An AI SDR agent is software that does part of a sales development rep's job on its own: building lists, researching accounts, drafting and sending messages, answering inbound chat, logging activity. The useful question is not whether to deploy one but which tasks to hand over. Our data from 1,000,000 B2B cold calls says the live conversation is the one task you keep human.

Two things people mean by "AI SDR agent"

The label covers two products that behave very differently.

The autonomous agent runs the whole loop: picks accounts, writes the email, sends it, handles the reply, books the meeting. The rep is not in the loop. It sells volume.

The copilot agent prepares the work and the rep executes: research summary, draft to review, call list ordered by signal, summary after the call. It sells time.

The first one looks good on a spreadsheet. The second one is what we see working for B2B teams in France and Europe, where buyers pick up fewer than 12 calls in 100 and delete templated emails in seconds.

Where agents win

Be generous here. There is a lot of SDR work an agent does better, faster and without complaining.

• Account research: the website, the last funding round, the hiring page, the LinkedIn activity of the three people you want. Ten minutes per account by hand, seconds for an agent.
• List building and enrichment: filtering an ICP, finding the mobile number across several providers, flagging the ones that bounced.
• Signal monitoring: watching the CRM for job changes, dormant leads, closed-lost anniversaries and website visits. Reps use about 20% of the signals already in their CRM; an agent can watch 100% of them.
• Drafting: a first version of the email or the opener, built on the trigger. The rep edits and sends.
• Inbound triage: answering the website chat, qualifying, routing to the right rep.
• Admin: call summary, disposition, next step, written back to HubSpot or Salesforce.

Every one of these is a task reps hate and do inconsistently. Hand them over.

Where humans still win: the call

The live conversation is where the meeting is booked, and the numbers show it is a human skill.

• A discovery question starting with "how" lifts conversion by 190% compared with closed questions.
• Asking "What do you mean?" before answering an objection lifts conversion by 197%. The rep has to hear the objection, pause and get curious.
• Calling a known contact with "How have you been?" reaches about 10% success, against 1.5% for a standard cold open. That is a relationship, not a script.
• Trigger-based openers book meetings 7 to 10% of the time versus 2 to 3% for pitch-first openers. The agent finds the trigger; the rep uses it in a way that does not sound like a mail merge.

An autonomous agent can send 2,000 emails a day. It cannot hear a CFO hesitate and ask her what she means.

The European constraint nobody mentions

Agents that make voice calls on their own are a legal problem in Europe before they are a quality problem.

The ePrivacy directive requires prior consent for direct marketing by automated calling machines without human intervention. In France, consumer cold calling has had legal hours and attempt limits since 2023, and the ARCEP numbering rules ban 06 and 07 mobile numbers as the displayed number for prospecting calls. An AI voice agent that dials prospects "like a rep" walks straight into all of that.

Email agents face the GDPR side: legitimate interest has to be documented, the data source has to be traceable and the prospect has to be able to object. A fully autonomous loop that scrapes, writes and sends makes that paper trail hard to produce when the CNIL asks for it.

A realistic split for a European outbound team

Here is the division of labour we recommend. It is the one we run ourselves.

Before the call block (agent): watch the signals, build today's list of 60 to 100 contacts, enrich mobile numbers, write a prep card per account with the trigger and a suggested opener.

During the block (rep, with the agent on screen): parallel dial, take the conversation, use the trigger, ask "how" questions, answer the objection with a question first. Target: 20 conversations a day.

After the block (agent): transcribe, summarise, update the CRM, draft the follow-up email, flag the coaching moment for the manager.

Inbound, all day (agent): chat triage and qualification, with a human handoff the moment a real buyer shows up.

The rep's day flips from mostly preparation and admin to mostly conversation. That is the productivity gain, and it is bigger than anything a fully autonomous agent delivers, because it compounds on the one step that books meetings.

How Sonik approaches AI SDR agents

Sonik is the copilot model. Our agents watch the CRM signals (website visitors, job changes, closed-lost, dormant leads, social intent), build the call queue, prepare each account and write the summary back to HubSpot or Salesforce. The rep makes the call on a parallel dialer that produces 5x more conversations than one line at a time.

We do not sell an AI that calls your prospects for you. In Europe we think it is illegal in most cases and unwelcome in all of them. Our mission is 2x more meetings via 5x more conversations with the right contacts, with a human on every one of them.

Conclusion

Deploy an AI SDR agent for research, signals, drafting, triage and admin. Keep the call human, because the data says that is where the 190% and 197% lifts live. This week, list every task your SDRs did yesterday and tag each one "agent" or "rep". If more than half the hours are tagged "agent", you know what to automate first.