Where AI meets reality: what we learned at CityJS London.
If you work in digital, you’ll know the AI conversation can feel relentless. Every week there’s a new tool, a new claim, a new reason you should be worried or excited. It can be hard to cut through the noise and work out what actually matters for your organisation.
We spent two days at CityJS London, taking in workshops, talks and the kind of corridor conversations that tend to stick with you longer than the slides. The picture that emerged was more nuanced than the usual "AI is eating the world" narrative. We came away with some key takeaways we felt were worth sharing, and thoughts on what they mean for the organisations we work with.
The industry isn't unanimously rushing into AI, and that's healthy
One of the things we noticed most at CityJS was how divided the room was on AI, and how refreshing that felt. This wasn't a conference full of true believers. Across multiple talks and panel discussions, developers were candid about where AI is genuinely useful in their workflows, where it produces output they don't yet trust, and where they're still figuring things out.
Douglas Crockford, who invented JSON and is something of a JavaScript elder statesman, opened the conference with two workshops and made no secret of his scepticism. He dislikes AI outright. Meanwhile, Kitze's talk on "vibe engineering" offered a more pragmatic take, arguing that the real skill isn't in letting AI loose on your codebase but in knowing enough to steer it precisely. That means understanding model limits, passing the right context, and writing clear instructions.
A panel discussion, "The future of the AI bubble", brought this tension into the open. There's no consensus, and we think that's actually the right place to be. When a technology is moving this quickly, a healthy mix of enthusiasm and scepticism keeps everyone grounded. The organisations we work with, from museums to universities, are navigating these same questions about where AI adds real value and where it's just noise.
Several talks also made a strong case for running AI models locally rather than relying on cloud-based services. Speakers walked through using open-source tools to run AI models on standard developer hardware, with results that are increasingly competitive with the big cloud-hosted services. The concern driving these talks was clear: handing everything over to large AI corporations carries real risks, for privacy and dependency. For organisations handling sensitive data, whether that’s research findings, collections records or student information, this is a conversation worth paying attention to.
People are finding information in new places, and your data needs to be there
The talk that gave us most to think about was Tejas Kumar’s session on what he called “the new UX.” His central argument was simple but striking: the interface for interacting with a product is no longer fixed. People increasingly reach for AI tools like Claude or ChatGPT to get things done, and if your organisation’s data and content isn’t accessible in those contexts, they’re effectively invisible there.
To make his point, he demonstrated something called MCP (Model Context Protocol). Think of it as a way to plug your organisation’s data directly into AI tools, so that when someone asks a question, the AI can pull live information from your systems rather than relying on whatever it scraped from your website months ago. Tejas built a live demo on the day: a CityJS conference companion that let attendees query speakers, talks and the schedule directly in ChatGPT. No app, no separate website, just the data showing up where people were already looking for it. It's a relatively small piece of infrastructure that could significantly extend the reach of content that already exists. Tejas has since published the full code and a walkthrough of how it works if you want to dig into the detail.
If AI is becoming a surface, your brand should show up in it
Making your data available in AI tools is valuable. But what Tejas Kumar showed next opens up a different question entirely.
Tejas’ demo didn’t just serve raw data into an AI interface. It served styled, branded components, a fully themed conference schedule card, a speaker grid and individual speaker profiles, all appearing inside the chat interface with their own visual identity. The emerging MCP Apps specification makes this possible, allowing organisations to control not just what information tools return, but how it looks when it gets there.
This raises an interesting question for the organisations we work with. If someone asks an AI tool about an event, a course, or a piece of research, and the response includes a branded, designed component from that organisation, that’s a meaningfully different experience to a plain text answer scraped from a webpage. It's also something that can be designed and maintained, rather than left to however the model decides to summarise your content.
We already build design systems that give our clients’ teams flexibility within their CMS. Extending those into AI interfaces feels like a natural next step, your brand identity travelling with your content, wherever it’s consumed. We're at an early stage with this, and the tooling is still developing. CityJS gave us both the technical grounding and the conviction that it's worth the time.
The bigger picture
The thread running through all three of these takeaways is the same: your website is no longer the only place your content needs to work. The organisations that recognise this early, and start making their data available beyond their own site, will be the ones whose content actually gets found as AI tools become a bigger part of how people look for information.
If your organisation has structured data that could be working harder in an AI-driven landscape, get in touch. We'd love to talk it through.