I came across two very contrasting posts this week from founders. One was from TailwindCSS, and the other from The Browser Company. The TailwindCSS founder shared how the company is struggling despite increased adoption, while The Browser Company shared how they are rethinking hiring due to Claude Code.

Both posts show the real-world impact of AI. On one hand, it highlights the fragility of the open-source ecosystem; on the other, it shows how designers and product managers can prototype much faster and achieve more.

AI is not fixing fragile open-source business models

I am an open-source enthusiast. Much to my own chagrin, I have made fewer contributions than I would have liked, but I have built part of my career around open-source tools like Terraform and Kubernetes.

The world of open source has been struggling with business models since before generative AI. Multiple projects like MongoDB, ElasticSearch, Redis, Grafana, Terraform, and others have changed their licenses from “true” open-source licenses to licenses more friendly to the commercial companies behind the projects; with mixed reactions by the community. Even before AI, usage and sustainability in open source were not always aligned.

The post from TailwindCSS shows a different form of struggle with the business model of open source. Because LLMs are so good at generating TailwindCSS code, the founder saw traffic to the documentation drop by 40% and revenue drop by 80%. The documentation for TailwindCSS is the main funnel into their commercial offering, and by seeing less traffic, their revenue dropped. This finally led to them having to let go of 75% of their software engineers. This is, however, counter to the actual adoption of TailwindCSS itself, which is still growing.

When AI can explain, generate, and contextualize results without sending anyone back to the source, usage and sustainability decouple. More people benefit from the tool itself and can create value, without that value necessarily flowing back to the people and companies behind it.

AI is accelerating experimentation

The job posting by The Browser Company paints a completely different picture. It shows AI as a huge lever on the productivity of both designers and engineers. Designers no longer have to wait for an engineer to build their prototype, while engineers can do more experimental work without sacrificing their core responsibilities.

What they are not saying is that they are launching vibe-coded features into the wild. In a reply to the post on X, the CEO shares that Claude Code is helping with prototypes and experiments, which don’t end up in the final product in that way.

This makes total sense, and it’s something I’ve heard in other places as well. Semi-technical designers can now use tools like Claude Code to build a working prototype instead of spending their time mocking up how a new feature would look. This helps both accelerate the iterative design process and makes it easier for other stakeholders to try out new design ideas.

Same Shift, Different Outcomes

I read both stories in the same week, and both stuck with me. Both companies are impacted by the same shift in technology, but the outcomes are completely different.

AI is increasing leverage at the individual level and can make individuals much more productive. However, the outcomes of that leverage might not be positive for everyone. In the case of TailwindCSS, it is clearly a negative. In the case of The Browser Company, it sounds like a positive.

This is why narratives like “AI is destroying everything” and “AI is pure upside” both feel incomplete. These statements require context and don’t apply to every situation.

I don’t have answers. But “AI increases productivity” is actually the least interesting part of these two stories. They raise a more important question: who gets paid when AI does the work? Until we have better answers, we will keep seeing cases where value creation rises while the people and companies behind them struggle to sustain themselves.

This blog was written by myself, with inspiration and review by Microsoft Copilot and ChatGPT. The heading image was generated using Adobe Express.