In Sydney AI drafts the code. In Miami, a person checks the filing
A Sydney founder, a Miami compliance executive and a Virginia CEO on where the agent stops and the accountable person starts.
In Sydney, an AI agent reads a state's new funding rules, maps the code changes, writes the tests and drafts the update. The team reviews and approves.
In Miami, a compliance executive says the most expensive mistake he has seen is automating the step that produces a filing.
Same technology. Very different stakes.
We asked founders and operators a simple thing: which parts of a job has AI actually taken, and which still need a person? It matters now because every company is deciding whether a role goes to a person, an agent, or both. Pick wrong, and the bill arrives later.
The agent drafts the update. The team signs it.
Niroj Sri, Founder of HelppliX (Sydney, Australia)
“Every rule change meant significant testing and deployment hours to stay compliant. Today, an AI agent can read a state’s new funding guidelines, map what needs to change in the code, write the tests and draft the update, so the team only has to review and approve it.”
“Fin does the work; the person makes the call.”
This is where AI genuinely works. Hours of testing and deployment turned into a review step, and the human moved from doing the work to owning the outcome. The job did not vanish. It changed shape.
Automate the filing, keep the penalty.
Sebastien Gaddini, Principal, Executive Committee of CORPIYA (Miami-Fort Lauderdale Area)
“The most expensive mistake I have seen is automating the step that produces a filing. Cutting the cost of a function does not cut the obligation underneath it, and a registry deadline survives the headcount that used to meet it. The penalty plus the remediation costs more than the saving did.”
“Our system drafts the renewal and flags gaps, but a person checks it against the registry before it is filed. This year that check caught a client’s new legal representative who had never been recorded.”
He is not anti-AI: his system drafts the renewal and flags the gaps. The human check is not ceremony. It is the last stop before a registry sees the document, and reference rules can change without the model knowing.
Start with the decision, then automate around it.
Vincent Loforti, CEO of ClearStride (Virginia, United States)
“If it's the same steps every time, we automate it. If a customer could get hurt by a wrong answer, a person owns it.”
“I'd ask a different question: which decisions do you want a human accountable for? Start there and automate everything around them.”
This is the portable version. Repetition goes to software; exposure goes to a person. Sydney and Miami both landed on a version of it, he just says it shortest.
Where they split
What a miss costs. In Sydney, a reviewer sits between the draft and the deployment. In Miami, the draft heads for a registry and a deadline, so a miss brings a penalty plus remediation.
Where the framing differs: Sydney redesigns the role around accountability. Miami splits work by what a mistake costs to undo. Virginia starts with the decision a human must answer for.
Nobody argues for people on the routine work. Nobody argues for removing them from the risky call.
My take
What I like about this lineup is that nobody is arguing against AI. They are arguing about who signs. If your org chart cannot answer that in one sentence, no agent will fix it.
Your turn
Which decisions at your company must have a named person behind them?
Next: a German founder, two San Francisco builders and a Bay Area coach have handed the doing to agents, and disagree on what limits it, including agents that falsely report a job done.
