Touching AI

No industry does frothy better than the tech industry and the “AI” boom is hitting all the marks and some:

Buzzwords galore? Check.

AI as savior and the world-will-never-be-the-same? Check.

Or AI as a dark angel that will lead to extinction or enslavement? Check.

So we’re either saved or damned. A familiar story.

So how did we get here?

Well, up until very recently, the cost to play in AI was prohibitively expensive. You needed to spend many, many millions on talent, compute, data (though this appears to have been, ahem, “borrowed”) and fund years of research without guarantees on ROI. There weren’t many players in position to make this level of investment.

In order to justify the costs, the potential upside had to be big (big as in Billions) – only well-funded companies with large TAMs could play at scale. As such, the major tech platforms were the ones best positioned to see a return given the size of the investment required. And invest they did – both internally (hard to calculate but assumed large) and externally. These investments have outpaced typically sources of innovation capital.

It made sense for them – they have the cash and, just as important, outsized profits to defend. As consumers we were less aware of these investments – AI/ML was leveraged to drive underlying features and UX (ads, feeds, recommendations etc) that powered billion dollar revenue lines but they weren’t the eye-popping, front-facing generative features we’ve experienced in the last year.

For the most part, the initial wave of AI (ML mostly) consumer tech applications were optimization and efficiency oriented. The “magic” was abstracted away into the application logic layers of products and experiences we used daily. Supervised learning algorithms that just got better and better at anticipating our needs. We were blissfully ignorant of Deep Learning or neural networks. In a sense, we were blind to its power, pervasiveness and potential.

There were exceptions – smartphone cameras rely heavily on AI/ML tech to process photos and power all the amazing features we’ve come to expect during each upgrade. But I’m not sure “we” (collectively) understood what was happening underneath the hood or thought to ourselves “this filter is going to take my job some day” etc. I’d also argue that more explicit AI forward experiences like onsite chatbots or assistants (Siri, Alexa) were so poor and frustrating that users were less than impressed.

The Day After ChatGPT
But that all changed quickly with the release of ChatGPT. We were all exposed to the power of generative AI applications and it was shocking. So much so that the conversation has been dominated by existential and apocalyptic pronouncements – we went from “Siri, you suck” to “Siri, have mercy on my life.”

Although still expensive to build and develop, AI tools have begun to be productized leading to more reasonable cost structures. This is driving the explosion of generative applications and our UX expectations have changed overnight. New generative functionality is being integrated (shoe-horned) into productivity, communication and workflow tools (see Office, G-Suite, Teams, Slack, Salesforce, Jira, Monday, HubSpot etc) on a daily basis. Seemingly all software coding environments have generative features already rolled out with more coming everyday.

Smaller (re: market-cap) industry-specific productivity applications are racing to integrate generative feature sets as well. These companies, however, will have to be more judicious with their AI investments. They can’t expect the same level of return given their relatively small user bases and TAM. At the same time, they can’t ignore the existential risk staring at them so they must move. Maybe the proprietary data sets already residing, silo’d inside their applications will provide them with an edge – maybe.

These smaller players will benefit the most from the large investments to productize LLM functionality, model hosting, code co-pilots and other AI development tools that lower the cost to develop and compute.

However, there are companies building “AI forward” applications from the ground up and taking aim at incumbent SaaS providers in every category. TBD whether or not incumbents can do gen AI well enough to retain current customers and continue to capture new ones.

Short Term Thinking