AI2 Incubator Eats Its Own Dog Food: The AI-Native Chapter
So here we are, packing up our old digs and moving into the shiny new space at Pier 70 in Seattle, complete with the launch of AI House—our new hub for AI experts, entrepreneurs, community leaders, and builders who are creating the next era of AI in Seattle (with a shoutout to our partners at the City of Seattle Office of Economic Development for making this happen). Sure, we could tell you this is just about getting better WiFi and more espresso macchiatos on tap (though both are true), but honestly, we're doing something much more interesting: fundamentally rewiring how we think about building and nurturing AI-native startups. And with our portfolio companies having raised nine figures in funding and achieved a collective 10-figure valuation, we figure we might know a thing or two about this whole startup thing. Spoiler alert: it turns out the secret ingredient was AI all along. Who could have seen that coming?
Well, actually, looking back at our track record, maybe we did see it coming. Let me tell you how we got here.
Our Journey So Far: From "That's Interesting" to "Holy Shit, They Were Right"
Looking back at our portfolio, we've had the privilege of backing some companies that basically said "hold my beer" to conventional wisdom and ended up being ridiculously ahead of their time. Take Xnor, for instance, which led the charge in efficient inference—a capability that seemed nice-to-have back then but has become absolutely critical now that we're living in an agentic AI world where your model better run efficiently or your users will revolt faster than you can say "token limit exceeded." Then there's Lexion, which emerged as an early pioneer in document understanding back when most people thought AI and legal documents belonged in the same sentence only if that sentence was "AI will never understand legal documents." They had the audacity to build on ELMo—the grandfather of language models that was invented right here at AI2—and basically paved the yellow brick road for today's document AI revolution. Meanwhile, WellSaid was busy breaking new ground in speech synthesis, creating AI voices so natural that they made us question whether we'd ever want to hear actual human voices again (jury's still out on that one).
More recently, we've been like startup fortune tellers, spotting the rise of AI coding tools, conversational programming, vertical LLM solutions, and specialized AI applications before they became the cool kids on the block. Hell, we were discussing the power and potential of LLMs long before ChatGPT made everyone suddenly realize that maybe this AI thing was actually going somewhere. Materia decided that technical accounting was ripe for an AI makeover, which honestly takes guts because accountants are not known for their enthusiastic embrace of newfangled technology. ChipStack took one look at chip design and thought, "You know what this needs? Cursor, but for hardware," because apparently even hardware engineers deserve to have conversations with their code. Casium decided that the legal industry needed a complete makeover and went full throttle on building a full-stack AI-powered law firm (because apparently disrupting one aspect of legal services wasn't ambitious enough), Yoodli pioneered AI-powered roleplay (and no, not that kind), Vercept built domain-specific foundation models for human-computer interaction, and Preemptive decided that health monitoring needed more AI and fewer clipboards.
Watching all these companies succeed by going AI-native from day one, we started to have an uncomfortable realization: we were great at spotting AI opportunities for others, but were we actually walking the walk ourselves? Sure, we could identify which founders should build AI-powered this or AI-enhanced that, but when it came to our own daily operations—customer discovery, deal sourcing, founder support—we were still doing things pretty much the same way incubators had been operating for the past decade. Talk about the cobbler's children having no shoes.
The Next Chapter: Time to Eat Our Own Dog Food
As we look forward, we're confronting a question that's both deeply philosophical and embarrassingly practical: How do we, as an incubator team alongside our founders, actually infuse AI into every aspect of our work? Not just the fun stuff like "let's build an AI that does X," but the nitty-gritty, everyday operations that make the difference between thriving and just surviving. Basically, how do we evolve to become AI-native ourselves, not just in the shiny products we help build for customers, but in how we actually get shit done every day?
The zeitgeist is pretty clear at this point, and it's not subtle: AI is eating white-collar work for breakfast, lunch, and dinner, then asking for seconds. A doctor who doesn't use AI is going to get absolutely schooled by one who does—it's not even a competition anymore, it's just sad. The same brutal logic applies to entrepreneurs and incubators—we're all white collar workers too, and we're not immune to AI's impact. Use AI or else NGMI, as the kids say. And speaking of not gonna make it, people are increasingly talking about the potential for one-person unicorns—the idea that a single founder with the right AI toolkit could eventually build what used to require entire engineering teams. It's still mostly theoretical, but the rapidly advancing capabilities of AI are making founders increasingly able to do more with less. Honestly? That's exactly the kind of founder we want to support. The ones who see AI not as a nice-to-have but as their secret weapon for building something massive without needing to hire half of Silicon Valley first. The question isn't whether AI will transform how we work (that ship has sailed, hit an iceberg, and been replaced by an AI-powered submarine), it's whether we'll lead that transformation or become the cautionary tale that future business school case studies use to explain what happens when you stick your head in the sand.
