AI sales agents: the five kinds, what each one costs and what to check before you buy
What makes software an agent rather than an assistant or a sequence, the five kinds sold today, how each is priced and the checks to run before one writes, replies or calls in your name.
Quick answer: An AI sales agent is software that takes a sales goal and carries out the steps itself, choosing each next step within limits you set: it researches accounts, finds people, writes and sends outreach, answers replies or makes calls. The five kinds sold today are research, prospecting, outreach, reply-handling and voice agents, priced per action, per person, per agent, per conversation or per minute. Before you buy one, check what it may do without your approval, where it sends from and when it hands a conversation back to a person.
On this page 7 sections
A sequence sends step three on day five whether or not anything changed. An AI sales agent works differently: it looks at the situation, picks an action, takes it and checks what happened, inside the limits you give it. That one difference, that it decides, is what you pay for and what you need to test.
The label covers a lot of ground: a research bot, a voice agent that answers inbound calls and an agent that runs your whole first touch on LinkedIn are all sold as AI sales agents. This guide sorts them into five kinds, shows how each is priced on its vendor's own site and ends with the checks to run before one acts in your name. We make FOUND, one of these agents and it gets one short section near the end.
What makes software an agent
Three kinds of software help salespeople today. What separates them is who decides the next step.
| Kind | Who decides the next step | What you give it | What you get back |
|---|---|---|---|
| Automation, such as a sequence | You, in advance, for everyone on the list | A list and a set of steps | The same steps, on schedule |
| Assistant | You, each time | A question or a request | A draft, a summary or an answer and you act on it |
| Agent | The software, inside your rules | A goal, a buyer and limits | Actions taken, each with a record of why |
Most products mix the three. An outbound agent may write like an assistant and follow up like a sequence, yet still decide by itself who gets a message this week. So the useful question in a demo is which steps it takes without you. Assistants have their own comparison, by the job they do.
Adoption is not the hard part. Salesforce's State of Sales report says nine in 10 sales teams use agents or expect to within two years. Gartner predicted in July 2026 that by 2028 AI agents will outnumber sellers ten to one, yet fewer than 40% of sellers will say agents improved their productivity. Its analyst Dan Gottlieb put it plainly: "more agents will not automatically mean more productivity." The gap between those two numbers is the case for buying an agent by the job it does, not by the label on it.
The five kinds at a glance
Most products sold as AI sales agents do one of five jobs or bundle two or three of them. Knowing which job you are buying makes the rest of the comparison simple.
| Kind | What it does on its own | What stays with you | Usually priced by |
|---|---|---|---|
| 1. Research | Answers questions about an account or a person from the web, your CRM and data providers | Choosing the questions and spot-checking the sources | Actions or credits |
| 2. Prospecting | Turns a description of your buyer into ranked companies and people, often from signals | Describing the buyer and pruning the first lists | Per person, credits or a bundle |
| 3. Outreach | Writes and sends first messages and follow-ups by email, LinkedIn or both | The offer, the voice and reading the first messages it sends | Per agent, per seat or by contact volume |
| 4. Reply handling | Sorts each reply by intent, answers simple questions and offers a meeting | What it may promise and the threads it hands back | A plan tier or per conversation |
| 5. Voice | Holds calls with a synthetic voice or dials for a rep until a person answers | Scripts, consent and the moment a person takes over | Per minute, often plus a platform fee |
Each kind, up close
For each one: what it does, what to look at closely and products that sell it, described from their own websites on 25 Oct 2026.
Research agents
A research agent answers the questions a rep would otherwise search for by hand, the same way across hundreds of accounts. Does this company sell to hospitals? Who runs revenue operations there? The useful ones show where each answer came from, so a person can check it before it goes into a message.
- Look closely at
- Accuracy on accounts you already know. Run it on ten of your customers before you trust it on strangers.
- Examples
- Clay's Claygent is an AI agent with web research that fills any custom data point in a table. LinkedIn's Sales Navigator writes AI summaries of accounts and buyers, Account IQ and Lead IQ, on its Advanced and Advanced Plus plans, in select languages.
