Building with AI: Insights from Matt Cynamon and USV

October 30, 2024 4 minute read

I’ve written before about some key themes I see in building with AI: the importance of creativity and curiosity, the increasing accessibility of tools, and most importantly, the benefit of just getting started. These themes came alive recently while listening to Matt Cynamon from Union Square Ventures (USV) on the Every podcast. I highly encourage you to listen to the episode, even if you haven’t dug into any content in this space before. It’s highly approachable and you’ll come away excited to try something yourself. Below, I’ve summarized my key take-aways from the episode.

For those not familiar with USV, they’re a venture capital firm that has consistently been ahead of the curve in identifying and supporting transformative technology companies. I’ve had the privilege of working at two USV-backed companies, and I’ve always been impressed by their thoughtful approach to technology investment and community building. They’re also incredibly generous with sharing their insights - from Fred Wilson’s legendary 16-year daily blogging streak at AVC to their extensive collection of thought pieces at usv.com.

What struck me most about Matt’s discussion was his genuine excitement and energy around building in the AI space. The people I see making the biggest strides in learning about LLMs share this same mindset. Curiosity is a key ingredient.

The “Just Build It” Mindset

Matt shared a compelling analogy during the podcast about building with AI. You can start by building some software with AI, even if you don’t fully understand how what you’re building works. You can then use the technology itself to understand what you’ve built. He made an analogy to ‘taking apart a clock radio’ to see how the pieces work together. You can do the same with AI tools - build a thing, and then use LLMs to learn about what you’ve built. It creates this amazing flywheel of building, learning, and building again - a perfect environment for self-directed, curious learners.

The key ingredients aren’t a deep technical background - they’re time, patience, persistence, and getting started.

Evolution in AI Tool Design

One particularly interesting insight from Matt was his evolving perspective on tool design when using AI. Initially focused on building comprehensive solutions, he’s shifted towards favoring more specialized AI components (like custom GPTs) over monolithic applications. USV’s own experience reflects this - they eventually broke down their librarian tool – which is a way to help them surface historical USV knowledge – into specific GPTs, finding more value in targeted solutions.

This reminds me of a classic software development principle: do one thing, and do it well. It’s fascinating to see these established patterns emerging in AI development as well. Also, the speed with which you can actually build a thing with AI makes it an easy story to consider building very niche software tools, versus big ‘solve all the problems’ platforms.

The Human Element in AI Applications

A critical theme that emerged in the discussion was the importance of viewing AI as an assistant or partner rather than a replacement. Matt emphasized that tools should help with tasks rather than completely take them over - something that strongly resonates with my previous observations about effective AI integration. It really matters how the AI integrates with the human workflow, and that problem space deserves a lot of thought and energy. It’s probably the most exciting aspect of building with AI right now, with plenty of room for experimentation and innovation. Some recent UX patterns that have emerged (e.g. NotebookLM podcast-style summarization) seem almost obvious in hindsight, yet are incredibly valuable.

Building Internal Tools First

One of the most practical insights from the discussion was around the value of building internal tools first, and doing it ‘scrappily’ - with higher tolerance for inaccuracies. Matt shared that USV has focused primarily on internal applications, where there’s a higher tolerance for AI’s current limitations. If something is wrong in the output, you can build ways to work around it, and those limitations just become part of the ‘fabric’ of how you adopt the tools. Internal users can adapt to and work around these limitations in ways that external customers might not accept.

Form Factor and UX Innovations

The discussion about tool integration brought up some fascinating points. Despite the proliferation of specialized AI tools, Matt finds himself gravitating toward all-in-one solutions like ChatGPT or Claude. His reasoning? The benefits of having different ‘jobs to be done’ accessible as ‘menu options’ within the same application outweigh the advantages of specialized but separate tools.

The conversation around NotebookLM was particularly interesting, highlighting some innovative approaches for using it:

Using it to generate insightful questions from meeting notes
Better citation handling in meeting note summaries
Creating “meeting podcasts” as an alternative to traditional note sharing

These examples show how the form factor of AI tools can fundamentally change how we interact with information. It’s not just about what the AI can do, but how it fits into our existing workflows. For those of you who have yet to try NotebookLM, I highly encourage you to check it out. I wrote a bit about that here

Final Thoughts

What struck me most about this conversation was how it reinforced something I’ve been saying about building with AI: the limits are primarily in our creativity and willingness to experiment. Whether it’s Matt’s exploration of different tool formats, the Every team’s creative use of AI for Slack summaries, or the various experiments with meeting documentation, the common thread is a willingness to try new approaches.

The landscape of AI tools and applications is evolving rapidly, but the fundamental principles remain consistent: start building, focus on human-AI collaboration, and don’t be afraid to experiment with new forms of interaction. As I’ve found in my own explorations, the most valuable insights often come from just diving in and trying things out.