An agent wrote ten hook lines in under an hour. None sounded Malaysian
A founder in Australia, an automation founder in Hamburg and a consultant in Dubai say agents took the volume, and people kept the standard, the exceptions and the outcome.
An agent gave an Australian founder ten hook lines for one post in under an hour.
None of them sounded like a real Malaysian until he pushed back several times.
Fast is not the same as finished. That gap is the whole job market debate in miniature.
We asked three builders in Australia, Germany and the UAE where the gap sits, and who gets paid to close it.
Every company is deciding right now whether a role goes to a hire, an agent, or both, and the answer shapes who gets hired next. We asked for specific examples, not predictions.
Ten hooks in an hour. The ear stayed human.
Brian Wong, Founder of Suna (Unaryx) (Australia)
“This week an agent gave me ten hook lines for one post in under an hour. None sounded like a real Malaysian until I'd pushed back several times.”
“So it's "both", but not 50/50. The agent does the volume; the person sets the standard and owns the outcome.”
Drafting outreach, fact-checking and metrics tracking moved over completely. What stayed with the founder was the ear, the sense of what sounds right to a real reader, and the patience to push back until it landed.
The bot runs the steps. A person reads the exceptions.
Laurent Bellandi, Founder of NEWMIND PARTNERS (Dubai, United Arab Emirates)
“The pattern we build most often is simple: the AI runs the routine steps and a person reviews only the exceptions before anything goes out.”
In Dubai the split is by task, not by job title. Re-keying orders, matching documents and routing requests go to the workflow, while judgement, exceptions and client relationships stay with people. It is a plain division of labor, with a human at the last gate.
Count errors, not heads.
Gleni Isaku, Founder of metanow.com (Hamburg, Germany)
“When we implement automation for clients, the win is rarely “replace the role”; it is “remove the busywork so the same person can ship more.”
“Practically: start with one low-risk process, define ownership and escalation before the agent goes live, and measure cycle time and error rate - not headcount cut.”
Triage, data entry between systems and routine CRM follow-ups are the easy handoffs. The metric he picks tells you what he thinks AI is for: a faster person, not a smaller team, with ownership and escalation settled before launch.
Where they split
Nobody here picked a side. All three leaders said both, and all three drew the line with a different ruler.
Brian measures volume against standard: "The agent does the volume; the person sets the standard and owns the outcome."
Laurent measures routine against exception, with "a person reviews only the exceptions before anything goes out."
Gleni measures output against headcount, and says to track "cycle time and error rate - not headcount cut."
My take
AI rarely takes a whole job. It takes the pieces that repeat. Look at what these three leaders kept for people: the standard, the exceptions and the outcome. If you are writing a job description, start there and let the agent have the rest of the bundle. Then ask who gets named when something goes wrong. That is a better hiring brief than a headcount target.
Your turn
Where does your line between agent volume and human judgment sit today, and who owns the outcome when the agent gets it wrong?
Next: an HR services CEO, an Irish consultant and an Indian software CEO argue the outcome should decide, with one finding the bottleneck was the owner, not the staff or the tools.
