AI took half the SDR job. The other half can burn your TAM.
A New York founder who fixes CRMs, a Tampa sales VP and a Dubai firm founder agree on what AI took from the SDR job, and on the one thing it shouldn't touch.
Here's the good news: AI has genuinely taken the research half of the sales development job. Account lists, signals, first drafts. Largely done.
Here's the bad news: the other half sends real messages to real buyers.
Joe LaGrutta, a founder in New York, says an AI SDR on a dirty CRM "burns domain reputation and TAM in a quarter." That is a very fast way to run out of market.
Three experts, three cities, one dividing line. Here is where they draw it.
Research went to the bots. The buying group didn't.
Richard Parkinson, VP Americas at Clarify B2B (Tampa, United States)
“In sales development, the research half of the SDR job has largely moved to AI: building account lists, watching intent and trigger signals, summarising a company's priorities and drafting a first-pass message.”
“The costliest mistake we see is automating first-touch outreach at volume into a short list of enterprise accounts. You can't unsend a bad message to a CIO, and when your market is a few hundred accounts, a burned account doesn't come back next quarter.”
Notice where he draws the line: The trouble starts when volume meets a short list. Enterprise tech, a few hundred names, no do-overs. That is the yes, with a very large asterisk.
The agent is fine. The CRM underneath it is the problem.
Joe LaGrutta, Founder of JayOh (New York City Metropolitan Area)
“The most expensive mistake: deploying an AI SDR on a dirty CRM. We've built or fixed 500+ HubSpot and Salesforce instances since 2016, and most teams buying an AI SDR can't trust their own lead status field.”
“The agent sends confidently wrong emails at 10x the volume and burns domain reputation and TAM in a quarter.”
The agent did exactly what it was built to do. It just did it fast, on data nobody could trust. The fix starts before the purchase, with the field the agent will believe.
Automate the prep. Keep the call.
Pratap Chandra, Founder of Keel Partners (Dubai, UAE)
“research preparation, monitoring, first-level data analysis, content drafting and routine follow-up tracking have moved almost completely to AI-assisted workflows.”
“The expensive mistake is automating judgment instead of automating preparation.”
Pratap Chandra's list of what moved is long, and none of it is a complaint. His line is short. Preparation goes to AI-assisted workflows. Client decisions, pricing and mandates stay with people. Same rule as the other two, from a Dubai desk.
Where they split
Where they differ is the failure mode. LaGrutta blames bad data: the agent amplifies whatever the CRM gets wrong, at 10x the volume.
Parkinson blames the market: a few hundred enterprise accounts leave little room for error. Chandra blames the category: judgment was never the thing to automate.
So one says fix the inputs, one says protect the list, one says draw the line at judgment. They are not three opinions. They are three checks to run before you press send.
My take
AI rarely takes a whole job. It takes the prep, and this lineup shows the prep is the easy yes. The expensive part is letting it keep going into work that needs a person answering for it. Before you ask hire or deploy, ask which steps would hurt if they went wrong, and who owns them. Give the agent the first set. Give people the second.
Your turn
So who owns the number at your company when an agent does the sending?
Next: in Sydney an AI agent reads new state funding rules and drafts the compliant code update, while a Miami compliance firm calls automating the step that produces a filing the most expensive mistake it has seen.
