AI sales agents: what an AI SDR does and which model fits

What is an AI sales agent, or AI SDR?

An AI sales agent is software that takes over steps of outbound prospecting that a sales development rep (SDR or BDR) would otherwise do by hand. The work is the same as it has always been: pick the accounts that fit your ideal customer profile, find the right people and their contact data, research why now is a good moment, write a first message and follow-ups, handle the replies and hand interested leads to an account executive. "AI SDR", "AI BDR" and "AI sales agent" are mostly vendor names for the same idea.

11x describes its AI SDR, Alice, as picking target accounts from live signals, researching each prospect, writing per-prospect copy rather than filling template variables, classifying replies and booking the meeting when interest is confirmed. That list is a fair map of the category. Where tools differ is which of those steps run without a person looking, where the lead data comes from, and how you pay.

• Account and lead selection: who to contact, from a database, signals or your own CRM.
• Research: company and person context that makes a message specific.
• Writing: first touch, follow-ups and replies, per channel.
• Sending and pacing: mailboxes, LinkedIn accounts, daily limits and stops on reply.
• Reply handling: triage, drafted answers, meeting booking or hand-off to a rep.

Sources: 11x docs: Alice overview

Two models: hands-off AI SDRs and research-first agents

Hands-off AI SDRs run outbound for you once a campaign is live. Artisan says its AI BDR, Ava, runs outbound end to end: finding and enriching leads, writing and sending outreach, handling replies and booking meetings. AiSDR calls itself a platform that automates outbound across all stages, with an agent, Ami, that proposes plays from your website and reviews each campaign after about two weeks. 11x's Alice runs one cadence across email, phone, LinkedIn and WhatsApp.

Research-first agents do the same research and drafting, then stop before anything leaves your workspace. A person reviews the list, the copy and the spend, and approves the send. The work is still mostly done by software; the decision about who hears from your company, and with what message, stays with your team.

The line between the two is softer than the marketing suggests. All three hands-off vendors ship approval controls: 11x lets you review every message, route messages to a named approver or run on Autopilot, set per segment. Artisan has a "Require approval before sending" setting, and without it Ava sends on her own once the campaign is live. AiSDR keeps your first campaign in draft until you approve it and offers a co-pilot mode for replies. The real question is which mode a tool is built around, and what it costs you to stay in that mode.

Sources: Artisan homepage, AiSDR LLM information page, 11x docs: Alice overview, 11x docs: brand protection and approval modes, Artisan help: campaign settings

What hands-off AI SDRs do well, and where they fall short

Hands-off AI SDRs are good at volume and consistency. They bundle the parts most teams struggle to assemble: 11x includes managed Gmail mailboxes with domain setup, warm-up, inbox rotation and monitoring in every plan, with contact data bundled in. AiSDR includes research, lists and writing inside $1.50 per person contacted on pay-as-you-go. Artisan caps how many new prospects Ava contacts per day and can answer replies on Autopilot, including booking meetings. For a team with a broad market, a short sales cycle and no SDRs, that is real capacity without hiring.

Their weak spots follow from the same design. Copy that no one reads before it sends can misstate your product or misjudge a prospect, and 11x's own documentation warns that teams who start on Autopilot and later find a problem have usually already sent the message that caused it. Pricing tends to assume volume: 11x charges per lead on an annual contract, with Growth listed from $3,750 a month billed annually for 2,000 new prospects, while its FAQ says Growth starts at $36,000 a year. AiSDR positions itself as added capacity, not a replacement for your SDRs.

• Good at: steady daily volume, bundled mailboxes and data, reply handling at any hour.
• Watch for: unseen sends on Autopilot, annual contracts, and prices tied to people contacted.

Sources: 11x Alice pricing, AiSDR pricing, Artisan help: campaign settings, 11x docs: brand protection and approval modes, AiSDR LLM information page

When high-ACV, sales-led teams should prefer approval-first

Which AI SDR tools work for B2B companies with high contract values? Mostly the ones you run with approvals on. When a deal is worth tens of thousands a year, your addressable market is a few hundred or a few thousand named accounts, and a buying committee has five people on it, each contact is expensive to waste. One wrong message to a VP at a target account costs more than the hours an SDR saves by not reading it.

