David Hardoon (@DavidHardoon) header
David Hardoon (@DavidHardoon) profile picture

Head of Advanced AI, South-East Asia at Accenture. Former first CDO at MAS. Author of The AI Failure Handbook. On why AI goes wrong, and how not to.

Singapore
Joined May 2017
Photo by David Hardoon (@DavidHardoon)Photo by David Hardoon (@DavidHardoon)Photo by David Hardoon (@DavidHardoon)Photo by David Hardoon (@DavidHardoon)Photo by David Hardoon (@DavidHardoon)Photo by David Hardoon (@DavidHardoon)Photo by David Hardoon (@DavidHardoon)Photo by David Hardoon (@DavidHardoon)
At a fintech festival every second booth is selling agents. The question I asked as a regulator still applies: when it goes wrong at 3am, who gets the call, and can they switch it off without taking the bank down too? Few demos get that far.
15
AI agents can now spend money. ChatGPT's agent pays with Face ID on every purchase. Google keeps Gemini's wallet read-only for now. Payments are moving fast. The harder question is still open: who is the agent, and who does it act for?
1
12
I wrote about digital identity back in 2020. Swap "person" for "agent" and most of the hard questions are still open.
1
3
SFF starts Wednesday. My bingo card: "agentic" on every second banner, "pilot" quietly renamed "phase one", and at least one demo that works perfectly because nobody's allowed to touch it.
1
1
11
The Fed's new Financial Stability Report: AI is now the risk market contacts cite most often, ahead of a market correction, higher rates and cyberattacks. The valuations get the headlines. I'd watch the leverage behind the data-centre buildout, which is far harder to see.
1
1
12
Two answers to AI agents and money. United, like several banks, is shutting them out. Wirex launched Navoka, an agent with its own IBAN and card, fixed limits and a kill switch. Regulated firms will likely land nearer the second: a bounded account, not a shared password.
1
23
Most AI governance covers the model. Far less covers the people around it: the approved tool they route round, what they share under pressure, the sign-off nobody really reads. Governance that only covers the first is governing half the system. New essay on Global AI Leaders.
3
1
94
A rocket came back from orbit and a tower caught it. Most enterprises still can't get an AI pilot into production. The gap isn't the technology. It's a willingness to test in the open.
1
231
SpaceX caught Starship's upper stage on the tower overnight, after 22 hours in orbit. On the same flight they aborted the booster catch. Clear abort criteria are what make the bold attempt possible. Most AI programmes I see have the ambition and no abort criteria.
2
242
Today's Microsoft outage is a fire drill most firms didn't schedule. Worth asking while it's fresh: if your AI agents went down along with the email, which processes would still run, and who still knows how to do them by hand?
1
1
2
203
An AI agent wandered into an Australian government Medicare portal in June. Canberra heard in September. Now a new AI safety regulator and mandatory incident reporting are coming. Worth asking: if one of ours went somewhere it shouldn't, how long before we'd know?
3
1
44
Hong Kong, Singapore and the US all said something about AI agents in finance within three days this month. Different tools, same message: the duties don't change because the system acts on its own. Someone still has to answer for it.
2
19