Klarna's rollback does not mean what you think it means
Why Chamath Palihapitiya misreads the impact of AI in organizations
Earlier this year, Klarna rolled back it’s big bet on AI replacing customer service. I wrote about it here. Chamath argues this means that AI struggles in production, and so it’s impact is overblown:
Not nearly enough people are talking about the implications of Klarna rolling back some of their AI bets.
Not knowing any of the details, I can guess why:
Replacing determinism or humans with probabilistic code is fraught with edge cases and require new ways of software development and process engineering that aren't well solved yet.
The implications to an entire generation of AI "apps" will be severe as more companies come to terms with the difficulty in getting products to work reliably in production with AI in the loop. Customer Service may be the first funeral signpost.
The result will be that many startups will need to pivot to simply use AI for narrow use cases and otherwise act deterministically. So what have people funded then? An AI company? Not really. Just a really overpriced SaaS company at AI valuations.
I disagree.
It’s true that AI works probabilistically, although it’s more accurate to say associatively, and so it struggles at tasks that need deterministic control. Fortunately, we have a solution — it’s called traditional software. The value of generative AI will come from applying it to associative tasks which are difficult or impossible to codify. Even this is too much of a generalization, as it’s hard to predict precisely which kinds of associative tasks AI is actually good at. Nevertheless, such tasks are common enough that any company built around automating them will have a radically different cost structure than incumbents, which make it different than a typical SaaS company.
Developing with AI will require new ways of software engineering (but is pair programming really that new?) as well as process engineering (evals have a long way to go) but both seem tractable. It may be difficult to retrofit them into an existing org, but there’s no reason they cannot be foundational in a new enterprise.