Vendors say the same thing in their docs. 11x recommends an approval workflow for regulated industries and named strategic accounts, tells you to escalate from review to Autopilot in order, and suggests reading 15 to 20 drafted messages on a new segment before anything sends. In practice, high-ACV teams get most of the value from the research and drafting, and keep the send decision with a BDR or an account executive who knows the account.

• Prefer approval-first when: deals are large, the account list is finite, accounts are owned by named reps, or your industry is regulated.
• A hands-off AI SDR can fit when: the market is broad, deals are small, and a bad first touch costs little.
• Either way: start on one narrow segment, read the drafts, and move to less review only once the output holds up.

Sources: 11x docs: brand protection and approval modes

How to evaluate an AI sales agent

Run a pilot on one segment you know well, so you can judge the research and the copy, and ask each vendor the same six questions. Two of them are about rules other companies set. Google requires bulk senders, those sending more than 5,000 messages a day to Gmail accounts, to authenticate with SPF, DKIM and DMARC, offer one-click unsubscribe and keep reported spam below 0.30%. LinkedIn says it does not permit third-party software, including bots, that automates activity on its site, and that members who use it risk having their accounts restricted or shut down. Any AI SDR that runs LinkedIn steps carries that account risk, whoever runs it.

• Data source: its own contact database, a waterfall across providers, or your CRM, and what you pay when nothing is found.
• Channels: email only, or LinkedIn, WhatsApp and calls in one sequence that stops when a lead replies anywhere.
• Approvals: what waits for a person by default, per segment or per step, and who can switch it off.
• Deliverability: whose mailboxes and domains send, warm-up, daily limits, and what happens on bounces.
• LinkedIn account risk: how actions are paced, which account sends, and what happens if it is restricted.
• Cost model: per person contacted, per seat, or per workspace in credits, and the contract length.
• Pipeline and ROI: count meetings held and pipeline created per segment, not emails sent.

Sources: Google: email sender guidelines, LinkedIn help: prohibited software and extensions

Where OXYGEN fits: research-first, with approvals

OXYGEN is research-first. It is a GTM workspace where AI does the research and drafting and your team decides what goes out. The Copilot in the web app builds lead lists, writes sequences and runs the work, and asks before sends and larger paid steps. AI columns research and write over every row of a table, and a research agent cites its sources. Auto-approve is optional per chat and off by default; with it on, the Copilot stops asking, so keep it off for named accounts.

The tools that send sit in the same place. One sequence mixes email, LinkedIn, WhatsApp and call tasks (calls are in beta), tests up to 26 versions of a step, and stops a lead when they reply on any channel. LinkedIn runs in the cloud with no browser extension at $34 a month per connected account. Lead data comes from waterfalls across 17 email and 13 phone providers, charged only on a hit. Replies land in one inbox with AI triage and drafted answers, and on a built-in CRM with a deals pipeline.

There is no hands-off AI SDR in OXYGEN that books meetings on its own: a person, or your own AI assistant through the MCP server or the CLI, runs the work. Pricing is one plan per workspace from $49 a month in credits, with no seat per user, so BDRs, account executives and founders share it.

• Copilot and AI columns that research accounts and draft outreach.
• Approvals before sends and larger paid steps, unless you turn on Auto-approve.
• Email, LinkedIn and WhatsApp in one sequence, with a stop on reply.
• A CRM with a deals pipeline on every plan, the free one included.

Looking for a ranked list of AI SDR tools?

This page explains the category. For the shortlist, the best AI SDR tools guide compares seven tools on pricing, approvals, channels, data, CRM and MCP, from each vendor's own pages. If you are weighing OXYGEN against a specific hands-off AI SDR, the Artisan and AiSDR comparisons go row by row, and the pricing page shows what a workspace costs in credits.

What OXYGEN does not do here

• No hands-off AI SDR that finds leads, sends and books meetings on its own; a person or your own AI assistant runs the work.
• No contact database of its own: lead data comes from provider waterfalls on credits, or Apollo's database on your own Apollo key.
• No Salesforce sync; HubSpot syncs both ways.
• No SMS step, and calls are in beta.
• LinkedIn automation carries account risk on any tool, OXYGEN included.