Reflections on Taste
Juan Pablo Flores Cortés and Clement Hugbo examine why taste has become central to conversations about AI, design, and developer relations. They use the process behind Charli XCX’s Brat artwork and CodeRabbit’s community experiments to show how context and iteration shape judgment. They argue for explaining choices, learning from audiences, and keeping taste open to people beyond an established inner circle.
Key takeaways
- 🧭 Explain your choices Show the intent, constraints, and trade-offs behind work before asking people to judge the result.
- 🧪 Use rough drafts to learn Share experiments with a community, gather feedback, and improve before scaling the format.
- 🤝 Build with your audience Let community knowledge and collaborators shape the work instead of imposing one taste.
- 🚪 Keep taste inclusive Question whose standards define good work and whose voices those standards may exclude.
Transcript
Why taste matters now
Juan Pablo Flores Cortés: Hello. Hello. Is it working? Okay. Great.
Hey, everyone. Thank you so much for joining us for this conversation. Earlier this year, I was talking with John about how much DevRel has changed, and one of the biggest changes we noticed is how often people now mention taste. If you have attended other talks at this conference, you have probably heard it too.
In conversations about technology, developers, and our work, it is increasingly common to hear the word “taste.” We wanted to understand why tech has become so obsessed with it. A New Yorker article even asked why tech people were so focused on taste, describing gatherings in places like Napa Valley where people discuss technology over wine.
Those gatherings are one way to offer a different experience. But when we looked more closely, we found no common definition of taste. For some, it is an aesthetic associated with Silicon Valley; for others, it is a human quality that AI cannot replace.
Others claim their taste is better than yours, or that taste is the only thing that matters when building. But what does it mean for products and DevRel? In San Francisco, people talk about taste while billboards celebrate 10x engineers and promise that everything can be automated. There is a gap to bridge.
How AI changed the pace of making
Clement Hugbo: As Juan said, taste is everywhere. Why has it become so relevant now?
In tech communities in San Francisco, New York, and elsewhere, taste has become a buzzword. But people often judge the finished work without discussing who made it or the context behind it. We want to talk about that context.
We will also talk about how taste shapes developer communication and offer a practical framework for DevRel. Before AI changed the pace of production, launches were less frequent. Teams had more time to consider the use case, the solution, and the community they hoped to reach. Three factors mattered: the scarcity of launches, control over the story, and polish.
Teams thought carefully about the story they wanted to tell and the intention behind it. Along with polish, that work helped people trust the product. Since 2024, agents, templates, demos, and AI-generated images have made it much easier to create and publish. Viral work can be copied quickly, often without acknowledging the work and decisions behind it.
When everything can be pushed out quickly, we can lose the patience to ask what we are trying to communicate. Design, art, and documentation may be published without the thought and feeling behind them. That is the tension we want to explore.
From craft to judgment
Juan Pablo Flores Cortés: The other piece that was important during the research was that we noticed that most of the conversations around taste before early 2024 were more about craft, and then craft evolved into taste. And it all starts with Claude 3.5 Sonnet, specifically because Claude 3.5 Sonnet was the first model that allowed these agentic workflows. After that, we start to see agents taking larger pieces of work, larger pieces of features. Tools such as Amp and Continue, and community workflows such as Plan and Act, let people run agents on larger features with less direct attention to the code. So if you compare it with what GitHub Copilot was in 2021, 2022, it was essentially more of an autocomplete.
In 2025, the conversation shifted. Early in the year, some developers said you no longer needed to read the code: just ask an agent to do the work. By the end of the year, people were asking whether they still needed to understand the code they shipped. That shift created a bottleneck, especially for DevRel. If you've worked in smaller teams, smaller startups, the execution or the work that engineering was doing stopped becoming a blocker, and the blocker became the marketing team, and especially how do we go and present these features to the developers.
This is what Developer Relations is good at: talking with developers and finding ways to help them use a product. But there's a couple of things that change with this perception on DevRel, specifically with some of the startups that we've been working with. The three things are, number one, there's a very strong pressure right now for DevRel to do viral content, specifically on X. X came back to become kind the platform where everyone is sharing.
The pressure we hear from companies is: we have a new feature, and DevRel needs to get it in front of developers and make it go viral. The second request is for stronger, more on-demand communities. How can we create a community that helps new work spread and gain adoption?