Prospecting agents
A prospecting agent turns "heads of RevOps at software companies with 50 to 500 people" into a ranked list of names. Some search a contact database by filters. Others watch for signals, such as a company hiring the team you sell to, a new leader or a fresh round and add people as the signals appear.
- Look closely at
- The filter after the search. Ask how each person is checked against your buyer and whether every lead shows the reason it is there.
- Examples
- Instantly's AI Sales Agent starts from your website, searches its database of 450M+ B2B leads for buyers that match, enriches them and writes the outreach. Salesforce's Agentforce ranks prospects with intent signals from enrichment data and researches each one across Salesforce, call recordings and the web. AiSDR charges $1.50 per person prospected and runs email and LinkedIn in one sequence.
Outreach agents
An outreach agent writes, sends and follows up. What separates a good one from a mail merge is whether each message says something true and specific about the person and whether it knows when to hold a message back.
- Look closely at
- Which account it sends from, yours or the vendor's, the daily and weekly caps and whether you can approve messages until you trust them.
- Examples
- Lavender's Ora writes and sends cold email; you build agents at no cost and pay $500 per agent per month once one launches, for 1,000 emails a month. Artisan's Ava, which it calls an AI BDR, finds leads, enriches them, sends personalized messages and books meetings, on plans scoped by leads contacted a month. 11x sells agents for outbound, inbound and expansion across voice, chat, email, social and SMS.
Reply-handling agents
The minutes after someone replies matter more than any template. A reply agent reads each answer, sorts it (interested, a question, not now, wrong person), stops the sequence, answers what it is allowed to and offers your meeting link to people who want to talk. Everything else goes to a person.
- Look closely at
- The list of things it may never promise: discounts, delivery dates, contract terms, anything legal. Then test it with the awkward replies your team really gets.
- Examples
- Instantly's AI Reply Agent reads every inbound reply and answers interested prospects in under 5 minutes, around the clock; its Human-in-the-Loop mode holds each reply for your approval and Autopilot sends. Salesforce's Agentforce answers lead questions from your CRM data and company content and offers real-time calendar availability for a meeting.
Voice agents
Voice agents come in two shapes. Some hold the call themselves, with a synthetic voice that qualifies an inbound lead or follows up on a list. Others work behind a human rep: they dial, get past phone trees and voicemail and put the rep on only when a person answers.
- Look closely at
- Consent and calling rules. In the US, the FCC ruled in February 2024 that calls made with AI-generated voices count as "artificial" under the Telephone Consumer Protection Act, so that law's rules for artificial-voice calls apply to them. Check the rules in every country you call before an AI voice dials anyone.
- Examples
- Retell AI sells voice agents pay as you go, at $0.07 to $0.31 a minute. Bland charges $0.14 a minute on its Start plan or $0.12 a minute plus a $299 monthly platform fee on Build. Orum's AI dialer pairs each rep with agents that find the number that answers, handle phone trees and voicemail and score calls.
What each kind costs
Vendors price agents in at least five different units, which makes side-by-side comparison hard. These are real entry points from each vendor's own pricing page, on monthly billing where the page offers it.
| Kind | Pricing unit | An example |
|---|---|---|
| Research | Actions and data credits | Clay Launch: $185 a month or $167 a month on annual billing, for its smallest bundle |
| Prospecting | Per person | AiSDR: $1.50 per person prospected, with research, lists and writing included |
| Prospecting and outreach | A bundle | Instantly's Starter bundle, with its AI Sales Agent: $94 a month |
| Outreach | Per agent | Lavender's Ora: $500 per agent per month once it launches |
| Reply handling | A plan tier | Instantly's Scale bundle, with its AI Reply Agent: $194 a month |
| Reply handling | Per conversation | Salesforce's Agentforce: $2 per conversation |
| Voice | Per minute | Retell AI: $0.07 to $0.31 a minute. Bland Build: $0.12 a minute plus $299 a month |
The price page is rarely the whole bill. Before you compare totals, add what it sends from (warmed-up mailboxes on a separate domain or a LinkedIn account in good standing), data credits for emails and phone numbers, the CRM and dialer seats of the people who take over and your own time reading its work in the first weeks. Then divide by the meetings you expect, not the messages.
