Before MozFest, I wrote that I was curious how much of the «Wilding» conversation would reach data governance, interoperability and the question of who writes the rules. Here is my answer, with a caveat. I attended about 15 of the 200+ sessions, and I skipped most of the advocacy ones. This is my route through the festival, not a report on it.

A well-run festival

MozFest 2026 took place in Barcelona from 28 to 30 October 2026, with about 3,000 people in attendance, according to Mozilla. Sessions started on time and left room to walk between rooms (anyone who has been to a conference in Spain will know how unusual that is). Production and moderation were excellent, and the sessions I attended had many questions from the room. Thank you to Mozilla and congrats to Zeina Abi Assy, the festival director, and her team.

Interoperability came up as a worry about lock-in and portability

The issue came up in several sessions. For example, in the fireside chat on the open future of AI, the point was that a company can invest so much in one provider that leaving feels like losing something. A talk on open AI devices named three places where people get locked in: memory, connectors and hardware. In the session on the stacks of 2046, a speaker argued that fragmented stacks don’t help, and that stacks need to connect with each other. Several people discussed the difficulty of taking your AI memory with you.

I don’t remember anyone mentioning standards bodies or specifications, but with 200+ sessions, someone surely did. In the current context, I see protocols that start inside single companies. MCP and A2A, for example, were created by Anthropic and Google, and have since moved under the Linux Foundation. That is a step forward. In my view, though, multi-stakeholder governance, with civil society and public bodies at the table, is being downplayed. This is a part of the conversation I know well, and I would have liked to hear more of it.

My bet: portable AI memory

In a participatory session on leaving the big office suites, we started designing a process to migrate email as an easy example, then realised that emails are full of links to files in the cloud drive, and those would be lost when migrating. We agreed that security and integrity trumps everything else and even the easiest perceived change is more complex in practice than on paper.

Making AI memory portable may look much harder than that, but I’m slightly more optimistic, for a few reasons:

• There has been less time for capture. People have not been invested for as long as they were in something like MS Office.
• Tools appear non-stop, and vibe-coding and reverse engineering are becoming ordinary.
  • Models are becoming commodities faster than I expected, which widens the options and makes local solutions possible.
• People have seen capture before and are more aware of it.
• Some regulations exist and others are being developed to avoid further lock-in.

I may be wrong. I believe disruption is possible but not easy (it never is).

Who oversees the agents?

This is the «who writes the rules» question again, this time for agents. In «Delayed Rules, Faster Agents: Who Governs AI in 2026?», speakers said it is ridiculous to think that humans will oversee AI agents. There was one good reason given that stayed with me: an agent can make more decisions in a minute than a whole team in a month.

If that’s the case, what’s the solution? Humans must choose the policies, but they don’t need to watch every action. Supervision still matters, and this means setting the rules and checking that the systems follow them.

Technology is a tool

The mood was pessimistic about tech capture and about the political situation around the world, but I think it was a bit too optimistic about the role of technology in solving significant societal issues. Part of that is due to my own route: I picked many sessions about building (local AI stacks, AI devices, personal context) and few about advocacy. The organisers describe the theme in terms of care and collective action, so I may have missed that side.

My optimism about portable AI memory is technical too, so let ellaborate: technology can make switching possible. Getting people to switch is a different problem, and I would have liked more on adoption. In one session, a speaker asked how many people had changed their default browser. Most hands went up. That tells you about the room, and little about the general public. In another, a speaker said we need to eat our own dogfood. True, and not a criticism of anyone: if migrating were easy, many would have done it already.

Déjà vu

I have been in versions of these conversations for more than 20 years. In some sessions I felt we were repeating ourselves. I wrote something similar after the OGP Summit last year, when I heard once more why opening data matters (OGP Summit 2025: Same old, with a twist?). Two different communities, same feeling.

Does the needle move? Slowly. One speaker reminded the room that even the Internet took about four decades to succeed. It’s also fair to say that some discussions will never go away. They need to be had.
That said, I still wonder how useful it is to repeat them to a room that mostly agrees already. I don’t have a full answer.

Two things from the sessions seem worth testing. One speaker told the story of a community in Chile that was already worried about a drought. They linked it to the water that data centres use, so that citizens could better understand the pros and cons of building a data centre in that region and ultimately oppose it. That made the issue click for people, both in Chile and in the room. Another speaker suggested using an open tool in addition to whichever is the dominant one for a certain purpose, not instead of it straight away.

If you were there, and we didn’t meet, please say hello. I usually spend more time in the corridors, but I focused on attending sessions and information gathering for a change this time. And if you went to sessions on standards or on the governance of interoperability, tell me what I missed.