The empirical study has drawn interest from a16z's Martín Casado.
Ohh excited to read through this.
🤖Excited to share a new working paper.🤖
The phrase "AI-Native firms" it everywhere, but are they any different? Is it just hype?
In our new paper we show AI ventures are organized differently, but not for the reasons you think.
Users appreciate the new paper and thread on AI-native firms organizing differently than expected because it makes their own work potentially easier.
No Digg Deeper questions have been answered for this story yet.
We argue that building a firm around AI isn't just about the *process* changes (e.g. using Cursor) but also involves a *product* channel whereby firms embed AI into what they sell which then changes how the firm should be organized.
Taken together, our findings suggest something bigger than workers simply becoming more productive.
Along with recent arguments by @martin_casado and @abhishekn and the work of @professor_ajay @joshgans and @avicgoldfarb, the evidence increasingly points to AI changing the firm’s underlying production function.
And like when past GPTs changed production, we should expect (radically) new organizational forms.
🤖Excited to share a new working paper.🤖
The phrase "AI-Native firms" it everywhere, but are they any different? Is it just hype?
In our new paper we show AI ventures are organized differently, but not for the reasons you think.
The results above are for YC startups, but this isn't a YC story. We show the same patterns hold when we we analyze the universe of US startups that raised funding over the same time period in PitchBook.
To shed light on these questions, we link data on all startups in recent @ycombinator batches and the universe of US ventures that raised between 2020-24 to the amazing data from @RevelioLabs capturing who works at these firms.
To measure if a startup is AI-Native we use the fact that YC startups can apply the tag "AI" to their public YC startup directory profile.
Without AI tools like Cursor, Gamma would need to hire more coders for sure. But without an AI product, they would need 10,000+ deck creators and a BPO business model, or a marketplace with as many freelancers. The startup simply couldn't exist as organized.
It was not only that workers used "railroads" or "motors" in their day-to-day work, but that the economics of building firms around these new technologies required new types of organizations.
Indeed, the first Org Chart (1854) stems form the need to manage complex rail lines (though they put the CEO at the bottom not the top!)
They have hierarchies that are half a layer flatter on average (echoing ideas in Ide and @EduardTalamas work building off of @lugaricano https://www.journals.uchicago.edu/doi/10.1086/737233)
This doesn't mean using AI for your work or the firm's internal processes don't matter!
Applying AI to work processes may simply require more managerial imagination or better tools like Claude Codex / Codex, something you can read about in our other working paper exploring AI's impact on entrepreneurship:
AI startups start smaller and stay smaller.
Writing the @HarvardHBS case on @GammaApp and @thisisgrantlee helped @hyunjinvkim and me see this difference.
Gamma uses AI tools in engineering, marketing, customer service, ... to move fast with a lean team. We refer to this channel as the *process* channel and its the focus of most AI research.
Within YC, AI startups are smaller than non-AI peers.
AI startups are not just scaled down versions of large firms. They have a greater shares of engineers, more folks in senior roles, and fewer managers.
Our argument echoes the work of economic historians from Alfred Chandler to Paul David.
Railroads and electricity enabled scale and distribution that in turn required new ways of organizing in order to benefit from the technology.
Caveat Emptor: This doesn't mean there are fewer jobs!!! We see a huge surge of entry (something folks like @JHaltiwanger_UM and others are seeing too) and so unclear what the labor market impacts are.
The firms are smaller, there could be many more of them!
Final result!
We find the differences between AI-native and non-AI startups are largest in startups working in services industries like education and healthcare.
Why? These are exactly the businesses that historically scaled by hiring large numbers of knowledge workers (tutors, therapists, ...)
But Gamma has also built a *product* around AI. Each day Gamma's 100m users are delegating to millions of AI "analysts", "researchers," and "designers" that build the users' decks.
By building around an AI product Gamma scales knowledge work with many fewer knowledge workers.
You can find the paper w/ @hyunjinvkim here:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6905079
@orgRem @hyunjinvkim @martin_casado @abhishekn @professor_ajay @joshgans @avicgoldfarb Thanks for the extensive thread! Makes my work potentially easier!
While you might worry YC's AI tag is just cheap talk, but we find robust evidence that AI startups are organized differently based on the talent they employ and manage. These firms are smaller, flatter, and more expert dense.