Eight checks before you buy
Ask each of these in the demo, then confirm it in the product in the first week.
- What it does without you. Ask for the list of actions it takes alone and whether you can require approval for each kind: first messages, follow-ups, replies and meeting offers.
- Where it sends from. Your own mailbox and LinkedIn account or the vendor's? Who owns the sending domain if you leave?
- How it decides who fits. Every lead should carry a reason you can read and rejecting a lead should change the next list.
- What it says about how it found someone. A first line that mentions a like, a comment or a profile visit reads as being watched. Ask to see the rules that keep it out.
- Limits and pacing. Caps per day and per week for each account and what it does when a platform pushes back.
- The handoff. What it may promise, which questions go to a person, and how fast that person hears about them.
- The off switch. Can you pause one person, one campaign or everything and does a reply stop every queued step for that person?
- How you will measure it. Agree on the number before you start, qualified meetings or pipeline rather than messages sent and look at it after 30 days.
If the agent you are weighing works the top of the funnel, our guide to what an AI SDR does and how to choose one goes deeper on these questions and what an AI outbound agent should never do lists the rules worth holding any of them to. To compare products, the best AI SDR software shortlists eight agents, the best AI sales tools groups ten by job and what an AI BDR does covers testing one in a month.
Where FOUND fits
FOUND is an AI SDR for LinkedIn: a prospecting, outreach and reply agent in one, for buyers who answer on LinkedIn. You describe your buyer or paste your website and it drafts the buyer for you. Each agent watches up to 15 buying signals every day, such as people posting about the problem you solve, people engaging with a competitor's posts, new roles, companies hiring the team you sell to and fresh funding.
It checks every person against your buyer twice, the headline first and then the full profile and shows why each lead is there. It finds their verified work email and writes the connection note on a Premium account, every message and every email from what that person did, in one sequence sent from your own LinkedIn and mailbox at a human pace: nothing in a new account's first 48 hours, then 35 invitations a week, stepping up after two clean weeks to at most 125 a week and never more than 25 in a day. Review mode holds every invitation, message and email until you approve it, a reply stops everything queued for that person and anything the agent cannot promise comes to you. Replies from both land in one inbox, where the agent can answer them and offer your meeting link. Ask turns one sentence into up to 50 companies, with the right person at each and their email.
It costs $69 a month for two agents, one LinkedIn account, one mailbox and 100 credits (see pricing) and you see your first real leads before you pay.
Questions
What is an AI sales agent?
Software that takes a sales goal, such as meetings with a defined buyer, and carries out the steps itself: research, finding people, writing, sending, replying or calling. Unlike an assistant, it decides the next step within limits you set and unlike a sequence, it changes what it does based on what happens.
Is an AI sales agent the same as an AI SDR?
An AI SDR is one kind of AI sales agent. It works the top of the funnel: finding people, writing first messages, following up and booking meetings. The wider label also covers research agents, voice agents and agents that keep the CRM current or prepare quotes.
How much does an AI sales agent cost?
It depends on the unit it is priced in. On 25 Oct 2026, AiSDR charged $1.50 per person prospected, Lavender's Ora $500 per agent a month, Salesforce's Agentforce $2 per conversation and Retell AI $0.07 to $0.31 a minute for voice. FOUND is $69 a month for two agents, one LinkedIn account, one mailbox and 100 credits.
Can an AI sales agent replace a salesperson?
It can take the repetitive part of the job: research, list building, first messages, follow-ups and quick replies. Discovery calls, negotiation and the judgment calls stay with people, so the useful question is which hours of a seller's week you want back.
Is it legal to let an AI voice agent call prospects?
The rules differ by country. In the US, the FCC ruled in February 2024 that calls made with AI-generated voices count as artificial under the Telephone Consumer Protection Act, so that law's rules for artificial-voice calls, consent included, apply to them. Check the rules where your buyers are, with a lawyer if you plan to call at volume.