In case you missed our intro post, you can read about Machine Earnings, my new newsletter + community focused on AI’s impact on the business world, here.
Before we jump into today’s essay, some programming notes:
📊 AI Market Reports & Playbooks Are Coming → Starting this quarter, Pro members will get exclusive AI market reports breaking down key vendors, trends, and strategies. Executive members will also get monthly deep dives—ranging from enterprise AI adoption to regulatory risks.
💬 Private Community Launching This Month → We’re rolling out a Slack community for AI operators, investors, and enterprise buyers to discuss AI vendors, deal flow, and best practices in a signal-over-noise space. If you signed up on the form, check your email next week for details on how to join.
If you haven’t signed up for our waitlist, do so here!
Now, onto this week’s essay. 🚀
The AI Developer-Prosumer Flywheel: How AI Labs Get Users Through Developers
The playbook for launching new AI products is being rewritten in real-time. Coordinated PR blitzes feel like a bygone era, with more companies following Lulu Cheng Meservy’s manifesto of “going direct.”
Today's most successful AI products spread through a different pattern – one that starts with developer enthusiasm, gains momentum through tech media, and ultimately reaches mainstream professionals through a powerful combination of authentic discovery and strategic distribution. This new model is transforming how AI labs increase developer adoption and awareness, acquire users, and build sustainable businesses. You can see it at work in the most unexpected places – like my family's group chat.
A few weeks ago, my younger brother sent an unexpected message: "Have you tried Deepseek? Just downloaded it - pretty impressive."
This group chat is my litmus test for most AI products. I usually hop in to ask my mother and brother their thoughts on different products I’m playing around with, or new features from ChatGPT. (My personal favorite is asking my mother if she thinks videos from Sora are real.)
So when my younger brother posted about Deepseek I was pretty surprised.
My family's interest in Deepseek highlights a crucial shift in how AI products spread – particularly in how they reach what the tech industry calls 'prosumers.' These aren't your typical consumers or enterprise customers, but rather professionals who use software to solve work-related problems. While their technical aptitude and expectations follow regular consumers, their needs remain business-focused.
This market has exploded beyond its origins in the startup world – today, it includes virtually any knowledge worker using AI to enhance their productivity.
This evolution in the prosumer market coincides with the obsolescence of traditional customer acquisition strategies. Almost all customer acquisition strategies from the SaaS era of software—both in enterprise and consumer—are outdated now. Competing on paid ads on platforms like Google and Facebook is an expensive game to play. Relying on social media referrals for growth is hard when there’s no more authenticity in referrals and influencers are selling out for a few bucks. Even in enterprise, referrals are getting hard—any smart startup knows well networked people that can intro them to a potential customer. From the customer side there are only so many vendors you’re going to use; even if your best friend started a new competing vendor, the significant switching costs often outweigh close ties.
Every major technology shift comes with a change in consumer acquisition strategy. The App Economy and the growth of mobile devices led to a massive shift away from banner ads towards more attributable mobile and social media ads.
You can start to see the beginnings of that with the new AI economy. The easiest example is the evolution of “SEO,” or search engine optimization. As a first adopter prosumer, I can’t remember the last time I found something because of an SEO-optimized articles. Smart companies are shifting towards “answer engine optimized” content—making sure their products come up high on the list of referrals from AI LLMs and chatbots. There are even companies popping up to help others figure out AEO, a tell-tale sign of a growing industry trend.
For AI labs like Deepseek, OpenAI, Anthropic, etc, the customer acquisition strategy is doubly hard—they need to attract users to their consumer-facing products while also attracting developers & large businesses to serve as more stable sources of revenue. You’re basically building two customer acquisition strategies at the same time. Whether intentional or not, this has led to a new flywheel: the AI Developer-Prosumer Flywheel.
I’m generally very eager to hear what you think about this framework—feel free to email me at ian@machineearnings.com with your thoughts! If I get enough I’ll try to write a follow up post.
Phase 1: X Gets Flooded With News About A New AI Breakthrough
The first stage of this flywheel is on X, which has became a massive information source for most of the world.
Usually what happens is that there’s something really interesting/cool/exciting/groundbreaking that goes somewhat viral on X. In the Deepseek example, it was primarily the inaccurate information around Deepseek’s AI model cost $5.6 million to train, and that its reasoning model was on par or better than OpenAI’s O1, at a fraction of the cost. But the key part here is that most of the virality comes from genuine and authentic posts from developers. Everyone is now keenly aware that developers are the ones that are building any software ecosystem, and because of that smart observers closely track what’s trending with that group.
