May 23, 2025 · 30m · y-combinator
How To Design Better AI Apps · Y Combinator
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of Y Combinator's 'The Breakdown,' General Partner Pete Koomen joins hosts Tom Blomfield and David Lieb to discuss why modern AI applications fail when constrained by legacy software paradigms and how developers can build true AI-native experiences through custom system prompts, background agent workflows, and unconstrained user interfaces.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The partners as informed peer 4.5 Guest teaching 2.9 Guest disagreement 2.0 The partners pushing back 2.0
speaking balance: gold is the partners, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 24:05 Pete defends daily prompt writing
Pete counters Tom's suggestion that users won't touch system prompts by citing real YC internal agent usage proving people will actively write and modify prompts daily.
Hardest push from the partners ▶ 21:00 Tom rejects raw system prompt writingTom explicitly challenges Pete's model by comparing manual prompt drafting to expecting an executive to write a 30-page manual for a new assistant instead of iterative onboarding.
Biggest teaching moment ▶ 5:29 Pete deconstructs developer gatekeepingPete breaks down why modern AI interfaces fail, showing that engineers are inappropriately applying decades-old code-hiding paradigms to natural-language system prompts.
The partners hold their own ▶ 11:49 David's technological wave breakdownDavid enriches Pete's thesis by citing concrete historical precedent, detailing how early internet directories and early mobile web-wrappers took years to find their native paradigms.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The partners as informed peer | Guest teaching | Guest disagreement | The partners pushing back | Why |
|---|---|---|---|---|---|---|
| Welcoming Pete Koomen and Introducing 'AI Horseless Carriages' | 3 | 2 | 2 | 1 | The hosts introduce Pete Koeman and discuss his viral essay critiquing corporate AI agent design. Pete explains the contrast between powerful developer tools like Cursor and frustrating, repetitive integrations like Gmail's Gemini drafting tool. | |
| Ideal AI Workflows vs. Defensive Corporate Software Design | 5 | 3 | 2 | 1 | The panel discusses why large tech companies ship bland AI tools, citing risk aversion, PR concerns, and legacy development mentalities. The hosts actively contribute context regarding Google's early transformer models and corporate caution. | |
| Custom System Prompts: Personalizing AI Output Tone | 3 | 4 | 2 | 1 | Pete demonstrates how personalizing a system prompt with user context produces natural email responses rather than lowest-common-denominator corporate prose. He explains how developers traditionally gatekeep software logic. | |
| The 'AI Horseless Carriage' Analogy and Technological Shifts | 6 | 3 | 1 | 1 | David Lieb draws historical parallels between horseless carriages and early tech paradigms, like digitized Yellow Pages directories and mobile web-wrapper apps. Pete showcases his inbox automation agent using natural language rules. | |
| Interactive Demo, Vibe Coding, and the 'Cursor Moment' | 5 | 3 | 2 | 1 | The conversation shifts to vibe coding and why developer tools are outpacing domain-specific tools in legal and accounting. Pete attributes this to dev tools granting raw access without patronizing guardrails. | |
| Will Everyday Users Write System Prompts? | 4 | 2 | 3 | 5 | Tom Blomfield and David Lieb push back on Pete's assumption that everyday users will comfortably write system prompts, citing non-technical users like their parents. Pete defends his view by comparing prompt writing to general computer literacy. | |
| Iterative System Prompting and AI-Assisted Configuration | 5 | 3 | 3 | 5 | Tom Blomfield argues for a layer of abstraction where AI auto-updates system prompts based on iterative feedback rather than users raw-dogging text files. Pete agrees with tooling improvements but maintains prompt visibility is essential. | |
| Moving Beyond Chatbots: Tool Use, MCP, and True Automation | 5 | 3 | 1 | 1 | The panel explores tool use, MCP servers, and transitioning away from conversational chatbot interfaces toward background automation that completes real-world tasks across multiple apps. |
Statements from this episode (23)
Koomen: Cursor and Windsurf are the most powerful software tools ever used
“When I use tools like Cursor and Windsurf to build software with AI, it feels like the most powerful tool I've ever used, right? It's this feeling of being able to create anything I want, anything I can picture in my head you know, will materialize in front of…”
Koomen: Gemini is incredible, but Gmail's UI makes it frustrating to use
“Gemini is amazing. The model itself is absolutely incredible. We've been using it a lot at YC to automate our own work. I'm so impressed with what they built. But a lot of that power I think is hidden behind a UI that makes it really frustrating to use.”
