Revisiting the DREAM

The developer adoption funnel during the age of LLMs

In his book about building developer-facing startups, Adam Frankl argues that developers adopt technology in a sporadic, non-linear fashion driven by the fear of obsolescence. He codifies the technology adoption for developers through a craft acronym, DREAM:

• Discovery → The non-stop pursuit of keeping up with the ever-changing technology landscape

• Research → The process by which a developer dives deeper into a topic they deem relevant to their quest for relevance

• Evaluation → The process by which a developer figures out if a solution or a set of solutions they’ve discovered can solve a problem

• Activation → The work a developer does actually to solve the problem they face

• Membership → The point at which a developer has adopted a solution to their problem, and they are now bought into the community surrounding the solution

Traditionally, the discovery, research, and evaluation loops have been done through blogs, newsletters, trusted peers/influencers, video content, meetups, conferences, documentation, tutorials, Stack Overflow, forums, and sandbox environments/free trials. These loops have been so prevalent that developer advocate is a sought-after, well-compensated role in tech, developer influencers are now some people’s full-time careers, and an entire analytics stack exists to measure conversion from content created for developers to product installs.

But LLMs have completely upended this process. Today, developers can, in their IDE (where they spend the majority of their day), ask an LLM to scour the web for the latest and greatest solutions to the problems they face.

Struggling with query performance for your user-facing dashboard? Ask the LLM how to speed up the queries. Need to manage state in your frontend? Ask the LLM for the most popular state management library for your framework of choice. Getting started with a new project, but getting bored with the same-old same-old? Ask the LLM for the up-and-coming framework in the languages you love.

Unfortunately for startups and builders of niche tools, LLMs standing between content and users means that the traditional channels to reach prospective users are becoming harder to leverage. The first shoe to drop has been the steep dropoff of StackOverflow traffic over the years.

Why would anyone choose to subject themselves to shitty code and another dev’s snarky comments about test driven development when you can ask an LLM to write a quick test in the shiny new testing framework you’ve decided to take for a spin?

So, what are startups and dev tool creators supposed to do now that LLMs are the arbiters of the choices developers make daily?

The obvious answer is that the brave new world of SEO for LLMs is becoming critically important. You have to make sure that the LLMs are aware of and surfacing your tool to your potential users. Any company not putting the same amount of effort/dollars/tracking/etc that used to be put into traditional SEO is missing a huge opportunity. And the window on this opportunity is likely closing quickly - it won’t be the wild west for long, as the LLM training improves and easy shortcuts get harder and harder.

Beyond that, the honest take here is that I’m still trying to figure it out. I’ve been interviewing people across the industry, from Hashicorp to MongoDB to dozens of startup founders. I’m running a few experiments with the products my team and I have built. In all this research that I’ve done, there are a few things that seem to be key:

1. Content is still king, but not in the same way as before. Here is a key set of differences

   1. Visual styles don’t matter anymore; developer content has always had a utilitarian beauty to it. From Stripe’s developer docs to HackerNews’ purposeful undesign feel, devs used to care about a certain kind of aesthetic, no more…

   2. Volume with moderate quality over craft. Unfortunately, it seems like the volume of information has more impact on a tool’s mentions across different user prompts than a masterfully written piece of content. That said, you can’t just blatantly throw out massive amounts of pure garbage content

2. Real, verifiable people (even the sellouts) build trust faster than robots

   1. People trust video content more than written content now. This seems to be both for passive entertainment (things that distract you from writing that next line of code, but you can pass off to your boss as learning) or highly technical tutorials, where you follow along with the content and pause occasionally

   2. Influencers are propping up left and right, hard-to-game community sites like HackerNews and Reddit remain popular discovery mechanics even in the wake of LLMs, since keeping up with AI still requires some community support, and what better mechanisms than the democratic vote of your peers?

This space is moving so fast, I’m finding it hard to keep up with everything. I’m going to continue to learn, and I’m hoping to share what I can here.

If you happen to know someone interested in exploring this topic, please share this post with them. If you’d like to participate in the conversation, join our RCL chat!

.