While it starts out with developers on X, it quickly expands to other audiences and other platforms. On the audience side, there’s usually a pick up of interest from VC’s, who need to look smart and that they’re ahead of the curve, and tech thought leaders, the newsletter and X thread creators. While playing around with new developer tools and API’s requires coding experience, that’s getting easier and it’s also getting easier to learn about AI progress without having to know coding at all. On the platform side, you primarily see a pickup of posts on Reddit and Hacker News, both places where developers spend a lot of time trying to learn about new developments.
This initial burst of developer enthusiasm on X lays the groundwork for broader media attention.
Phase 2: Trending On X Generates Media Interest
X is becoming a major information source for media publication. It always was, to be honest; there used to be journalists that would just write up stuff that was trending back in the SEO days, trying to tap into the curiosity around trending topics.
But all this got accelerated dramatically after Elon’s headfirst dive into politics. Between his posts and Trump’s history of using social media as an announcement platform, the mainstream press started to cite X as a source more and more. Now, if something’s trending massively on X and capturing the X zeitgeist, you can be sure that media outlets will pick it up. It might not be the CNN’s and Wall Street Journal’s of the world, but the TechCrunch’s and The Verge’s certainly will. Not to mention the hundreds of creators with their own audiences following suit too.
This media interest is usually a good thing. The articles talk about a) that this topic is trending and here’s why b) some general context about progress in AI and c) how you can play around with it yourself (that last part is critical).
The media articles usually pushes this information from the confines of the tech bubble into the general prosumer ecosystem. Knowledge workers and professionals all over the world are using AI tools to get better at their job and do their job faster. Usually when there’s AI news, a lot of that consumption comes from normie prosumers. They immediately go try the consumer version of product: whether its a chatbot like Deepseek or a new feature within ChatGPT, like Operator or Deep Research.
With mainstream media amplifying the buzz, the conversation naturally shifts toward wider prosumer adoption.
Phase 3: Mainstream Prosumers Jumpstart “Consumer” Product Growth
Not everyone is going to learn how to code or be salivating at the mouth to try out the newest AI tools. If mainstream, normal, people were to have that curiosity they’d just be in tech. Most people want to only learn about the cream of the crop—the top 1-5% of products that are actually making a difference. The assumption is, if its going viral on X and a publication is writing about it, its worth checking out.
When most of us think about ChatGPT or Claude or Deepseek’s mobile app, we think about it as the consumer product on top of their infrastructure and API’s. But most normal people think about these tools in the context of work; there arguably still hasn’t really been a breakout consumer AI product. While these products are designed for “mainstream” use, they’re still primarily adopted by professionals looking to improve their professional lives.
This phase usually lasts awhile and can be a bit self-fulfilling. It’s a bit like an avalanche—it starts off slow at the top of the mountain (a niche audience) and gets more momentum and speed as it expands. Deepseek is a great example of. While most of the coverage initially focused on the training costs and functionality, the story changed once Deepseek’s app hit the top spot of the App Store. Then, the Deepseek was story about a Chinese company building better AI than American companies that have raised billions. The bigger more prosumer and business interest there was in the story, the more media publications were eager to fill the information gap.
As prosumer interest grows, developers are inspired to innovate further—thus restarting the cycle.
Phase 4: Innovation Goes Back To Developers, & The Cycle Restarts
Developers quickly capitalize on the growing consumer interest in different technological innovations. The open-source nature of AI makes this easier too; companies can easily add new functionality and swap out models quickly. While brand new products are a bit more difficult to conceptualize and build (we still haven’t see much around Anthropic’s Computer Use API, though it came out months before OpenAI’s Operator), there’s an initial wave of developer adoption.
It requires some creativity and product sense, but is very doable. Perplexity adding Deepseek R1 is a great example. After seeing Deepseek hit the top of the App Store, the media narrative started to incorporate some of the geopolitical dynamics between the US and China, particularly around data privacy. Deepseek’s data centers are in China and theoretically all your data from Deepseek goes to China. What they do with it is largely unknown, its enough to spook the general public.
Perplexity capitalized by quickly using Deepseek’s open-source model and integrating it into American data servers so all the information stayed domestically. Then they added the model to its already robust list of AI models available for users.
First mover advantage is important here—you have to move quickly if you’re going to do something like this. But it also depends on your product. If you’re building an AI Agent for doctors or something, integrating Deepseek quickly doesn’t make as much sense. For Perplexity, there was a strong user case (their users are fast adopting prosumers), a strong business case (they launched it primarily to their Pro users that pay $20 a month), and they were able to solve a perceived problem (domestic vs Chinese data servers.
It feels like this is a relatively straightforward flywheel—go viral on Twitter, get some articles, and maybe your app can hit the number 1 slot in the App Store too! In reality its extremely difficult to pull this off. Why? It requires genuine interest and curiosity.