Pete Koomen: Gmail's AI drafter sounds unnatural and saves no effort
“There's two big problems with this. The first problem is like you're pointing out that doesn't sound anything like me, right? If you got this email, you wouldn't, you would assume somebody else wrote this. My account got hacked. I got phished. And then the sec…”
Koomen: AI models can handle complex, multi-step scheduling workflows
“These AI models are now capable of doing things like that, where it's actually anticipating all of the things that will need to happen as a result of my daughter waking up with the flu and helping me do those things.”
Koomen: Legacy software mindsets prevent unlocking AI's full potential
“Part of the problem is that We're using old software development mentality, old software development techniques to build these features, and we're not actually taking full advantage of what AI can do.”
Blomfield: Google lost its transformer lead to OpenAI through excessive caution
“And so, I think, like, Google were really on the cutting edge of this transformer technology, and were like, leapfrogged by OpenAI, just because I was too cautious to put stuff out there.”
Koomen: Editable system prompts remove repetitive prompting for AI
“By editing this system prompt, I'm able to explain to the AI model how I write emails in general, so that I don't have to do it every single time.”
Koomen: AI developers mistakenly treat system prompts like legacy code
“My contention is that A lot of AI app developers, including the Gmail team in this case, are treating the system prompt the same way they've been treating code for decades.”
Lieb: Tech shifts take years to move past legacy software paradigms
“When mobile came out, the first mobile apps, or a lot of the first mobile apps, were basically just websites wrapped in a native app wrapper, but they didn't take advantage of any of the new technologies available on the mobile phone, like GPS, like multi-touc…”
Lieb: LLM technology is capable enough for non-programmers to build apps
“And this is like a great example of how the LLM technology is actually good enough to let non-programmers
Program these apps.”
Koomen: LLMs are like smart fresh grads needing specific instructions
“The way that I think of these models, a super smart, you know, fresh grad right out of college that can do anything pretty well, but it has no idea what to do, you know, and the missing step is giving me the ability to teach it how to do work that I don't want…”
Blomfield: Rebuilt Tumblr blog into custom software in one hour using AI
“I was on Tumblr and I migrated my blog to custom software I wrote in one hour on a train journey.”
Blomfield: Coding agents are far ahead of other general-purpose AI agents
“The coding agents especially just seems so far ahead of all the other general purpose agents right now.”
Koomen: Every profession will have its Cursor or Windsurf moment
“It's basically every profession I think will have its cursor moment or its windsurf moment.”
Koomen: Agents excel at instruction processing, not blank-slate generation
“These agents aren't good at writing from scratch. They're good at processing instructions and doing and turning them into some text output, right? For code, that's actually really useful, right? If I can describe in English what I'm looking for, An agent can o…”
Koomen: Developer AI tools lead by avoiding defensive corporate restrictions
“Developer tools are power tools, right? Developer tools by definition allow you to get to the bare metal, right? To get under the hood of what it, whatever it is you're working with. And so the teams that built these tools allowed me Full access to have this a…”
Koomen: Mainstream users will adopt AI prompting faster than personal computing
“And I think the same thing is going to happen with prompting. So I think this is actually going to happen a lot faster with prompting, because writing a prompt is a lot more accessible than, you know, operating your file system manager on, on your computer. Yo…”
Blomfield: Training AI assistants requires gradual feedback, not 30-page upfront instructions
“The idea that, take a human analogy, you hire a new employee, you hire an assistant, the idea that you're going to sit down and write out 30 pages of instructions of how you do every single thing in your life is possible, but none of us actually do it. You sor…”
Blomfield: In 5 years, 99% of people won't touch their system prompts
“But I, I'd suggest, I think in In five years, for 99% of people, they're not touching their system prompt, but the system prompt is custom to them.”
Koomen: Writing prompts will become part of daily workflow for many
“Some of the work that we've done at YC, building agents to automate some of the repetitive work that we and our teammates do on a regular basis, actually suggests to me that writing prompts is going to be part of the day-to-day flow for a lot of people.”
Blomfield: YC startup Den is building a 'Cursor for knowledge work'
“There's a company in the current batch called Den, basically building cursor for knowledge work. That's trying to chain together all of these different MCP servers, which basically a way to call tools for agents.”
Koomen: Y Combinator built internal AI agents to automate employee tasks
“We built an early version of this inside of YC and we're already seeing employees of YC automate parts of their job that are easy to automate.”
Blomfield: Embedding chatbots into software is almost always the wrong approach
“I kind of hate the kind of chatbot paradigm. Like it was the easiest It was the way to bring the most basic LLM experience to the whole world. You know, we had GPT before ChatGPT and ChatGPT brought it to the mainstream, but now every product is just embedding…”