Fireside Chat with Itai Damti of Unit + Thomas George of Housecat

It was a pleasure to host a fireside on AI and fintech last week with Itai Damti and Thomas George.

Itai is the Co-Founder and CEO of Unit and has spent years building one of the most important pieces of embedded finance infrastructure out there. Thomas spent 10 years at building at Monzo + a stint as a fintech VC before starting Housecat, which helps companies actually implement AI tools (in fintech and beyond).

The conversation was honest in a way these conversations usually aren't. A few takeaways:

• It's still verrrrry early in agentic payments.
• Inside companies, the middle is getting absorbed.
• Chat is great for discovery + bad for finishing the job.
• AI fraud is massively underhyped.
• The "death of SaaS" is overhyped.

It's still verrrrry early in agentic payments.

Most of what gets called agentic payments today isn't real money moving. The headlines are way ahead of the volume.

The interesting question underneath: will agents hold money natively in their own accounts, or will they just touch the money already sitting in your bank + card accounts? Itai's feeling was that both will exist but the answer matters a lot for who ends up owning the customer relationship + the economics on agentic transactions.

Inside companies, the middle is getting absorbed.

There are two roles within companies right now. The people who build systems and the people who talk to humans. Everything in between (the paper-pushing middleware work) will get absorbed by AI.

That's a useful lens for thinking about org design + where vertical AI lands. The work that compresses is the work that lives between the builders + the customer-facing humans. If that's most of what your tool automates, you're in the path of value creation. If you're trying to replace the builders or the front-line humans, the bar is way higher.

Chat is great for discovery + bad for finishing the job.

Chat is genuinely good at discovery. It lowers the activation energy to get started and makes complex products approachable.

But chat is bad for finishing complex flows. Real work needs real UI. The pendulum is swinging back toward proper software experiences, with the software getting smarter underneath rather than the chat box getting bigger on top. The companies that figure out how to blend the two will win. The ones betting everything on chat as the universal interface are setting themselves up for friction at exactly the moment users need to get something done.

AI fraud is massively underhyped.

This was Itai's pick for the most underhyped story in fintech right now. Every form of identity in financial services is becoming forgeable. Deepfakes, AI-generated malware, synthetic identity at scale, voice cloning for account takeover. The institutions that built their fraud stacks on legacy assumptions about what's hard to fake are dangerously exposed. We'll almost certainly see a wave of major attacks in the next few mos and years. The companies that have been quietly investing in modern identity + fraud infra will look prescient very quickly.

The "death of SaaS" is overhyped.

This was Thomas's pick. It's easier than ever to get to 80% of any software implementation. AI has genuinely collapsed the cost of a first build.

But getting from 80% to the 99% needed for production software is as hard as it's ever been. Maybe harder. Users still want proper UX. Edge cases still need to be handled. Security still matters. Some existing SaaS businesses are exposed. But SaaS as a category is going nowhere. The ones that survive are the ones that take the foundation seriously + use AI to extend their surface area, instead of getting replaced by a chat wrapper.