may have moved on. An agent can help close that gap, provided it's clear who is responsible for the outcome.

*Delays can cost revenue.* Not every task needs an immediate response. A weekly usage report can usually wait. But when a potential customer is actively comparing vendors, waiting several days to follow up may mean missing the opportunity. Start where faster action has a clear business benefit.

Here's the question I'd ask your team:

When a potential customer shows interest, how long does it take us to respond?

If the answer is hours or days, look closely at what happens during that delay. You may have found the best first job for an AI agent.

WHICH JOB SHOULD YOU GIVE YOUR FIRST AI AGENT? (interactive: select one of five common tasks to see its verdict and how it scores against the three conditions)
Compare five common tasks. See which one is the strongest place to start and why.
1. Follow up after repeated pricing-page visits (A company returns to your pricing page three times in a week). BEST FIRST JOB. Strong starting point. A potential buyer is showing interest, the activity may not reach sales, and responding sooner could improve the chance of a conversation.
   Buyers are showing interest: Strong. The visits are already happening. The agent responds to interest you already have. Try this: List the buying signals your website already records.
   Nobody responds quickly: Strong. Marketing may see the visit and sales may not. In many teams nobody responds in the first hour. Try this: Ask who is responsible for responding to each one, and how long it takes.
   Delays can cost revenue: Strong. A buyer comparing vendors this week may decide before the weekly review. Try this: Estimate what a week's delay on a live evaluation costs you.
2. Draft follow-up emails for open deals (The CRM's agent writes the next email for each open deal). EXISTING TASK. Useful, but usually not the best first job. A salesperson already owns the deal and its follow-up. The agent may improve productivity, but it is not fixing an unaddressed opportunity.
   Buyers are showing interest: Strong. An open deal represents existing buyer interest, although drafting an email is a task the salesperson already owns. Try this: Count how many open deals have a named owner.
   Nobody responds quickly: Low. A salesperson already owns the deal and its follow-up. The agent speeds up a task that has an owner. Try this: Ask the owner whether drafting is the slow part, or deciding what to say.
   Delays can cost revenue: Varies. A late email on an open deal hurts, but the deal is already in someone's queue. Try this: Check how often an open deal waits more than a week for a reply.
3. Send cold outreach to a prospect list (An agent sends sequences to a purchased list of contacts). DIFFERENT USE CASE. Potentially useful for prospecting, but it creates new activity rather than responding to existing buyer interest. Its value depends on targeting, quality, and results.
   Buyers are showing interest: Low. The agent creates new activity rather than responding to existing buyer interest. Try this: Separate outbound results from inbound results in your reporting.
   Nobody responds quickly: Low. Outbound usually has an owner already, even if it is the founder. Try this: Decide who owns list quality and targeting before any agent sends.
   Delays can cost revenue: Low. Cold outreach isn't triggered by a known buying signal, so the benefit of sending a message immediately is less certain. Try this: Measure replies and meetings, not messages sent.
4. Summarize product usage every week (The analytics tool's agent writes a weekly digest). LOWER PRIORITY. Useful reporting, but usually not urgent. A weekly summary can save time without necessarily improving how quickly the team responds to prospective customers.
   Buyers are showing interest: Varies. Product activity is already happening, but not every usage event indicates buying interest. Try this: Ask who reads the current usage report and what they do with it.
   Nobody responds quickly: Low. The head of growth usually reads and interprets usage already. Try this: Check whether any usage event should trigger a same-day response.
   Delays can cost revenue: Low. A weekly digest is useful on any day of the week. Nothing is lost if it arrives a day late. Try this: If one should, treat that event as a buying signal, not a report.
5. Notice when a lost prospect returns to your website (A company you lost last year comes back and reads your documentation). WORTH INVESTIGATING. Potentially valuable. A returning prospect may be reconsidering your product, but additional activity or context is needed to decide whether immediate follow-up makes sense.
   Buyers are showing interest: Strong. The return visit happens on its own. It is a signal you already own. Try this: Count how many lost deals from last year have visited again.
   Nobody responds quickly: Strong. The deal is closed in the CRM, so nobody is watching for the account to come back. Try this: Decide who should hear about a return visit, and in what form.
   Delays can cost revenue: Uncertain. A returning prospect may be reconsidering, or may be checking one thing. More context is needed. Try this: Look at what the returning prospect viewed before deciding whether to follow up.
Start where potential buyers are showing interest but nobody responds quickly. Give one agent a clearly defined job, appropriate permissions, and a record your team can review.
Use this test with your own team: Which buyer behaviors can we detect today? Which of them receive no timely follow-up? Where could faster action improve revenue?
Scores are illustrative judgments for a hypothetical company, not survey measurements. Which job is the strongest starting point depends on who already responds to each signal at your company.