Being all-in with AI requires a radical rewiring of how we think and operate, and let me tell you, it feels about as comfortable as trying to pat your head and rub your stomach while riding a unicycle. In the early stages, this whole thing feels messy, counterintuitive, and frankly a bit hypocritical. Picture this: we might be frantically building a sophisticated meeting note assistant for our customers while we're still over here taking notes by hand with actual pens during our own customer discovery sessions like it's 1995. This contradiction isn't just embarrassing—it's endemic. It's equally true for founders and for incubators and accelerators like us, and honestly, it's a special kind of ironic that the people building the AI future are often the last to actually live in it.
The Real Bottlenecks (Hint: It's Not What You Think)
Here's what we've learned after years of watching startups succeed and fail in spectacular fashion: the biggest bottlenecks for early-stage startups aren't what most people think they are. Plot twist—it's not coding! Cursor, Replit, Devin, and Lovable have basically turned coding from a mystical art into something closer to having a really smart, slightly sassy conversation partner who happens to be fluent in Python. It's not designing mockups either—tools like Figma and the new wave of AI design assistants have made it so easy to create beautiful interfaces that we're running out of excuses for ugly products.
No, the real bottlenecks are hiding in plain sight, lurking in the earliest and most crucial stages of the startup journey: finding the right idea that doesn't suck, validating it without fooling yourself into thinking your mom's enthusiasm counts as market validation, and getting to actual seed-ready traction that makes investors think "shut up and take my money" instead of "bless their hearts, they're trying." This means getting genuinely good at customer discovery (not just talking to your friends), building real customer traction (not just people who say they'd "probably use it"), and successfully navigating fundraising without losing your soul or your sanity.
Here's our radical bet, and yes, we're going all-in on this one: all of these core startup challenges should be transformed with AI. Not just improved or optimized or made slightly less painful—completely transformed. We're talking about fundamentally changing the game, not just getting better at playing the old one. Now, you might wonder why these workflows haven't seen much AI transformation yet. Simple: founders are typically ignored as a profitable user segment for AI innovation because, let's be honest, they don't have big enterprise budgets to throw around. Precisely this is where an incubator like ours needs to step in and build tools that actually serve the people doing the hardest, most important work in the startup ecosystem.
Our AI-First Approach: Two Experimental Projects (Or: How We're Eating Our Own Dog Food)
We're not just talking about AI transformation like it's some distant future where robots do our laundry and write our emails (though honestly, that future sounds pretty great). We're building it, right now, with our own hands and probably way too much caffeine. We're developing AI agentic systems to tackle two critical areas that have been the bane of early-stage startups since the dawn of time: idea discovery and customer discovery support.
Project 1: AI-Powered Idea Discovery and Mapping (Because "Uber for X" Deserves Better)
Our first project focuses on systematically discovering and evaluating startup opportunities, and yes, we're fully aware of how meta this is—using AI to find better ideas for AI startups. It's like inception, but for entrepreneurs who've had way too much coffee. We're building agents that systematically prompt LLMs to generate candidate ideas using comprehensive combinations of techniques, pain points, customer segments, and industries. Think "Uber for X" but executed with the kind of systematic rigor that would make a McKinsey consultant weep tears of joy, at scale, and with genuine intellectual honesty about what might actually work.
We believe in the abundance of startup opportunities waiting to be discovered—and we're not talking about the same tired SaaS plays that get recycled every six months on Product Hunt. We're talking about the lesser-known pockets, the industries that don't get TechCrunch coverage but represent massive opportunities, the spaces traditionally neglected by prototypical 20-something founders who frankly don't know enough about unglamorous industries to realize they're sitting on goldmines. There are countless whitespaces in sectors that would never make it into a Stanford entrepreneurship class but could absolutely change the world (and make some serious money while doing it).
Each candidate idea then goes through a comprehensive vetting process that would make a PhD committee proud, involving deep research across multiple dimensions: market opportunity (is this actually big enough to matter?), go-to-market strategy (can we actually reach these people?), competitive landscape (who else is trying this and how badly are they failing?), risk assessment (what could go spectacularly wrong?), and more. The result is a detailed product and investment memo with an executive summary that doesn't sound like it was written by a committee and an attractiveness score that actually means something. Ideas scoring low get the boot; high and medium-scoring opportunities get the attention and resources they deserve.
Project 2: AI-Enhanced Customer Discovery (Or: How to Practice Your Pitch Without Traumatizing Real Humans)
Our second project addresses one of the most critical yet criminally under-supported aspects of early-stage startups: figuring out what customers actually want without accidentally building something nobody asked for. We're building a web-based portal for founders to sign up and manage their customer discovery processes, and yes, it's basically NotebookLM meets Granola AI, but specifically tailored to the unique needs of pre-seed founders who are one customer conversation away from either brilliance or a complete pivot.
