Doing the work

Should I let an AI SDR send our cold email, and how do I stop it burning our domain?

Treat an AI SDR as a sending machine, because that is what decides whether it hurts you. The tools are good at research, list building and a first draft, and the damage comes from volume: an agent that writes and sends hundreds of lightly varied emails a day trips the same limits a careless person would, only faster. Gmail and Yahoo since February 2024, and Microsoft's Outlook.com since May 2025, expect SPF, DKIM and DMARC from bulk senders (Gmail and Microsoft draw the line at 5,000 messages a day), and Gmail and Yahoo want spam complaints under 0.3 percent. So keep the agent on secondary domains, never your main one, cap each mailbox at the twenty to thirty sends a day you would set for a person, verify every address, and watch bounce and complaint rates weekly rather than reply rates alone. Have someone read a sample of what went out every week, because generic generated mail is exactly what recipients mark as spam. If a vendor's pitch is how many emails it can send, treat that as the warning, not the feature.

Go deeper

5 resources, 5 link-checked.

📄 Article
✓ Link checked Free Advanced

The counterweight to every AI SDR pitch deck, with the failure modes named and numbered: sender reputation collapse, intent-data false positives at 31 to 47 percent, and the eighteen month reply-rate half-life. Read it before you sign an annual contract.

The Case Against AI SDRs: Contrarian Analysis 2026

From Digital Applied 18 min read

  • Claims median sender reputation falls from 92 to 54 within 90 days of deploying an AI SDR.
  • Reply rates on AI-SDR campaigns decay over 60 percent in 18 months (11.2 to 4.4 percent) as recipients pattern-match the template.
  • An audit of 14 B2B SaaS firms found 31 to 47 percent false positives in intent data, so roughly 1 in 3 flagged accounts was not in-market.
  • Concedes four places they do work: mid-funnel nurture, campaigns under 500 accounts, research synthesis with human delivery, and reply routing.
Open digitalapplied.com →
📄 Article
✓ Link checked Free Advanced

It gives you a four part test (autonomy, reasoning, tool use, memory) to tell a real agent from a renamed feature, which is the skill a GTM leader needs in every vendor meeting this year. The numbers are sobering and useful: 62 percent of teams experimenting, only 23 percent in production.

The best AI agents for B2B sales in 2026: What's real vs. hype

From Mutiny by Matt Ratchford 20 min read

  • 88 percent of AI agent pilots never reach production, and only 11 percent of enterprises fully deploy.
  • 62 percent are experimenting but just 23 percent have scaled agents to even one function.
  • Total cost of ownership runs 3.4x higher than API-only estimates once integration and monitoring count.
  • Independent test of 11x Alice: 847 emails, 11 replies, 1 meeting from 200 leads over two weeks.
Open mutinyhq.com →
📄 Article
✓ Link checked Free Beginner

The clearest single page on the boring setup nobody explains to a first-time sender: why you buy secondary domains, what SPF, DKIM and DMARC actually do, and a three week warm-up schedule with real numbers rather than vague advice to go slow.

Cold Email in 2026: Domains, Deliverability, Replies

From Unify by Austin Hughes 15 min read

  • Warm a mailbox for about three weeks (roughly 5 a day, then 10, then 20) before full volume.
  • Cap at 25 emails per mailbox per day, going to 65 only when warranted, and never send cold from your primary company domain.
  • Keep reported spam under 0.10 percent (about 3 complaints per 1,000) and never let it hit 0.30 percent.
  • The first email captures 58 percent of all replies; average reply rate is 3.43 percent, top 10 percent clear 10.7 percent.
Open unifygtm.com →
📄 Article
✓ Link checked Free Intermediate

Microsoft's own announcement of the sender rules for Outlook.com, Hotmail and Live addresses, which made Microsoft the third of the big mailbox providers, after Gmail and Yahoo, to require authentication from bulk senders.

Strengthening email security: Outlook's new requirements for high-volume senders

From Microsoft Tech Community (Outlook blog) by Puneeth 6 min read

  • Applies to domains sending more than 5,000 emails a day to Outlook.com, Hotmail and Live addresses.
  • SPF and DKIM must pass, and DMARC needs at least a p=none policy aligned with SPF or DKIM, preferably both.
  • Enforcement began on 5 May 2025, with non-compliant mail first routed to Junk and rejection to follow, and the Safe Senders list does not bypass it.
  • Also recommends a visible unsubscribe link, regular removal of invalid addresses, and mailing only people who consented.
Open techcommunity.microsoft.com →

Browse all 795 resources →

The same ground, at another level

How outbound that works reads from a different seat.

Terms in this answer

People also ask

Also in Starting Up

The same ground, over in Get your first customers, our Starting Up track.

Also in D2C

The same ground, over in Grow organically & retain, our D2C track.

eChai Partner Brands The partnership →