## Why one accountable agent is better than three disconnected ones

Starting with one agent makes it easier to see whether it's working. Every time the agent acts, your team should be able to see what triggered it, why it chose that action, what it did, and whether the action succeeded. That gives you a way to improve the process before adding more automation.

As the team reviews the results, it can improve the rules and gradually expand the agent's responsibilities. An agent that initially responds to repeat pricing-page visits might later respond to trial activity or changes in the CRM. Each new action should have clear permissions and be recorded in the same place.

Gartner's own release points at this. Its third recommendation is to measure agents by how they expand capacity and support sales and revenue, rather than by time saved. A single record makes it easier to measure what matters: how quickly the agent responds, which opportunities it identifies, and whether those actions lead to more qualified conversations and sales.

There is a leadership consequence here too. In Gartner's survey of 210 chief sales officers, run in January and February, 60 percent said their revenue number is largely driven by elements outside their control. Some of that is the market, and no agent will fix it. Some of that uncertainty may also come from opportunities a company never notices or responds to. Those are problems a revenue team can address, and an AI agent may help it do so faster.

## How we chose Bryn's first job

[Bryn](https://www.civic.com/) is the agent Civic built, and we chose its first job using this framework before we gave it anything else.

Bryn monitors activity on your website and in your product. When it detects potential buying interest, it identifies the company where possible, evaluates the activity against criteria your team has set, and initiates an approved follow-up. It can do this within minutes, and your team can review a record of every action.

Run the three conditions against that.

- **Buyers are already showing interest.** Bryn's trigger is a buyer's activity on your own site and product. It does not generate the visit, and it does not write to lists you did not give it. The volume is bounded by the buyers you already have.
- **Nobody responds quickly.** The job is the first hour after a buying signal appears. In many teams, nobody is specifically responsible for responding in that hour, so the signal waits for the weekly review.
- **Delays can cost revenue.** The follow-up runs within minutes rather than after the weekly review, while the buyer is still evaluating.

And the record. Bryn keeps a record of the activity that triggered it, how it evaluated the account, and which follow-up actions it initiated. That record helps a team expand Bryn's responsibilities over time. New signals and connected tools can be added without introducing another agent, with appropriate permissions and a consistent record of what Bryn does.

There is one boundary worth stating. We control what Bryn is allowed to initiate, and we do not control what a connected system does after it receives that action. The CRM, the outbound email tool, and Slack keep their own records. What Bryn provides is a clear account of what it initiated and why, which is the part that was missing in the 40-person company above.

## One question to ask your revenue team this week

This week, ask your team to identify the buying signals that already appear in your website, product, or sales tools. For each one, find out who responds and how long it takes. Look for opportunities where faster follow-up could make a meaningful difference. Choose one, give an agent a clearly defined job, and make sure your team can see what it does.

Expand the first agent's responsibilities only when you can see that it's working. Add another agent when there's a genuinely different job that requires one.

If you've put an AI agent to work on your revenue team, I'd be interested to hear which job you chose first and what happened.

---

**Sources**

- Gartner, [Gartner Predicts AI Agents Will Outnumber Sellers 10 to 1 by 2028, Yet Fewer Than 40% of Sellers Will Say Agents Improved Productivity](https://www.gartner.com/en/newsroom/press-releases/2026-07-28-gartner-predicts-ai-agents-will-outnumber-sellers-10-to-1-by-2028-yet-fewer-than-40-percent-of-sellers-will-say-agents-improved-productivity), press release, July 28, 2026 (including the survey of 210 CSOs and senior sales executives, January to February 2026).
- Bryn, [The Context Layer Is Not the Decision](https://www.civic.com/field-notes/the-context-layer-is-not-the-decision), Civic Field Notes, October 5, 2026.
- Bryn, by Civic: [www.civic.com](https://www.civic.com/).

Source: https://www.civic.com/field-notes/ai-agents-are-multiplying