Notes from Web Summit Vancouver 2026
May 22, 2026
One word ran through Web Summit Vancouver 2026 more than any other: sovereignty. Over data, over infrastructure, over IP, and over the rules that govern all of it. The keynote framing was largely national - which country's stack ends up running the world a decade from now - but the conversations on the floor and at our booth were about something more immediate. They were about the architectural decisions enterprise teams are making right now, on real platforms, that will quietly determine who owns the stack later.
At the booth
Across the four days we talked with engineering leaders about the platforms we build and maintain and what those teams are doing as they add AI capabilities to them.
The first thing that came up, almost every time, was trust. The wording varied, but the substance did not: when a customer-facing AI feature gives a wrong answer to a real user, who owns that problem? The sharpest version came from teams who had seen enough to know that an AI feature bolted onto a fragile platform is usually the first thing to fail visibly when something underneath goes sideways.
The second was data flow. Most teams were already mid-build, with a search vendor chosen or a hosted LLM integration on the roadmap, and most of them had not yet sat down and traced what happens to a single query end to end - from the moment a user types it to the moment a result comes back. The AI integration market has moved quickly enough that mapping the data path is rarely the first thing anyone gets to.
The third was cost. Hosted AI infrastructure looks affordable at pilot scale and starts to sting at production scale, and the teams most interested in alternatives were almost always the ones who had already run the math at projected volume rather than current volume.
What we covered in the Masterclass
The session I presented was Best-in-Class AI Site Search Without the Middleman, built around Scolta, the open source AI site search toolkit we have been developing at Tag1.
The standard vendor playbook for AI search routes the user's search through multiple outside services, each touching the content along the way. That might make sense for global multi-tenant platforms with billions of documents across dozens of languages, but for most of the teams that visited the booth that heavy lift is unnecessary. They spent the week hearing that the data path matters, and the standard playbook is what that conversation is about. Scolta is a more practical answer, and it happens to be the answer that keeps sovereignty intact at the same time.
Scolta answers sovereignty questions by reorganizing where the work happens. The keyword search runs in the browser and the AI layer calls whatever LLM the team already uses, which means the only outside service in the data path is one the team already has a relationship with. AI summaries are grounded in the team's content and costs scale with usage.
Questions to take back
A few of the questions that kept coming back at the booth are worth carrying home:
Which third parties touch your AI features? Map every AI-touching feature you ship - search, summarization, support routing, recommendations - and write down exactly which outside services see the prompt or the output. Most teams have never made that diagram, and the exercise tends to be clarifying on its own.
What does the bill look like at the volume you are actually heading for? Run the projection at your real expected volume, then compare it against the cost of owning the index and the stack around it. The answer surprises people in both directions.
Who owns user trust if an AI feature breaks? When a customer-facing AI feature hallucinates or leaks, the customer does not see which vendor caused it; they see whose logo is on the page. That fact belongs near the top of the architecture decision, not the bottom.
Where Tag1 fits
At Tag1, our day-to-day work is keeping web platforms stable under real production load, protecting data on systems the team actually controls, and making sure the AI layer on top does not destabilize what is already running. That is the work we do for the enterprises that bring us in, and it is why the Vancouver conversations felt so familiar. They were about systems people already run and the pressure those teams are under to make the next layer of AI both useful and defensible.
If you want to dig into the material we brought to Vancouver, you can find it here: Vancouver 2026.
Interest in the Masterclass ran well past the time we had on the floor, so we are hosting a webinar.
Related Insights
-
/