Y Combinator just published its new Requests for Startups for Summer 2026. For those unfamiliar with YC: it's the most influential startup accelerator in the world, with Airbnb, Stripe, Dropbox, Reddit, and Coinbase on its track record. Twice a year they publish a list of business directions and ideas they're actively looking for founders to build, as part of their accelerator program. Think of it as a compass showing where the smartest investors believe the biggest opportunities for (new) businesses lie in the years ahead.
The reason I find this list so interesting is twofold. First, because they're often very early in spotting (digital) opportunities. When they wrote in 2012 that they wanted to fund "the future of payments," Stripe was still in its early days. And second, because for the upcoming round they're betting very explicitly on the opportunities AI is enabling.
The opening line of Summer 2026 makes that clear:
"AI has stopped being a feature and started being the foundation."
The fifteen categories range from counter-drone defense to inference chips for agents. We want to highlight three categories, because we strongly believe in them ourselves, but also see big opportunities for others.
1. SaaS Challengers: the end of traditional software power
Jared Friedman opens his case with an observation you'll hear in every software company's boardroom: everyone is talking about how AI-coding means the end of SaaS. Investors have written off billions in market cap over it, and you may have seen the term "SaaSpocalypse" in the news.
The SaaS model decisively beat custom software in recent years because the latter was too expensive and cumbersome to build and maintain. A five-person team simply couldn't compete with the Salesforces of this world. AI has lowered the cost of producing software by a factor of 10 to 100. The millions of lines of code, built up over decades, that made you choose an existing solution, is barely a reason anymore to make that same decision today.
If you're currently paying €50K per year for a software package of which you use 20% of the functionality, chances are that within 18 months an AI-native challenger will exist that covers exactly your use case for a tenth of the price. Or, taking it a step further, you can ask yourself whether the solution you need couldn't simply be built and maintained internally with the help of AI.
2. Company Brain: the missing piece of AI automation
This may well be the most important category on the entire list for getting AI deployed effectively within your organization. Tom Blomfield (founder of Monzo, now a YC partner) offers the following diagnosis:
"The biggest blocker to AI automation of companies is no longer the models, they just got so good so quickly. Now the blocker is the domain knowledge."
Every company has critical knowledge scattered around. Some of it sits in people's heads. Some of it is buried in old emails, Slack threads, support tickets, and databases. The company works because people vaguely remember where that knowledge is and how to apply it. But AI agents can't operate that way.
Blomfield argues we need a new kind of solution: a company brain. Not company-wide search. Not a chatbot over documents. But a living map of how a company works. Think about how refunds get handled, how pricing exceptions get decided, how engineers respond to incidents. And on top of that, an executable skills file that AI systems can use to carry out that work safely and consistently.
Organizations are increasingly saying "we want to deploy AI agents" and then buying a license from one of the big AI labs. But the real bottleneck isn't in the models — those are more than good enough to perform the desired tasks. The challenge lies in the fact that there's no structured answer anywhere to "how do we actually do this?". An agent that needs to draft a quote has no idea that customer X always gets an 8% discount because they signed a major project in 2019. That knowledge sits in Karin's head over at the back office.
Before you invest in agents, copilots, or any AI tool, you need to understand which knowledge is implicitly held in your organization and how to make it explicit and queryable. I recently wrote about how to set up a digital brain for yourself or a small team in this piece.
3. The AI Operating System for Companies: from open loop to closed loop
The third category, written by Diana Hu, is essentially the natural follow-up to the Company Brain. Hu describes what she sees in the best AI-native companies in the YC portfolio: they've made their entire company queryable. Every meeting recorded, every ticket tracked, every customer interaction captured — all readable by an intelligence layer that learns from it.
The core concept is the shift from an open loop to a closed loop. In an open loop, you make a decision and maybe look at the results weeks later. In a closed loop, the system monitors what's happening, compares it with what should be happening, and adjusts accordingly. Hu says she's seen teams cut their sprint time in half and ship twice as much.
Since we ourselves have only just started and therefore have no historical sprint time, I find it hard to express a difference for ourselves. But our own Operating System is what we're betting on hardest, and I wrote about it last week in this article.
The problem most existing organizations now run into is that building this still requires brute integration labor. Gluing together Slack, Linear, GitHub, Notion, call recordings, and a dozen other tools with custom code. There's no off-the-shelf product yet that connects all this context into one intelligence layer that can reason across it.
Hu's point: what you need isn't a dashboard, but a system that turns a company's own artifacts into a self-improving loop.
The companies that invest in this over the next three years are going to build up an enormous lead over those that don't. You make your organization less dependent on the knowledge of the people who've been there for years, and you end up with an organization that learns faster as a whole.
The stack needs to be rebuilt
When you put these three categories side by side, you get what we see as a golden combination. The software of the next fifteen years will be built for agents running loops, for organizations that continuously learn, and for teams that can build in months what used to take years — fully tailored to the specific work they need to do.
SaaS Challengers replace the traditional one-solution-for-everyone application layer.
Company Brain replaces the knowledge layer that's currently dependent on highly loyal employees.
AI Operating System replaces the orchestration layer.
What we're doing with this and what you can do
At Think Again we work right at the intersection of these three categories, because we strongly believe they reinforce each other. We have the advantage that there are no existing processes, roles, or ways of working — other than what we know from previous work. That makes us flexible to set things up now in a way that fits where we think the world is moving. That's why we're building our own custom software with the goal of becoming an AI Operating System that holds all the knowledge about the work we do.
Where this will be a year from now I find hard to say, and I mostly hope to learn a lot from people who are working on this too.
Three things you can do yourself this week:
Make a list of your five most expensive SaaS licenses. Ask yourself for each: are we using more than 30% of the functionality? If not, is there an AI-native challenger ready and waiting?
Identify three "Karins" in your organization. People whose departure would be a serious problem because they carry knowledge that isn't documented anywhere. That's your company brain backlog.
Look at one repetitive work process that's an open loop. Something where you make a decision and only see the effect weeks later, because you simply haven't properly documented why it had a certain outcome the previous times. That's likely your biggest opportunity to build a closed loop with AI.
If you need help with this, we're of course happy to assist. In the meantime, we'll keep sharing how we're working on this ourselves.
