Airtasker CEO: AI bosses are selling fear, not science

“Conjectures are being pushed as fact, manifesting the future,” says Tim Fung. And he’s not happy about it as he sits across the table from me in the restaurant. He’s referring to the pronouncements by AI CEOs, as reported in the media.

“That’s the job of any CEO,” says Fung, to amplify their company’s worth, “but we should not take that as just some sort of scientific fact.”

Tim Fung is a serial entrepreneur who co-founded the Australian company Airtasker with Jonathan Lui in 2012, which he leads as CEO. The platform lets people post small jobs online for local workers, known as Taskers, to bid on, across jobs from mowing lawns and moving furniture to taking photographs at a wedding.

I’m meeting up with him on one of his regular trips to London, because he has a small team here at Waterloo. We meet for lunch at King’s Cross, London’s Knowledge Quarter, where Google DeepMind is long established and both OpenAI and Anthropic have taken space.

CEOs of AI companies are the titans of this decade and Fung thinks they have done a good job in positioning themselves as “likeable characters” or, in the case of someone like Dario Amodei of Anthropic, someone who is “purely a scientist; he's not selling you anything, look at him, he doesn't look slick.”

This is what annoys Fung, even though his company is not in direct competition with them.

The pitch that sells by scaring

Then he slips into an impression of how that pitch sounds to him: “He'll always go like, ‘I got really worried because I looked at the AI and then it started doing things that I simply couldn't comprehend’. And I'm like, can anyone not read the subtitles here, ‘my product is awesome’.”

Fung skewers elements of AI fear-mongering as a pure sales tactic. Tell someone that Anthropic is the best, he argues, and many will disagree on the spot. "But if you actually can say to someone, ‘I used my product this morning, and I just got scared about what it was doing to my staff, and the fact that I'm going to have to lay them all off’, you're selling without selling, which I just think is bloody genius.”

In other words, the emotional aspect to this, the fear, becomes a sales tactic.

What if the fear is real?

In fact, an example of the realisation of this real fear unfolded this past week (after I met Fung). The ChatGPT maker was testing a system in a safely walled-off, isolated 'sandbox' so nothing it did could affect the real world. The model found a vulnerability and managed to access the internet, breaking into the server of tech company Hugging Face using stolen credentials; all on its own, without being told to.

Hugging Face confirmed the attack and OpenAI said it had happened while it had been testing its “state-of-the-art cyber capabilities”. On the one hand, it’s a salutary warning of what can happen when AI acts autonomously. But that global news story simultaneously showcased the power of the products OpenAI is developing.

Fung's workforce uses AI and he certainly sees value in it. He reckons that the efficiency gains are real but only around 50 per cent of that comes from the technology itself.

The other 50 per cent comes from the fact that “it's put a rocket up everyone's ass who is a CEO or founder, and basically said, ‘you're about to get screwed if you don't move faster’, and that [message] trickles down.” He sees his own teams shipping products much faster than usual, for example. There’s a new AI-inspired sense of urgency.

The business model Tim Fung doesn’t buy

However, despite AI’s usefulness and the fact it has picked up many enterprise clients, Fung also has doubts about AI as a long-term business model, on an industry level.

“If you started saying,” explains Fung, “every time you press the AI button, the query button on Claude or on Gemini, you’ve got to pay a pound. I don't think that business model is going to fly. I’d think, I'm just going to go to another one that doesn't charge me a pound. There's no moat, or rather, there are diminishing moats.

"There is probably a difference between a completely untrained model and a quite-trained model, but over the next years it seems like at some point that just taps out. So an open source model will be marginally different to a highly-trained excellent model.”

In other words, people may gravitate to the free models, cannibalising the industry. Unless of course the free model is then completely withdrawn, I point out, which often happens after the owners of a new technology have managed to entice customers in, capturing huge market share.

“Maybe that is where the rabbit hole ends,” muses Fung. “It just feels like either government or big tech is going to have to pay for all this AI, because customers want it, but no one's willing to pay for it.”

It may also become the case that the compute power required for AI-powered tasks makes it prohibitively expensive, however useful it is. Some companies, like Tesla, have already restricted employees’ access to tokens as they noticed the costs going up.

Fung doesn’t see AI as a threat to his core business model. Around 85-90 per cent of the jobs advertised on his platform have to be performed in-person (like garden clearing or help with a house move).

But when the robots come, he says with a smile, then we’ll have to rethink things for sure.

What does this mean for leaders?

Half your AI gain may not be the tool. Fung reckons only 50 per cent of his efficiency improvement came from the technology, the rest from the urgency it created. Some of that is available through faster decision-making at no licence cost.

Budget AI as a utility, not a licence. Uber spent its entire 2026 AI budget by April and Tesla capped staff at $200 a week only after engineers had run up thousands. Put per-head limits and usage monitoring in place before you need them.

Know what can't be automated. Around 85 to 90 per cent of Airtasker's jobs have to be done by a person in a physical place. Audit what share of your own revenue depends on presence, regulated sign-off or a relationship software can't hold.