Three ad agencies let AI run the mechanics. One client lost $40,000.
A Tel Aviv co-founder, a Bangkok CEO and a Tampa founder, all running ad agencies, agree AI owns the setup. The fight is over who owns everything after it.
A junior media buyer's 4-hour campaign setup now takes minutes.
A home services client, meanwhile, watched $40,000 vanish in 6 weeks.
Same technology. The difference was whether a person looked at the work.
We asked three agency leaders which parts of the job AI has actually taken, and which still need a person. Everyone is deciding this right now, and the bill for guessing wrong arrives fast.
Four hours of setup became minutes. Onboarding is another story.
Samuel Huang, CEO of Tele Ads Agency (Bangkok, Thailand)
“At Tele Ads Agency in Singapore, our junior media buyers used to spend 4 hours setting up one Telegram ad campaign. Budget splits, audience targeting, bid caps, all of it. The AI tools we run now do that setup in minutes with fewer errors.”
“The most expensive mistake I see is automating too much too fast. A company hands its whole client onboarding to an agent, the agent misreads a nuanced brief, and the client walks. That lost account costs more than the salary saved.”
Setup is repetition. A client's brief is judgment. Huang draws the line right there.
The bids went to the bot. The offer didn't.
Roy Danino, Co-founder of Lachi Media (Tel Aviv, Israel)
“I watched a home services client burn $40,000 in 6 weeks because they let an AI tool write their ad copy and set their bids with zero human review.”
“AI manages bidding and budget pacing. A person writes the offer, builds the landing page and checks whether the leads are actually worth what we paid.”
Not an anti-AI story. His own agency lets AI run bidding. What differs is who checks the output.
One question decides whether the agent flies solo.
Zack Bowlby, CEO & Founder of ROI Amplified (Tampa, United States)
“I’ve seen AI take over pacing, anomaly alerts, CRM hygiene checks, and first-pass attribution summaries. The human still has to decide whether a lead is profitable, whether intake quality changed, and whether CAC makes sense against LTV.”
“For hire vs. agent, I ask one question: can a bad decision burn cash or distort revenue reporting before someone catches it? If yes, we use both: the agent monitors and recommends, but a person owns budget moves and accountability.”
Zack's answer is a test, not a headcount. Agents watch and recommend. A person signs.
Where they split
Danino: AI runs bidding and pacing. People own the offer, the landing page and lead quality.
Huang: AI runs the setup. People own the nuanced brief and the client relationship.
Bowlby: AI monitors and flags. People own budget moves and accountability.
All three hand the mechanics to the machine. They split on where the human line sits: creative, onboarding or money.
Danino's filter for it: "which 20% of the work carries the revenue".
My take
Read these three side by side and the pattern is hard to miss. Nobody says AI can't do the work. They say it can do the work faster than anyone can catch its mistakes. So the hiring question may be less about headcount and more about who is assigned to look.
Your turn
Where does your team draw the line: which task did you hand to AI for good, and which one did you take back?
Next: an HR-tech founder, a strategy consultant, a recruiting COO and an HR consultant agree AI can screen candidates, then disagree on where it goes wrong, from auto-rejections and leaked exclusion notes to the talent you stop growing.
