Contemplating Hype

"Toto, I don't think we're in Kansas anymore."

It feels like highly capable, broadly available LLMs have changed the world. But has anything really changed? Not yet. We're at the start of a journey into a new technicolor world, a journey that will come with its own surprises and challenges. Our job is to navigate this yellow brick road. But unlike Dorothy, we're not just following a path. We're helping to shape it. The choices we all make will shape this new land of Oz even as it shapes us in turn.

Metaphors aside, what does that mean for us at Sourcegraph? Our mission is to make it so everyone can code. Our current point on that journey is to win devs’ hearts and minds by solving their Big Code pains. (And get paid by their companies for sustained, business-critical usage. We need to build a sustainable business to advance that long term mission!)

As stevey notes in Cheating is All You Need, that mission feels closer than ever, but we still have to figure out how to get there. Our strategy is still sound. Winning over the hearts and minds of developers who have the biggest, gnarliest coding pains still sets us up to make coding easier for everyone. However, the advent of LLMs that are good enough to apply to real world problems is going to impact our execution plan for getting there.

But why do we have to move so quickly? Why can't we morph our strategy slowly and steadily, finishing up what we've started and then adding new work as we go? Right now we're in a period of a million dandelion seeds flying into the air, most of which will fail to take root. Are we just setting up ourselves to be one of those failed seeds? Can’t we wait to see where the value is?

You might also be thinking about our own past. Folks who have been at Sourcegraph longer than I have (pretty much everyone 😝) know that moving past MVP/beta/experimental to a polished feature that feels integrated into our product has not exactly been our strength. Are we falling into the hype cycle? Are we turning back to our old habits after promising ourselves we were going to do better by focusing on quality and reliability?

Here is my belief: yes, this space is hyped up at this point. But there is something real here. If even 5% of what is being explored comes to pass, then it will change the way people produce and consume information in a way that compares to the advent of the world wide web, the television, the radio, maybe even the printing press.

Okay... didn't we hear that last year? Just to briefly address the elephant in the room — hi crypto! — a key difference between last year's hype darling and LLMs is that block chain based innovations were primarily system solutions. A distributed read only ledger is primarily useful for solving coordination problems in low trust environments. For personal ledger problems or high-enough trust groups, there are simpler solutions. That means that the problems that block chains are poised to solve generally require broad buy-in before the solution is useful. An alternate financial system isn't useful if you can't actually buy anything with it. Tracking property ownership is only useful if people respect your claim. Understanding the provenance of your goods only works if the whole supply chain is bought in. And verified ownership of digital goods only matters if enough people care about who verifiably owns something you can right click.

Generative AI is different in that it's much more conducive to solving single player problems. Can this make me more efficient at learning? At writing? At coding? At problem solving? If it can, then that's good enough. To put it another way, delivering 5% of the hype on a global coordination problem is equivalent to getting nothing. You don't reach the activation energy to make it valuable. Delivering 5% of the hype on single player problems is value that is now on the table. Generative AI will likely lead to system solutions, but the start of our road will be paved with solutions people can adopt independently.

Why now? One side of hype is that 95% of what's getting released right now is going to fail. The other side of hype is that whoever establishes themselves now is going to have a lead that is hard to overcome. This is a period in which the successful products, the ones that actually bring value to people's lives, are going to get the sort of recognition that gives them a durable advantage. We want to be part of that pack because the alternative is playing a catch up game where we're at a huge disadvantage. A big company might be able to invest in that catch up game until they win. However, a small company like us? There are big changes coming to the world of software development, and if we don't make it on the first boat we may not be relevant long enough to catch the next one.

P.S. If you want to learn more about what Sourcegraph is doing in this space, check out our AI coding assistant, Cody.