If you think you’re going to generate virality on X from your employees & investors posting about a new update, you’re wrong. Nowadays, the average consumer can sense when they’re being sold to—that extends to prosumers too. You might get some quick wins, but it’s a hard way to jumpstart this flywheel. You might think it makes sense to use paid social ads and a PR agency to make this process simpler. That might be true but that won’t have the same punch as genuine curiosity.
To me, the way to actually create and capitalize on this flywheel is to build really interesting stuff. Craft is becoming much more of a distinction for products and users are very savvy to well-designed products and work.
There are some interesting case studies I want to point out where the flywheel worked really well or didn’t work well at all:
OpenAI & ChatGPT’s Initial Launch: ChatGPT’s initial launch and GPT-4 is really the prime example of this. It generated a ton of hype on X, garnered media interest, captured consumer mindshare, and OpenAI was able to use that to catapult ChatGPT to the top of the App Store and millions of users. It worked so well because a) what they released was novel b) it was really impressive for developers and the average prosumer c) they had a massive first mover advantage. By comparison Anthropic and Google’s Gemini haven’t had the same impact at all. I think that’s because a lot of what they do doesn’t feel super fresh, but rather optimized versions of things that are already out there. Anthropic’s Claude 3.5 Sonnet is a much better writer and developer than OpenAI models, but Claude isn’t jump in the App Store rankings. Gemini was only able to tap into the flywheel when they released NotebookLM, which was fresh and innovative, but the original team left and it feels like the product has gotten copied enough that NotebookLM doesn’t feel critical. Deep Research was another innovation that probably should have gotten more attention but it was really hidden as a product, and with attention spans shortening nowadays, no one has the time to dig through dozens of Google pages to figure out where to use Deep Research (this was my main problem actually.)
Anthropic Computer Use API: This is a really interesting case study to me. Anthropic’s Computer Use API felt like a massive innovation and a push for Anthropic to take the lead on the AI Agent revolution. It came out months before OpenAI released anything and, while it was launched as an API for developers, it felt like this would inevitably be added to Claude.
But that never happened. Developers like Replit, Canva, and Asana have added it but Anthropic hasn’t released anything themselves. And these companies aren’t really marketing Anthropic’s tech (rightly so), they’re more focused on marketing the actual product that the API can unlock.
Anthropic seems like it just paused halfway through the flywheel, which is probably why there hasn’t been much media coverage about it and most of my normal friends haven’t heard of it. It’s a shame because it’s really a great product and deserves to get wider recognition. While it’s clear that Anthropic is building great stuff, unless they integrate it into their products, they won’t reap all the benefits of their innovations like OpenAI or Deepseek.
ChatGPT Pro & Operator + Deep Research: OpenAI’s $200-a-month tier is an interesting wrinkle in the flywheel. Operator and (especially) Deep Research both captured a lot of attention on social media and media publications but it hasn’t translated to ChatGPT rocketing up the App Store.
Obviously that’s because these products are locked behind the very high monthly paywall; even if users want to try it out, the price point is way too high for casual users. My guess is that this solves 2 problems for OpenAI: 1) they want to juice up the number of subscribers for $200/month tier. This makes sense—if they get to 50k users on that tier, they can hit a $100m run rate. For a business that is turning into a for-profit endeavor and is getting increasingly scruitnized on its revenue generation, this almost seems like a no brainer. 2) These AI Agents use cases are still pretty new. While OpenAI actually integrated it into their products, unlike Anthropic, there’s still a bit of trepidation around letting everyone use it en masse and opening the AI Agent floodgates. Safety is still a concern for AI labs, and no one wants to be the AI lab that launched Agents that did a bunch of bad stuff on the web. A smaller scale launch gives the flywheel a smaller impact but limits the damage while also improving the bottom line.
In the AI era, the Developer-Prosumer Flywheel isn’t just another marketing tactic. Unlike the old days, when consumer and enterprise software traveled down separate channels, AI is rewriting the rules. Here, genuine developer passion and real-world prosumer needs combine to create a self-sustaining cycle. By definition, a ‘prosumer’ is a professional who uses software to solve work-related challenges—blending consumer expectations with the rigor of business demands—and this market is exploding beyond its startup origins.
The most successful AI labs know that you don’t win by throwing money at ads. Instead, it’s about building products that naturally spark excitement and drive organic growth. As this flywheel gains momentum, developer enthusiasm becomes the leading indicator of mainstream adoption, blurring the lines between consumer and professional tools. In other words, the companies that thrive won’t be those with the deepest pockets, but those that can consistently earn their market through authentic, iterative innovation.
In conclusion, the Developer-Prosumer Flywheel isn’t simply a new distribution strategy—it’s a fundamental realignment of how breakthrough technologies penetrate the market. For AI labs, this means rethinking everything from product development to go-to-market strategy. Embrace this model, and you’re not just capturing attention; you’re reshaping the very landscape of technology adoption.