
28 segments available
In this episode of The Breakdown, Tom and Dave are joined by fellow YC General Partner Pete Koomen to lay out a new vision for how AI should actually work: not as a chatbot bolted onto legacy software, but as a customizable tool that helps people offload the work they don't want to do. From editable system prompts to agents that act more like collaborators, they dig into what it means to build AI-native software—and why the future belongs to products that let users teach machines how to think. Pete's essay, "AI Horseless Carriages": https://koomen.dev/essays/horseless-carriages Apply to Y Combinator: https://ycombinator.com/apply Work at a startup: https://workatastartup.com Chapters (Powered by https://chapterme.co/) - 0:00 – Intro 0:52 – Why AI apps are broken 2:39 – The problem with Google's AI App 4:00 – A better way to build AI apps 5:27 – The hidden system prompt 7:57 – What if you could access the system prompt? 9:40 – The developer-user divide in software 10:48 – The "horseless carriage" metaphor 13:35 – Email reading agent demo 14:34 – Everyone can be a prompt engineer 16:23 – Why coding agents feel magical 21:42 – Training AI like a human assistant 28:45 – The problem with chatbot interfaces 29:10 – Advice for founders
In this segment, Pete Koomen discusses the limitations of traditional software development techniques in leveraging AI's full potential. He compares AI to a 'rocket ship for the mind,' emphasizing the transformative capabilities of AI when users can program it using natural language.
"We're using old software development techniques to build these features and we're not actually taking full advantage of what AI can do. I think the promise of AI for many of us is that it allows us to..."
The hosts introduce Pete Koomen, a partner at Y Combinator and founder of Optimizely. They highlight his recent essay that critiques the current approach to building AI agents, setting the stage for a deeper discussion on AI applications.
"Welcome to another episode of The Breakdown. Today we're lucky to have Pete Kumman, our partner here at YC. Pete was the founder of Optimizely which built software to help companies AB test. Pete, wel..."
Pete shares his contrasting experiences with AI tools, feeling empowered when using advanced software like Cursor and Wind Surf, but frustrated with existing AI integrations in apps like Gmail. He illustrates how these tools can sometimes create more work instead of simplifying tasks.
"like cursor and wind surf 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..."
In this segment, Pete critiques the AI drafting agent in Gmail, which generates email drafts based on user prompts. He discusses the limitations of the AI's output, which often lacks personal tone and authenticity, leading to user frustration.
"phenomenal, right? 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. Um, but a lot of that ..."
Pete envisions an ideal AI experience where the assistant can autonomously manage tasks based on user context, such as checking calendars and composing appropriate emails. He argues that current AI models are capable of this level of functionality but are not being utilized effectively.
"over by like or he's been fished and I need to report fishing like something's going horribly wrong. accounts been hacked, right? There's two big problems with this. The first problem is, like you're ..."
Pete explains the concept of the 'system prompt' that guides AI behavior in applications like Gmail. He highlights the lack of user access to this prompt, which limits personalization and effectiveness in AI interactions.
"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 t..."
In this segment, Pete discusses how the generic nature of system prompts leads to AI outputs that do not reflect individual user styles. He emphasizes the need for personalized prompts to enhance the relevance and authenticity of AI-generated content.
"me. Why is that? Well, what's actually happening under the hood when I ask this Gmail agent for a draft is that this Gmail agent, this little UI, combines my prompt, which is the one we we just talked..."
Pete reflects on the cautious approach taken by companies like Google in developing AI features. He suggests that this caution may hinder innovation and the ability to fully utilize AI's capabilities, as seen in the limitations of the Gmail AI.
"actually built a little demo so that you can see the impact of a system prompt on the email draft, right? And so we combine our sort of supposed Gmail system prompt with my user prompt here asking for..."
Pete proposes a reimagined approach to system prompts, advocating for user-editable prompts that reflect individual user characteristics and preferences. This would allow AI to generate responses that are more aligned with the user's voice.
"to say anything that's going to embarrass Google. I I remember it was either Google or Facebook released like a scientific model like pretty early on and then very very quickly pulled it back because ..."