Founders can systematically manage hypotheses around customer segments, problem definitions, ideal customer profiles, MVP scopes, and all the other variables that keep you up at night wondering if you're building something amazing or just an expensive hobby. The AI assists with ensuring comprehensive coverage and rigorous thinking, because let's face it, when you're in the thick of building something, it's easy to convince yourself that your assumptions are facts and your hunches are data.
But here's where it gets really interesting (and slightly weird): founders can run simulated conversations with AI taking on specific customer personas. Imagine practicing your pitch with a "grumpy technophobic VP of sales at a midsize SaaS company who's been burned by the last three software implementations and whose idea of cutting-edge technology is Excel macros," or validating your assumptions with a "cost-conscious CFO at a growing e-commerce startup who treats every dollar like it's their own money and responds to every demo with 'but how much does it cost?' before you've even explained what it does."
Want to test your healthcare startup idea? Try pitching to a "hospital administrator who still communicates via fax machine, thinks 'the cloud' sounds unreliable because what if it rains, and whose last interaction with new technology involved a two-hour training session on how to use the new printer." Building something for restaurants? Practice with a "family restaurant owner who just wants to serve good food without learning another app, has been pitched 'revolutionary' POS systems by seventeen different salespeople this month, and whose response to AI is 'can it help me find good line cooks who actually show up?'" It's like having access to every possible customer archetype without having to bribe real people with coffee and the promise that the conversation will "only take fifteen minutes" (it never does).
And once you've nailed your customer discovery and have real traction? Well, that's when the fun really begins. We're already sketching out how to reimagine the entire seed raise process with AI—from investor matching to pitch deck optimization to managing the fundraising pipeline. Because if we're going to transform how you discover and validate ideas, why would we stop there? Every sacred cow of startup-building is fair game for an AI-powered rethink.
Taking Our First Steps (With a Frankly Irresponsible Amount of AI Tools)
I'll be throwing a combination of cutting-edge AI tools at building these systems—Lovable, Claude Code, Cursor—and dedicating what I'm calling "a significant but carefully unspecified portion" of my time to making this vision reality. Let's just say it's somewhere between "definitely more than a weekend hobby" and "maybe I should have negotiated better work-life balance with my spouse" and leave it at that. In fact, as I write this very blog post with Claude, I literally have Claude Code running in my terminal working on one of these projects. The recursion is getting almost embarrassingly deep at this point.
This isn't just about building better tools or optimizing existing processes—it's about fundamentally changing how we discover, validate, and build startup ideas from the ground up. We're betting that the incubators and accelerators that survive and thrive in the next decade will be those that don't just help build AI-native products, but become AI-native organizations themselves. Or else: NGMI. The ones that figure out how to eat their own dog food without getting food poisoning, if you will.
Now, let's be clear about one thing: this is all experimental as hell. I'm planning to fail fast, iterate frequently, and listen obsessively to feedback from our founders—because they're the actual customers of these tools, not us. If a founder tells me the AI customer discovery simulator is giving them useless advice or the idea generation system is spitting out garbage, I want to know immediately so I can fix it or scrap it entirely. And honestly? If we build something that genuinely works and could help the broader startup community, I'm seriously considering open sourcing parts of it. The whole point is to make startup building better for everyone, not just hoard cool tools in our corner of Seattle.
Come Build the Future (Seriously, We Mean It)
So here we are, officially settled into our new home at Pier 70 with AI House up and running (and yes, the WiFi is excellent and the espresso macchiatos flow freely). But more importantly, we're not just opening a new office—we're opening a new chapter in how startup incubation actually works. We're committed to practicing what we preach, using AI to transform every aspect of how we discover opportunities, support founders, and build companies. And yes, we're fully aware of how experimental (some might say slightly insane) this sounds, but honestly, if you're not at least a little bit crazy in this business, you're probably not thinking big enough.
The future of startup building is AI-native from day one, not as an afterthought or a nice-to-have feature you bolt on after product-market fit. We're here to make sure that future starts today, with real tools, real founders, and real companies that prove this isn't just Silicon Valley hype-cycle nonsense but actually how the world is going to work.
Come build the future with us at AI House. It's going to be weird, wonderful, and hopefully the kind of work that gets you excited to wake up in the morning.
P.S. This blog post was written in collaboration with Claude Sonnet 4 through careful prompting but no manual edits—because if I'm going to preach about being AI-native, I better practice what I preach. The tone is deliberately playful, casual, and slightly humorous. If anything here made you frown, please blame Claude. If it made you smile, I'll take partial credit.