A trusted community can also help keep developers engaged when switching AI tools is so easy. And it can create genuine excitement. Before Cline and CodeRabbit, I worked at GitHub on the Campus Experts program.
One of those Campus Experts was here yesterday. When we built the program, we spent a lot of time asking what would be useful to students and what would become part of their identity. During the pandemic, we paid close attention to the details. Evan from BrandMakers sent us many backpack options while we worked out which one was right for the students.
Even after graduation, students still carry those backpacks. The program boxes, which we thought might have little practical use, appear in the backgrounds of calls to show a connection with GitHub. Some graduates also display their flags. That attention to detail mattered, and Evan did excellent work finding suppliers. It made us ask whether each component was practical and meaningful to the students.
Some of these things align with taste. To make a judgment, you need to compare alternatives, consider feedback from the people who will use the result, and explain your trade-offs. Taste is contextual: what works depends on the audience, community, and geography.
A little bit related to what we mentioned before with the practicality, but this is more about how did you explain the decisions that you have taken. When AI obscures those choices and trade-offs, it can produce what people call AI slop.
The choices behind the work
Clement Hugbo: Juan has talked about DevRel; I come from a design background. Product designers often judge work publicly: “This design doesn't look good” or “This button doesn't look good.”
This feels unreal. This feels like AI slop and stuff like that. There's been a lot of context around what looks good and what doesn't look good. And while Juan and I were talking about this, we thought about this brand, for example. How many of you are familiar with Brat Summer?
Now, look at the cover of Charli XCX’s Brat. It appears low resolution. If it were released today, when AI can produce polished visuals so quickly, some people might call it lazy or say they could do better.
The Brat cover was designed by Brent David Freaney. At Config 2026, he showed how it was created. Designers and developer experience teams often do a lot of research and iteration that the audience never sees. Unless you explain the process, people may not understand how you arrived at the result. Would Brat have worked with a different cover? Maybe.
It is hard to answer without seeing the process. At Config, Freaney showed the decisions behind the cover in Figma: the type, color, culture, and audience. The final image may look low resolution, but that appearance does not show the work that went into it.
The cover worked because the choices around it fit together: the music, the identity, the moment, and the community. Charli XCX and her audience made the idea of Brat resonate. Taste is the relationship among choices and the story that makes them legible. Without that story, it is hard to explain the purpose behind a solution. For us, taste communicates intent and connects the work with the people it is for.
Experiment with your community
Juan Pablo Flores Cortés: Good partnerships also show that taste is not only our own judgment; it can come from communities and collaborators. We have given creators room to work with our brand in their own way. One successful example is Merge Cop, a video series made with ThePrimeagen about a cop who hunts bad code in code reviews.
Merge Cop became one of the creator’s most popular partnered shows. Another creator made an eighties-style training video about code reviews. It sounds strange, but it resonated with that creator’s audience. This leads to a more controversial point: slop has a place, though not in the way it is usually published. There is a useful kind of slop: low-fidelity experiments that let us try things out.
The harmful kind is digital contamination: publishing an unedited AI output as your own work and leaving the audience to work out what you meant. At CodeRabbit, we could test a different approach. We can benchmark how models perform at code review, a different task from generating code.
A model’s training and available tools can affect how it reviews code. We decided to publish model reviews and see what the community thought. The first post was an experiment.
It was nothing fancy, just a few graphs comparing Claude Opus 4.8 with previous models to see whether this kind of code-review content was useful. I shared it on Reddit to get frank feedback. Our Discord community was supportive, but people on Reddit were more willing to point out what could be better.
We listened, iterated, and published more reviews. They became some of the most-read posts on our blog. By the time Fable was released, people already knew that we compared models on code-review tasks.
Hendrik made a video soon after the model was released. The review and video were picked up by news outlets, and the video became one of our most viewed. It grew out of a rough first experiment and repeated feedback.
After our Fable post, competitors began publishing similar content. The format became crowded, so we need to change what we do next. You cannot simply tell a DevRel team to “do taste” and bring in developers. Taste is built through habit and accumulated work.
Who gets to define taste?
One last point we cannot fully explore here is that taste comes with power. Someone decides what counts as tasteful. Taste can celebrate good work, but it can also exclude voices that do not fit a group's standards. The relationship among power, capital, and taste deserves a longer discussion; I would be happy to continue it in the hallway.
Awesome. Thank you all.
Clement Hugbo: Thank you.