In this concluding segment, Pete emphasizes the importance of empowering users in AI development. By allowing users to define their own system prompts, AI can become a more effective and personalized tool, enhancing productivity and user satisfaction.
"like this in a way that looks a lot like software that we've been building for decades, right? And so just to illustrate what's possible, let's imagine that Gmail allowed me to not only see but edit t..."
Pete Koomen elaborates on the historical divide between developers and users in software design, emphasizing how this division has led to generic software solutions. He critiques the Gmail system prompt as a 'lowest common denominator' approach that fails to capture individual user needs. This segment underscores the necessity for a shift in how AI applications are built, advocating for a more collaborative approach that empowers users to define their own interactions with technology.
"did the Gmail team hide decide to hide this system prompt away? 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'..."
In this insightful segment, Pete Koomen introduces the 'horseless carriage' metaphor to illustrate the limitations of current AI applications. He compares early automobile designs to today's AI tools, suggesting that simply adding AI to existing software is insufficient. Koomen argues that true innovation requires a complete redesign of applications to leverage AI's capabilities fully, rather than merely replacing traditional components with AI functionalities.
"that developers today are using AI in the previous generation of software development that they've been used to? Yes. Um, and so you use this phrase, the AI horseless carriage. Maybe tell us what you ..."
Pete Koomen discusses the potential of AI to automate repetitive tasks, particularly in email management. He emphasizes that much of the time spent on emails is busy work that doesn't require full cognitive engagement. By using AI to handle these tasks, users can free up their mental resources for more important work, showcasing how AI can transform productivity in everyday tasks.
"the the deepest problem here is that when the when the Gmail team set out to build this, they kind of asked, "How can we slot AI into the Gmail application? How do we replace the horse and put an engi..."
In this segment, Pete Koomen presents a practical example of an email reading agent designed to streamline email management. He explains how users can program the agent to categorize and prioritize emails based on personal preferences. This demonstration illustrates the accessibility of AI programming for non-technical users, highlighting how intuitive it can be to teach AI to perform specific tasks that alleviate the burden of routine work.
"example. So this is an example of an email inbox. It's got a bunch of messages on the right hand side and there's an agent operating in this inbox and it's an instead of an email writing agent, it's a..."
Pete Koomen shares his experience with programming AI agents, emphasizing the intuitive nature of creating system prompts. He argues that while many people may not consider themselves technical, the process of teaching AI to perform tasks is straightforward and accessible. This segment encourages users to engage with AI technology, illustrating how they can effectively communicate their needs to AI models.
"thing I find interesting about this is this is really the code. This is the programming you are doing for this agent. But if you read it, it's pretty accessible, right? It says if it's a techreated em..."
In this segment, Pete Koomen reflects on the transformative power of AI in software development. He discusses how AI can enhance the coding process, allowing developers to create applications more efficiently. By sharing his own experiences with AI-assisted coding, he highlights the potential for AI to revolutionize various professions, enabling users to build tailored solutions that address specific needs.
"intuitive. And so, for example, here here's what happens when I apply this little agent to each one of the emails in this inbox. Amazing. I love this. Um, this is a this is an essay. Uh, but it's not ..."
Pete Koomen concludes by envisioning a future where AI agents are tailored for specific professions, such as accounting and law. He discusses the potential for these specialized agents to automate repetitive workflows, making professionals more efficient. This segment emphasizes the importance of adapting AI technology to meet the unique demands of different fields, paving the way for a more productive future.
"Yes. And to me uh like we will have caught up when everybody uh has the same experience that we have when we're using these coding agents in their particular domain. Right? So when accountants can bui..."
The hosts delve into the strengths of AI models in processing text and generating code. They explain how effective prompting can lead to remarkable outputs, showcasing the utility of AI in coding tasks. This segment highlights the importance of clear communication with AI tools to maximize their potential and effectiveness.
"accounting agents. I think there's two reasons. Um, the first is that these AI models are incredibly good at processing text, right? So, if I can write a good description of a thing I want, they can p..."
