A few things I’ve learned from Build

What launching 500+ projects has taught me about accelerators, hackathons, and the future of supporting builders

For almost a year we have been running increasingly large programs where we help builders launch something new. What started with “Builder week” with 10 builders has become “Build” which is currently in its 4th program with close to 2000 builders from all over the world. Build is now now powered by many of the coolest companies in tech — Google Cloud, Coinbase, Pinecone, Optimism, and dozens more.

“It’s still early” as they say, and we still have a lot to improve, but some pretty amazing things have happened already. We’ve helped launch more than 500 new products and distribute well over half a million in no-strings attached grants. Build projects have joined prestigious accelerators, raised millions of dollars in venture money, and grown to tens of thousands of users. Perhaps most importantly, we’ve helped thousands of people get that crazy idea out of their head and into the world.

I’ve been writing about new accelerator models and other systems for innovation for many years, so experimenting with one at a reasonable scale is a dream come true. Perhaps unsurprisingly though, the traction has only made me more impatient to improve things and move forward. So in part to force myself to reflect, here are a few learnings that stand out to me so far:

• It is very difficult to predict where quality projects will come from

• Text based applications are dead

• Enforcing short cycle times is magical

• One-off hackathons miss a huge opportunity

• Changing the business model changes everything

It is very difficult to predict where quality projects will come from

We score the thousands of applications we get for every program from 1-5, and while perhaps over time we will be able to build a better algorithm, so far we are quite bad at predicting which projects will get grants, raise money, or be finalists at the end of the program. Not only that, but I would have expected that the success outcomes would skew more towards the US and Europe than the program itself does. The reverse is true — 2/3 of participants are from the US and Europe but looking at the share of finalists or grant winners that number is closer to 1/2.

There are lots of potential explanations for the above which have to do with our own systems and biases. And of course there are still some projects that you can reliably predict will be successful — we’ve had staff engineers at big tech companies join Build to launch their startups or side projects and those will predictably have at least some baseline quality/success. But the point is that as the cost of building goes down, the long tail of builders is getting longer, and picking big winners upfront is going to get insanely difficult.

Text based applications are dead

We can’t get precise numbers, but a large % of people are clearly using ChatGPT or other LLMs to write their applications/project updates. In our case it’s no problem because the whole point is that we don’t select too aggressively on the front end of the program (we select on the other side based on what people build) but any system that relies on text-based applications to allocate scarce resources is going to have to change radically. Not to say all of those changes will be for the worse, but what a lot of applications test for is actually just the willingness to fill them out with detail. That skill is no longer testable via text-based apps.

I actually find a lot of room for optimism in this because the reason people are using LLMs in Build is mostly because they don’t speak English natively. I had an interaction the other day where I finally met someone virtually I knew via their application and project updates (which were awesome). It was only in talking with them that I realized they hardly speak english. It was a cool moment, the realization that previously they would not have been able to succeed in a program like ours where English is a requirement. Of course we — like so many others — will need to grapple with separating signal from noise as AIs can do more and more (including building the products) but I’m excited for that challenge because it comes with a much more level playing field.

Enforcing short cycle times is magical

As Visa founder Dee Hock said of the aggressive deadlines he gave to the tech team, “if you give computer people more time, they will just consume it.” Of course, that’s where the entire agile method comes from, but the issue that anyone who has worked in a tech company will tell you is that agile sprints easily spill from one cycle to the next. It’s rare to find companies that reliably ship products on time.

A shocking number of people have told me that the most useful part of Build is just having the forcing function to ship something within the 4 weeks, even if it requires cutting scope aggressively. That’s my bet on why 60-80% of people come back from one Build program to the next — it’s low overhead, and it’s a forcing function to ship something on time.

One-off hackathons miss a huge opportunity

One of the builders that has been in every program described Build as “the platform-ification of hackathons” and there’s certainly some truth to that. What I’ve become quite opinionated about is that a big issue with hackathons is that they are almost always one-off things. Because they are so expensive and high-overhead to run, they are either done as big-bang marketing events for companies or as sub-events within conferences. Either way, there is rarely any continuity between them.

The data we have so far suggests that builders love coming back to a similar program over and over again to make progress, and as a result the value for a given partner company within the program increases with time as well. It makes sense — the cycle time from awareness of a new piece of tech, to adoption, to actually building something useful takes months. One of the stats we look at is the number of successful connections we make between builders and companies, and the growth of that number is more than 2X the growth of the size of the program itself. As Build grows, the density of the network between partners and builders grows as well.

Changing the business model changes everything

A lot of people ask how Build is different from traditional accelerators like Y Combinator or Techstars or any of the dozens of others that operate in a similar way. I am increasingly excited by how much the business models of these kinds of programs drive everything about them. Any program that takes equity from projects as its core business model will behave in certain ways that we don’t because our business comes from infrastructure companies. If someone were to ask me for interesting ways to create “the future of Y Combinator” my answer would now be simple — look for other business model/incentive structures you can apply to a program focused on supporting builders and otherwise keep as much the same value as you can. Otherwise the only way to really compete with YC is just to focus on a smaller niche, and even that is pretty hard to do well. 

And by the way, the whole search for “the future of Y Combinator” is a bit of a wild goose chase because Y Combinator itself is not going anywhere anytime soon. The game of picking the very best founders out of a huge pool and anointing them with reputation in return for equity is by all measures a great game to play if you’re the best in the world at it. A better question is “what is the next Y Combinator type unlock for supporting builders at early stages, given how the process of building is changing?” That’s the one that keeps me up at night.

We still have a lot to learn, and some big iterations ahead as well. Build gets better as it gets bigger, and that means at a certain point if we are successful it will no longer look like an incubator “program” in any traditional sense. We have some ideas around that and I’ll keep sharing learnings as we have them, if I can remember to step back from the process of building it every once in a while! We’re certainly having fun with it.

If you think a lot about this stuff as well or want to work with us, reach out! And if you’re a builder, applications for Build v5 in late June are open — it’ll be our biggest and best yet. 

-Joey