📶 YC request for startups (S25)

Suppose you believe that LLMs are now able to automate a lot of legal work. There are two things you might do with that idea. You could build an AI agent and sell it to law firms. That's what most people do. Or, you could start your own law firm, staff it with AI agents, and compete with the existing law firms. That, my friends, is going full-stack. You could do this for any industry, especially one dominated by slow-moving incumbents. Instead of selling to the dinosaurs, you could make them extinct.

Over the next decade, as new coding tools make it easier than ever to build and ship products quickly, great design is going to matter even more. Especially as many designers are worried about AI replacing their jobs, the real opportunity is for designers to use AI themselves to launch their own products and build their own companies. The design job of the future will be "founder". Designers already have so many of the skills needed to be great – strong user empathy, a focus on solving problems, a high bar for quality, and taste. These are a must for every founding team.

Humans interact with businesses in many ways, but one way hasn't changed much in almost 100 years—and that's phone calls. You probably have experienced long wait times, voice trees, the press star or pound. In the end, we always want to talk to a human agent because the alternatives are just not that great. But the new voice models and conversational LLMs are now incredibly good. And startups who take advantage of them are are now making voice AI bots that are indistinguishable from humans. It's pretty amazing and the unlock was the quality of these models.

A lot of the software tools used for scientific applications in chemistry, biology, materials science, or operations research haven't changed much in decades. These rely on standard methods along PhDs to solve complex problems in drug discovery, chemical process optimization, metals & mining, or power grid optimization. In particular, test-time compute is unlocking new types of startups that can solve these scientific problems.

Think of an assistant that knows how you typically respond to certain kinds of emails and can draft those responses for you—or even send them with your approval. It understands which meetings you usually accept and which you decline, and that can auto-schedule your week in a way that optimises your productivity and minimises travel. It keeps track of tasks you've completed in the past and how you approach new ones, so it can handle recurring processes without your constant input.

The US healthcare system is over 17% of GDP or over 4 trillion dollars. There are estimates that one-third of that, or over 1 trillion, is just spent on administrative tasks. We have one of the best healthcare systems in the world, but sadly, a lot of this spending is just unnecessary administration that exists because different health systems are not interoperable, don't have APIs, or simply the only way of doing a workflow or task if a human manually extracts data from one system to another.

Today, with AI, we think it's finally possible to build a truly personal tutor for everyone. The latest reasoning capabilities let them break down complex topics step by step, in a way that should help explain even the most complicated subjects in a straightforward way. With the latest foundation models being multimodal, they can now present these concepts by generating animations, manipulating 3D objects, and explaining concepts with voice.

Robotics hasn't had its ChatGPT moment yet, but we think it's almost here. Everyone has known that robots are the future - But that proved elusive because previous generations of robots were expensive, brittle, and only worked in controlled conditions. With the rapid improvements in foundation models, it's finally possible to make robots that have human-level perception and judgment. That's been the missing piece.

The rise of AI—particularly large language models—promises to reshape how we teach, learn, and measure outcomes. These technologies have the potential to significantly improve educational access, personalize instruction, and free both teachers and learners from drudgery. We're just starting to see new personalized study tools for students and grading platforms for teachers, but we're still very early in figuring out what AI can truly achieve here. One big challenge is figuring out the business model - while more than a billion people are engaged in education, budgets are notoriously tight, and purchasing cycles are horrendously slow.

Consumers spend $20B year on home security. And the biggest players are legacy companies whose features haven't changed in decades. Meanwhile, AI is taking the world of commercial security by storm. Companies like Verkada, Lumana are showing just how powerful AI can be. In commercial buildings, doors are unlocked with facial recognition. Video cameras detect suspicious behavior and send push notifications to building managers. Virtual AI guards will even come on loudspeakers and scare away criminals – to prevent crime before it happens.

Soon, all companies will have one thing in common: every employee will build their own agents to automate the repetitive parts of their jobs. An internal agent builder is a tool that I can use to create an agent to handle mundane tasks I don't like doing. This infrastructure will need to access every other piece of software I use in my day-to-day. It'll need to manage permissions, and send sensitive data securely to the latest generation of LLMs.

Every day I drive to work for 20 minutes, and every day I wish I could be using that time to triage emails, draft replies, delegate scheduling. What if we could all arrive at work each day already at Inbox Zero? If you've used things like Vapi, Retell, or even ChatGPT Advanced Voice Mode, you know how good these voice agents are today. The use cases are far bigger than just driving. I know firsthand - It took me experiencing nerve damage in my arms to realize how much we still rely on keyboards for even the most basic tasks. And email just might be the perfect starting place to build the most useful all-purpose assistant. Anyone with access to my inbox instantly knows my friends, my plans, my writing style, and so much more.

Most humans are not rational about their finances. But just being alive comes with a set of decisions you need to make about your finances. How much should I save for the future, where should I invest my money to take the right risk, and how should I think about debt and taxes? How you answer these questions today will impact your freedom to do the things you want in the future. Until today, the answers have generally been asking friends, googling, hiring a financial advisor or maybe even asking your bank. These options aren't great. They are filled with bias and don't have the full picture of your financial situation and goals. And hiring a great financial advisor is very expensive. With LLMs, there is a unique opportunity to build software that gives every person access to personalized finance, investment, and tax advice at near-zero cost. They would use API's to access your complete financial situation and give completely personal unbiased advice.