Everyone is talking about AI agents right now. Vendors are announcing them, LinkedIn is full of people claiming they’ve replaced half their go-to-market team with one, and revenue leaders are being asked by their boards whether they’re “doing anything with AI.”
Most of the conversation is noise. Some of it is genuinely useful. The challenge is telling the difference quickly enough to make a good decision.
This piece is my attempt to cut through it — not to hype AI agents, not to dismiss them, but to give revenue operators, marketing leaders, and sales teams a clear-eyed view of what these things actually are, where they add value, and how to introduce them in a way that doesn’t blow up trust, waste budget, or create a mess your RevOps team spends the next year unpicking.
What an AI Agent Actually Is
An AI agent is software that can take action on your behalf — not just generate text or surface an insight, but actually do something: send an email, update a record, flag an account, trigger a workflow, book a meeting.
The distinction matters. Most AI tools in the revenue stack are still advisory — they analyse, they recommend, they summarise. An agent is different because it acts. Which means the stakes around getting it right are higher, and the consequences of getting it wrong are more visible.
In a revenue context, agents are being used (with varying degrees of success) for things like prospecting — identifying high-intent accounts and initiating outreach; expansion monitoring — detecting signals that suggest an existing customer is ready for more; churn prevention — flagging at-risk accounts before the renewal conversation becomes a crisis; and pipeline hygiene — updating CRM records, progressing stalled deals, or surfacing deals that need attention.
None of these are magic. They’re all workflows that humans currently do manually, with varying consistency and attention to detail. The agent just does them more reliably, at higher volume, and without forgetting.
Why Most Rollouts Fail
The teams that struggle with AI agents almost always make the same mistakes. They’re worth naming clearly.
They treat it as a technology project rather than an operating model change. Buying an AI agent tool is the easy part. Deciding exactly what it should do, how much autonomy it should have, what it should say, and what happens when it gets something wrong — that’s the hard part, and it’s where most implementations stall. The technology is ready before the organisation is.
They don’t start small enough. The temptation is to identify a big, high-impact use case and go straight there. Prospecting at scale. Full pipeline automation. End-to-end renewal management. These are compelling ambitions, but they’re also complex, data-hungry, and high-risk if something goes wrong at volume. The teams that make meaningful progress start with one workflow, prove it works, and expand from there.
They underestimate the data problem. An AI agent is only as good as the information it has access to. If your CRM is messy, your contact data is stale, your account records are inconsistently maintained, and your intent signals are unreliable — the agent will act on bad inputs and produce bad outputs. Garbage in, garbage out is not a new problem, but AI agents make it more consequential because the outputs are actions, not just reports.
They skip enablement. Sales reps and marketing teams need to understand what the agent is doing, why, and how to work alongside it. If they don’t trust it, they’ll override it constantly. If they trust it too much, they’ll stop paying attention. Neither is the right relationship. Human oversight isn’t a workaround — it’s a design requirement.
They don’t define what “good” looks like before they launch. What tone should the agent use in outreach? When should it act and when should it wait for a human to review? What’s the trigger for escalation? What does a successful outcome look like and how will you measure it? These questions need answers before go-live, not after the first wave of complaints.
A Framework for Getting It Right
Based on what I’ve seen work (and what I’ve seen go sideways), here’s how I’d approach a sensible AI agent rollout for a revenue team.
Start with trust and safety, not speed. Before anything goes live, answer these questions: Who has access to what data? What are the privacy and compliance implications of the agent acting on customer or prospect information? What’s the audit trail — can you see what the agent did and why? What are the fail-safes if something goes wrong? This isn’t bureaucracy. It’s the foundation that allows you to move fast later without creating liability.
Pick one workflow and go deep. Choose a starting point based on two criteria: where the pain is sharpest, and where the data is cleanest. If your CRM data for enterprise accounts is solid and your team is losing deals because of slow follow-up, start there. If your prospecting data is patchy but your customer health scores are well-maintained, start with expansion or retention. Work with what you have, not with what you wish you had.
Map what the agent needs to know. For any given workflow, an agent needs to answer three questions before it can act: Should I act here? What should I do? What should I say or send? Map the data sources required to answer each of those questions reliably. If those sources aren’t available, connected, or clean, address that first. Trying to automate a broken data process doesn’t fix the process — it just makes the errors faster.
Define “good” explicitly. Write down what a good outcome looks like for the first workflow. What does a good outreach message look like — in terms of tone, length, personalisation level, and what it should and shouldn’t include? What’s the right level of autonomy (fully automated, or human review before sending)? What are the guardrails — accounts the agent should never touch, situations where it should always escalate? These feel like small details. They’re not. They’re what determines whether the agent behaves consistently or embarrassingly.
Build in human oversight from the start. The most successful implementations treat the first phase as human-in-the-loop: the agent does the work, a human reviews and approves before it goes live. This builds confidence, surfaces edge cases, and creates a feedback mechanism. Once you have evidence that the agent is performing well and teams trust it, you can gradually extend its autonomy. Going the other way — starting fully automated and adding oversight after something goes wrong — is much harder.
Measure the right things. The metrics that matter in the first phase are not pipeline or revenue (these are too slow and too influenced by other factors). They’re process metrics: how many actions did the agent take, how many were reviewed and approved without changes, how many were edited, how many were rejected and why. These tell you whether the agent is calibrated correctly and where it needs refinement. Once the process metrics are strong, you’ll start to see the business metrics follow.
The RevOps Role in All of This
If you’re in a RevOps function, AI agents are both an opportunity and a responsibility. You’re likely the person who understands the data stack, the workflow architecture, and the gap between what the CRM contains and what it should contain. That makes you the most important person in the room when an AI agent rollout is being planned.
Your job in this context is to be the voice of data reality. To push back when a proposed use case depends on data that doesn’t exist or systems that aren’t connected. To design the measurement framework that tells leadership whether the agent is working. To own the integration between the agent and the rest of the revenue stack — CRM, marketing automation, customer success platform — so that actions taken by the agent are reflected consistently across the business.
You’re also the person who will be asked to fix it if it goes wrong. Which is the best possible reason to be involved from day one.
A Word on Vendor Claims
Every sales engagement platform, CRM, and marketing automation tool is currently announcing AI agent capabilities. Some of these are genuine step-changes in what the product can do. Some are existing automation features with a new name and a higher price point.
The questions worth asking when a vendor pitches you AI agent functionality: What is the agent actually doing — is it taking actions or generating suggestions? What data does it need access to and how does it connect to our existing systems? What are the controls around what it can and can’t do? How do I audit what it’s done? What happens when it gets something wrong?
If a vendor is vague on any of those, that’s information.
The Honest Summary
AI agents are a genuine development, not a buzzword. In the right context, with the right data, the right guardrails, and the right starting use case, they can materially change what a revenue team is capable of — not by replacing people, but by handling the consistent, process-driven work that currently eats time and attention that’s better spent elsewhere.
But they require the same thing that every meaningful operational change requires: clear thinking about what you’re trying to achieve, honest assessment of whether your data and processes are ready, and enough patience to start small and prove something before you scale it.
The teams winning with AI agents aren’t the ones who moved fastest. They’re the ones who were most deliberate about what they were building and why.
That’s not a new lesson. It just keeps needing to be said.
