API Renaissance: The Dawn of API Unicorns Supporting Invisible Apps
Reimagining the Software World for an AI-driven Future
What if I told you the next wave of billion-dollar startups won't even have a website or an app? And that most consumers will be unaware of the next seven tech unicorn ideas I outline below.
Current Apps Are Optimized Around Humans
A company’s website or mobile app has been meticulously a/b tested and hyper-optimized around human behavior. Every popup, hover effect, and button placement has been curated to appeal to our human instincts and guide us toward checkout.
Our most remarkable new AI tech meticulously scours this maze of pixels to execute commands. For example, watch how this demo by Adept, which recently closed a $350 million raise, takes over a user’s Chrome browser to run through search parameters on Redfin. Incredible!
And yet, this will soon be outdated and unnecessary.
Let’s Imagine a World Optimized Around AI
Let’s imagine a world where apps are rebooted from the ground up and designed for semi-intelligent AI’s performing tasks on behalf of their human masters. Instead of optimizing beautiful responsive websites for unpredictable humans, the next generation of apps only needs to offer efficient endpoints for AI bots.
We are about to have an API renaissance.
[API sidebar for non-engineers:]
APIs are how the brains (server-side logic layer) of App A can tell App B what to do. For example, imagine a social media management platform like Hootsuite needing to post a tweet to Twitter on your behalf. Twitter would need to have an API endpoint that lets other apps create a tweet on behalf of a user. When you schedule a tweet in the Hootsuite platform, Hootsuite’s logic layer waits until the appropriate time, and then automatically sends your draft tweet via a command to Twitter’s API. Twitter then confirms back to Hootsuite that the tweet was processed successfully, and Hootsuite displays that back to you. A developer had to manually code the Hootsuite logic layer to work with the Twitter API.
[And this next part is probably interesting for everyone:]
In the future, somebody’s AI will draft a tweet based on their voice note, their chosen news headline, or a hot take from a transcript from a recent conversation… and this same AI will then search the internet for Twitter’s official API, it will “learn” on the fly how to interact with Twitter’s API, and send the appropriate command. Nobody will need to manually teach the AI how to interact with Twitter’s API. And if Twitter changes its API, then the AI will automatically update itself the next time it runs.
A Bank Without a Website
Imagine a bank with no tellers, no ATMs — but also no websites and mobile apps. The entire bank is just a series of API calls. Users simply tell their AI assistants “send $300 from my checking account to Jerry Morgan’s” and their AI will direct the bank on what to do. The bank just needs to offer a super clear API workflow that a GPT/LLM will correctly interpret 99.9999% of the time (and have its own AI safeguards for the other 0.0001%).
Is it crazy? Maybe. But we’ve gone from accepting cash payments to accepting physical credit card payments to now having a $95 billion company called Stripe emerge that only exposes API endpoints. You rarely even type in your credit card information anymore. The brain built into your Google Chrome browser scans the page for fields that resemble a Stripe credit card form and offer to autofill it on your behalf!
Let’s keep going:
An insurance company just needs to offer a convenient way for your AI to fill in your application information to provide you with a quote and bind your coverage. The companies have already stopped sending actual adjustment agents and instead pay claims based on AI-analyzed photos of the damages
A real estate brokerage must simply offer a convenient way for your AI to search and discover the properties it knows you find most appealing (why make Adept’s ai jump through all those human hoops?!)
A grocery store just needs to let your AI run through your weekly shopping list (based on your past shopping behaviors, the feedback from your smart refrigerator, and that Thai dinner party you have coming up on your calendar this Tuesday)
A restaurant delivery company just has to let your AI re-order your favorite pizza from your favorite pizza joint (with blue cheese on the side, as usual)
A doctor’s office can expose its scheduling software so your AI can select a slot, optimizing for commute times in your neighborhood based on rush hours and the other meetings you have that day.
A fashion ecomm website can let AI efficiently scour collections of pre LLM-tagged product images to identify and purchase the most on-style products for their user’s upcoming dinner party.
You get the point.
If you have a company today, then now is the time to consider how you want to prepare your logic layers for this new invisible AI future.
The Picks and Shovels of this New World
If we can agree that APIs are about to have their breakthrough moment as we rebuild our world around AI, what tooling needs to be built for APIs servicing AI as the new consumer?
API generation: every company will need an API layer. If before, technology service providers competed to provide small businesses with the best-looking website design, the next wave of providers will compete on the most effective API layer. Platforms like Shopify, Squarespace, and Wix will likely double down on their API support for their clients. Companies with custom builds will need to write their own external-facing API layers.
API documentation: alongside the APIs, we will need robust documentation for each API endpoint in a format that LLM/GPT can effectively read
AI Accountability: with AI’s speaking to AI’s, the inevitable failures will often happen behind the scenes. How will we know that whether the AI correctly interpreted an API? Provided the correct information? Got the desired result?
API authentication: currently, the user must sign into a third-party app to give access (e.g., you might sign in with twitter to give Hootsuite access to your account). What happens when AI’s need to sign in on a user’s behalf?
API pay walling: how does a company charge for a simple data request? Paying $2.9% + 0.30 for each credit card transaction won’t fly. Will this finally be an area where the promise of low transaction cost blockchain-based micropayments emerges?
API revenue generation: imagine you have an app that has loads of data that AI’s would love to ingest – how can you monetize it? Recently, Reddit and Twitter announced that they will charge (or sue) for automated ingestion of their data. Companies like Prequel are helping companies monetize their API data flows, and this space is sure to blossom further.
API marketplace or exchange: how will we standardize all the various APIs, allow that someone’s GPT assistant can discover the API, and ensure it is using the latest verified version of the API? Perhaps we end up with several marketplaces based on industry verticals (e.g., a Financial Services, Security, Social Media Platforms)
Each of these opportunities will breed at least one unicorn.
ChatGPT’s plugins are likely the first incarnation of this “API data layer app store.” But I doubt the Google, Apple, and Amazon of the world will willingly give such authority to this wolf in the chicken coop.
Zapier might also be well positioned to capitalize on this massive trend, given its myriad of integrations. As I covered on a recent Bot Meets World episode, they’ve recently announced their Interfaces product, which will soon allow users to chain API endpoints with some GPT-powered generative AI steps.
Whoever emerges as the winner, I think we can expect to see new tooling around APIs that enables a new class of invisible apps optimized around their AI patrons.
Do you want to build in this space?
Great. Let’s talk.
I’m actively investing in ventures that explore these new paradigms.
Whether I fund your venture, we build something together via our Magnetic Venture Studio, or we just nerd out about the future of this space over coffee, let’s chat. Honored to support any AI-related innovation.
Thanks to Paige, Tal, and my brilliant engineer dad for reviewing early drafts.