Venture and corporate investment in Artificial Intelligence (AI) is at all-time highs (up to $9.3B according to some estimates). There is tons of optimism but also some high profile failures.
Watching the current trends, there are interesting parallels between the AI explosion and the early API / platform markets. Specifically, there is a challenge for all new companies on where to focus: on a general technology platform or service with broad re-use? or a specific application?
It also brings to mind an old adage from the AI world:
If it doesn’t work it’s AI. If it works it’s Engineering.Unknown, circa 1995
That self-deprecating sentiment certain used to often be true, though it’s more in vogue to be optimistic these days!
I suspect might see a resurgence of this sentiment as AI technology becomes more mainstream. If it’s highly optimized for one application, is it really Intelligence? or just a smart algorithm? The “general versus specific” dilemma is also very real for today’s AI companies.
In the early API market, there was a multitude of API-first companies that provided truly innovative generally reusable API based services. A number of these did eventually win out (Twilio, Stripe and Sendgrid amongst others), but there were also many that struggled. It turned out to be very difficult to build a successful business on a general API platform service alone since it depended on developers and other third parties inventing truly high-value applications on the platform. The applications had to be high value, otherwise, there was no budget to pay through to the company offering up the APIs. Paradoxically, as the value of the applications grew it created pressure for these third parties to move off the original API as costs rose, or at the very least, negotiate aggressively.
There are companies that made it through this trial of fire and found the right balance between amazing services, cost, and addressable market size. For many others, the challenge became “what is the killer app” using our APIs and who should build it.
One of the most successful answers turned out to be “lets us build that killer app”. In other words, many of the API success stories combined their API-led go to market with investments in applications/services which addressed an end user market directly. Even Twilio did this with its OpenPBX solution.
I suspect a similar dynamic will play out in the AI startup world. Possibly in an even more extreme fashion. Algorithms and methods are valuable in AI but data really is king. Without data it’s extremely hard to train systems, validate them and add any type of value. Algorithms and methods which work for one domain (let’s say healthcare data analysis) often don’t translate well to another (let’s say transportation logistics). These factors make it doubly hard to succeed with a “generic” API capability as a product/service.
I suspect we’ll see more and more cases where companies “use” AI to themselves become players in an industry of choice. This gives them full access to customers, to data and to all the dynamics in between. Going full stack is the only way to really be certain that things are working.
Just remember that when they work, for now at least people might question if it’s really AI!
Photo by Clyde Thomas on Unsplash.