In this segment, the discussion focuses on the empowerment of developers through unrestricted access to AI tools. The hosts argue that allowing developers to fully utilize AI capabilities leads to more innovative and effective applications. They contrast this with the cautious approach taken in other domains, advocating for a shift towards greater freedom in AI interactions.
"by definition allow you to to get to the bare metal, right? to get under the hood of what whatever it is you're working with. And so the teams that built these these tools allowed me full access to ha..."
The conversation explores the shift in responsibility from developers to users in AI applications. The hosts discuss how giving users control over system prompts can lead to more personalized and effective AI interactions. This segment emphasizes the importance of user agency in shaping AI behavior and outcomes.
"I hope we're able to move beyond u this this moment in time where there's so much assumed liability on the part of the application developers for what these models do. Well, I think the interesting po..."
This segment addresses the evolving landscape of prompt engineering and its accessibility to the general public. The hosts draw parallels between the learning curve of using computers and the future of writing effective prompts for AI. They express optimism about the increasing intuitiveness of prompting, suggesting that it will soon become a common skill.
"Yeah. It's it's a shift in mentality. Uh giving giving AI to the user as a tool and they can use it for whatever their purposes are versus feeling like you as a developer have to be responsible for ev..."
The hosts discuss the gradual learning process involved in interacting with AI systems, likening it to training a new employee. They emphasize the importance of iterative feedback and adaptation in refining AI responses, suggesting that users will eventually develop a collaborative relationship with AI tools.
"remarkable or interesting for somebody to use to be able to use a computer right? And so what happened is we all just sort of figured out how to use these things and the interfaces got better, right? ..."
In this segment, the hosts highlight the iterative nature of developing AI applications. They share insights from their experiences at Y Combinator, where they have built tools that adapt to user feedback. This discussion underscores the potential for AI to evolve alongside user needs, creating a more effective and responsive development process.
"able to do this. I'm not sure everyone will want to or to have the initiative or agent like we're three founders sitting here. We love to tinker with this stuff. Like you're telling me about your mom ..."
The conversation shifts to the role of AI in automating repetitive workflows. The hosts discuss the potential for AI to serve as a collaborative partner in various professional settings, enhancing productivity and efficiency. This segment emphasizes the transformative impact of AI on traditional work processes.
"check the drafts anymore. Send it." And then if you spot something that's not quite right, you say, "Oh, no, no, no. I would have phrased that this way or this other way." and the human would take tha..."
The hosts conclude by discussing the future of AI customization, where systems will adapt to individual user preferences and workflows. They envision a scenario where users can rely on AI to understand their unique needs without needing to manually adjust system prompts. This segment highlights the potential for AI to become an integral part of everyday tasks.
"autoupdate. Absolutely. Right. And and these are great examples of um UI conventions that we just haven't figured out yet, right? And and you can see people are experimenting with different models her..."
The conversation shifts to the future of AI prompting, where the speakers envision a world where users won't need to manually edit system prompts. Instead, they will provide feedback in natural language, allowing AI to automatically adjust and optimize prompts. This segment highlights the potential for a higher level of abstraction in AI interactions, making it easier for users to teach AI without technical barriers.
"possible. It'll be interesting to see how this plays out. 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 basi..."
This segment discusses the concept of 'tools' that empower AI agents to perform tasks on behalf of users. The speakers illustrate how AI can automate mundane tasks, such as managing emails and handling repetitive work, thereby freeing users to focus on more meaningful activities. They highlight the transformative potential of AI tools in enhancing productivity and streamlining workflows across various applications.
"the prompting, right? And and Tom is kind of saying like there's a going to be a bunch of innovation on how we write system prompts or how we edit them or or iterate on them. In the essay, you also ta..."
In the final segment, the speakers address how founders should approach building AI-native applications. They stress the need to rethink existing tools and design new solutions that leverage AI to offload repetitive tasks. The discussion emphasizes that the future of AI is not just about embedding chatbots but creating innovative tools that fundamentally change how users interact with technology.
"we built a 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. And and the difference that these tools provid..."