Agents can do part of the SDR job today. After a person writes the rules, they do account research, enrichment, email verification, first drafts and reply triage. A person still makes judgment calls, runs live calls and keeps relationships. Few companies have switched. In a 2025 survey of 351 B2B companies, 1% named "AI SDRs" in their team model.

Which SDR tasks do agents do well today?

Agents do the research and admin work before the first call. We automate each of these tasks in our own outbound. A person checks each one.

  1. Account research. The agent reads public feeds, job posts, funding news and on-chain data. A person writes the filter for which events count.
  2. Enrichment. The agent matches each event to one company, then finds the contact and title. A person writes the rules for namesakes and old titles.
  3. Verification. The agent checks every address before the send. A person decides how to treat catch-all addresses.
  4. First drafts. The agent writes an email from the signal's own fields. A person approves the template, the offer and the claims.
  5. Reply triage. The agent puts one of six labels on each reply. Four of the labels alert a person.

The gain is coverage. The agent reads every feed on every run and checks every address on every send.

Where do agents still make mistakes?

Agents make mistakes where the data is unclear. A person had to write a rule for each type of mistake.

  • Signal quality. Our catalyst monitor logged 6,069 buying signals from 26 July to 7 October 2026. Of those, 5,857 (96.5%) were too vague to email about (our data, October 2026). They were noise, the wrong company in the headline, drops in total value locked, or job posts outside sales. Another 91 named a company that our contact database could not match to one firm of 11 to 1,000 staff.
  • Namesakes. A hiring signal for one company matched a different company with a similar name. A second signal did the same. Our rule now takes the top exact-name match across all sizes, then applies the size filter.
  • Titles. One contact record had a job title that named a different company. We now drop any title that names another firm.

A person found each error in a sample of the agent's output. Now the agent applies each rule to every record.

Which parts of the job still need a person?

Three parts need a person. Our reply classifier uses six labels: positive, referral, question, negative, out of office and unclear. Four of them go to a person, who answers from the sending mailbox.

  • Judgment calls. A recent conversation, a partner conflict or a deal in progress decides if a signal earns an email this week. No filter sees these. Unclear replies land here too.
  • Live conversations. A buyer's question often has several parts and a hidden objection. The classifier labels it "question". A person answers it and runs every call after it.
  • Relationships. A referral, a follow-up months later and an introduction at a conference all depend on trust between two people. The classifier labels a referral. A person makes the introduction.

How many companies have replaced SDRs with agents so far?

Very few. The Bridge Group surveyed 351 B2B companies for its 2025 report. "AI SDRs" appeared there as a category for the first time. 1% of respondents chose it (Bridge Group, 2025). The most common model was one SDR for every two account executives, at 31%.

The same report sets the bar for an agent:

  • Median SDR on-target pay: $80K.
  • Average ramp: 3.0 months.
  • Median monthly quota: 10 first-stage meetings held.

Measure any agent on meetings held. Ignore activity counts such as emails sent. The survey data is from early 2025, and adoption has likely changed since then.

How should you split the work between agents and people?

Give agents the steps that a rule can decide. Set up the agents in this order:

  1. Write the buyer definition and the signal rules in plain words.
  2. Let agents collect signals, match each one to one company and find the contact.
  3. Read a sample of matches by hand each week. Write a rule for each new error type.
  4. Verify each address again before the first email if its last check is older than 30 days.
  5. Let agents draft the emails from templates that a person approved.

Then hand the replies to people:

  1. Let a classifier label the replies. Send positive, referral, question and unclear replies to a person the same day.
  2. Keep a person on every call and every introduction.

With this split, the SDR spends less time on lists and more time in conversations. For the system behind steps 1 to 6, read what GTM engineering is. To compare a team with a partner, read outbound agency vs in-house SDR team.

To see what these steps find in your own market, start with five named buyers, free. Each buyer comes with the reason they fit now.