Takeaways from Converge 2025

Madelin kicked off the conference with some big claims: docs are dead, AI is going to ruin everything, and design systems aren’t doing what we’d hoped. Thankfully, she went on to explain how things are much more nuanced than these blanket statements, giving a more positive outlook on the current state of the industry.

Elyse’s vision of the design system of the future built upon the points made by Madelin. We need to go past components and infrastructure to provide “design intelligence”. This means using tools like Dessn.ai to prototype using the codebase, or an AI-powered PR bot given design system context. This way we can detect using the code when teams are diverging, surface evolving patterns, and enforce guidelines at implementation level, removing the need for heavy design control and policing.

Fundamentally this flips the current design system workflow – it’s no longer design to code but code to design. The real source of truth becomes the final product, not the design system. The design system of the future stops focussing on all the things the design system could possibly have and looks instead at what it should be bringing: speed, efficiency, cohesion.

I noted down a couple of tips Guy gave in his talk that seem useful to any team.

- When you’re feeling doubt, try a pre-mortem – Imagine you’re in the future and the project has failed. Why? - Use consumers’ needs as minimum requirements – a simple filter that keeps the team focused on actual problems rather than theoretical perfection.

Both Guy Segal and Andy Bell gave great talks making the same point from different angles: better code starts with better communication, not more CSS knowledge or tutorial watching. While, they both stayed clear of the day’s otherwise very AI focused theme, their talks highlighted a skill that’s becoming even more critical as AI handles more technical implementation.

Bell in particular talked about how better communication skills unlocks decision making influence, to shape better code from the top down, rather than having to salvaging a bad brief or design. When AI can write complicated functions and components for you, making sound strategic decisions will be key to making better software.

AI dominated the day, but a secondary theme emerged: simplify your systems. Several speakers called out increasingly complicated token sets in design systems. Luis Ouriach from Figma talked about the danger of over-experimentation and the tech debt it can accumulate. A simple quote I wrote down was, “Am I missing something? Do these complex token hierarchies actually help, or are we just over-engineering everything?” When you’re staring at an over-engineered system feeling lost, don’t assume you’re missing something—question whether the complexity serves a real purpose.

Unstructured documentation websites are not enough for consumers to retain knowledge about the design system and how to use it. We need to provide structured learning paths. Ashley and Erin used Figma’s interactive docs as an example of the kind of hands-on learning to provide.

I really enjoyed Donnie’s talk about how we should use constraints to question established patterns and come up with better, more creative solutions. For example, how would an “are you sure?” flow work if we couldn’t use a modal? Operating systems have used rubbish bins for the same flow for over 50 years, are modals actually a better solution than this?

Bill Collins delivered a brilliant talk on design-development communication, focusing on creating shared understanding of patterns and tokens. His example: “What’s the relationship between a primary button colour and its hover state: two random values, or always 10% darker?” This shared understanding lets patterns be implemented differently in design and code, to create efficient, maintainable systems.

At Etch, Gav recently wrote about systematic thinking in design systems, which echoed Collins’s core message. The best systems treat values, components and tokens as results of shared patterns, not foundations. This shared understanding of patterns, creates consistent products whilst keeping both design libraries and codebases well implemented.

In host Luke Murphy’s words: “A witch on stage wasn’t on my Converge 2025 bingo card!” Natalya Shelburne gave an alternative talk about facing creative fears in the age of AI. Drawing from her background teaching and leading creative teams, she argued for embracing the magic of creative people. In times of change—like the evolving tech landscape we’re in—it’s not the tools that shape the future, it’s the people using them.

Julien demonstrated how Sage built a Figma plugin for their design system using Zeroheight APIs. When clicking on a component in Figma, it brings up the relevant component documentation. Another practical example expanding on Madelin and Elyse’s earlier talks of how it’s already possible to bring docs to where people actually work rather than expecting them to context-switch to the full documentation website.

I’m sure everyone left the closing talk in deep thought, at least until they made it to drinks at the afterparty! John argued we need to stop accepting AI being framed as uncertain and inevitable – this framing labels those who question it as rejecters of progress and allows AI to take over without any standardisation. We only need to look at the industrial revolution to see how this plays out. Let’s not build the factory again.