Apollo’s AI agents: what they do, and how to set them up

What Apollo released

Apollo has released two AI features this year.

On 4 March 2026 it launched its AI Assistant, which runs prospect searches, list building, sequence creation and research from plain-language requests. Apollo reported nearly 20,000 weekly users during the beta, and said those users were 36% more likely to book a meeting in their first 14 days.

The assistant waits to be asked. The agents, documented in September 2026, do not. They run on a schedule, do one job each, and bring back recommendations for a person to act on.

The three agents

Sequence Optimization proposing one change at a time is the right design. Change three things in a sequence at once and you cannot tell which one moved the reply rate. It still needs volume behind it: a change judged on 150 sends is mostly noise. The cold email test size calculator shows how many sends it takes before a difference is real.

How approval works

Each agent runs on a schedule you set, daily, weekly or monthly, or on demand. When a run finishes, Apollo opens a conversation in the AI assistant with the results. If Slack is connected, the recommendation also arrives as a direct message, and pressing approve in the thread tells the agent to go ahead and report back.

Two settings deserve attention. Each agent has a credit approval limit: the number of credits it can spend in one run before it has to ask. And Apollo’s documentation says it may switch an agent on automatically for some eligible users based on their activity. Check the Agents page in the AI assistant to see what is already running in your account.

The mistake agents make faster

Apollo was implemented in the SDR teams at Locus and GoComet, and the failure I see most has nothing to do with the tool: teams export thousands of contacts before the ICP and messaging are settled, the campaigns flop, and Apollo gets the blame.

Lead Discovery runs that same export on a schedule. With a sharp ICP, that saves an SDR hours of list building every week. With a loose one, it produces a steady supply of plausible-looking contacts who were never going to buy, each with a confident reason attached. Sequence Creation has the same dependency on the context you give it.

So the order matters. Write the ICP and the messaging first, then let the agents scale them. The tool comparison that usually comes before this decision is Apollo vs ZoomInfo.

Apollo publishes one customer result for the agents: an account executive whose reply rate went from 0.9% to 4.8% after applying the recommendations. It is a single testimonial with no detail on volume or segment, so treat it as a reason to test rather than a benchmark.

Apollo set up from scratch: NuVista AI

The clearest picture of what the agents need comes from an Apollo account built from nothing. For NuVista AI, an AWS consulting partner in Atlanta, the work started with several sessions with the founders on what the product solves and why it is different. Those sessions produced the markets, ICPs and personas, and the messaging followed from them.

Only then was Apollo set up for campaigns, sequences and activity tracking, and NuVista’s one remote SDR was onboarded and coached on it: how sequences are built, what a campaign should look like, and how LinkedIn runs alongside email rather than beside it. The prospecting around AWS Summits ran on the same stack.

Those founder sessions are exactly what Apollo’s AI context center now asks for. Skip them and Sequence Creation writes from a generic description of your company. The 90-day build behind that engagement is in the outbound 90-day plan, and the vertical view is GTM for AI startups.

At larger scale the constraint moves from setup to consistency. When the GoComet SDR team grew from 10 to 23, output rose from 2 to 6 SQLs per rep per month. How that was run is in how to scale an SDR team.

How to set them up

• Fill in the AI context center first. The agents read it for your products, services and positioning.
• Start with Sequence Optimization on sequences that already have volume. It improves what is running and needs no new list.
• Run Lead Discovery weekly, not daily, until the matches are consistently ones an SDR would have picked.
• Set a low credit limit per run. Raise it once the recommendations hold up.
• Connect Slack so recommendations reach the person who owns the sequence.
• Keep a log of what you approve and reject. The rejection reasons tell you what to fix in the ICP.

If you also use Claude, Apollo publishes an official Claude connector that searches, enriches and sequences contacts from a conversation. It is covered in Claude for sales teams.

Who should turn them on

Turn them on if you are on Apollo Professional or Organization, your ICP is written down, and your SDRs spend more time building lists and sequences than talking to buyers.

Wait if your current Apollo campaigns are underperforming and nobody can say why. The agents will produce more of the same. A pipeline audit or an SDR performance diagnostic will tell you whether the problem is targeting, messaging or follow-up.

For how Apollo’s approach compares with dedicated AI SDR products, see the AI BDR platform map. HubSpot’s equivalent is the HubSpot Prospecting Agent.

References

• Use Agents to Improve Outbound Work. Apollo Knowledge Base. Agents, schedules, credit approval, plans, automatic activation.
• Apollo Launches AI Agents for Smarter Outbound in Slack. Apollo. Slack approval flow and the customer result quoted.
• Apollo Launches AI Assistant. Apollo, 4 March 2026. Beta usage and meeting figures.
• Apollo connector. Claude connector directory. All accessed 2 October 2026.