Agents ran a 561-site study. Three founders still keep the client.
A Miami founder, a West Busselton founder and an Ahmedabad founder agree on where AI stops, and one warns that owners automate the wrong half.
A 561-site study that would have taken an analyst weeks by hand? Agents ran it.
Speaker onboarding, ticketing and sponsor follow-ups? Also handed to AI.
Three founders on three continents then drew the line in almost the same place. It is wherever a person, a relationship or a client win is on the other end.
We asked which parts of a job AI has actually taken, and which still need a person. Every company is making that call right now, so the line is worth finding.
Weeks of analyst work went to agents. The voice of the business didn't.
Adev Aarons, Founder of Specularis (Miami, United States)
“At Specularis, agents ran a 561-site study of Florida business websites that would have taken an analyst weeks by hand.”
“Anything that speaks for the business to a real person stays with a human, because a mistake there costs trust.”
Adev's rule is simple: agents take the work that needs no judgment, people keep the work with a face on it. The test is what a mistake costs, and who sees it happen.
Finding the lead is the easy part.
Nikhil Sharma, Founder of Event 1O1 (Ahmedabad, India)
“However, identifying a lead is not the same as winning a client.”
“My approach is to assign AI the research and preparation, while keeping humans responsible for verification, communication, commercial decisions, and execution.”
Nikhil does not ask what AI can do. He asks who has to answer for the result, and who has to hold the client's trust when it matters.
The admin went to AI. Owners handed over the wrong half.
Brooke Wright, Wright Mode and Techella at Techella (West Busselton, Australia)
“Speaker onboarding, ticketing and sponsor follow-ups moved to AI completely.”
“The most expensive mistake: owners automate the human part and keep the admin.”
Brooke's test comes before any tool: can you write down how the job gets done today? If not, no one is ready to take it, human or agent. Start with the map, then pick the worker.
Where they split
They agree on the easy part: agents take volume, research and admin.
They agree on the hard part too: people keep whatever faces a client or a relationship.
They differ on how to find the line. Adev prices the cost of a mistake, Nikhil asks who is responsible, and Brooke says to map the process first.
My take
Today, AI rarely takes a whole job. It takes the pieces nobody needed a person for. What stays is the piece with a name on it, the one where someone has to answer when it goes wrong. Before you buy the tool, write down who that someone is. Then check which half you are automating: if the admin is still on your desk while a bot talks to your best client, flip it.
Your turn
Which part of your own job moved to AI, and which part would you never hand over?
Next: a labor-data founder wants juniors to keep entry-level work, a UK HR director wants supervised judgement designed in, and a US educator says checking agent output can teach the same judgment.
