I went to SXSW 2026. Loved it as usual, but I couldn’t help coming away with the distinct impression that nobody — not even the people on stage who sounded confident — actually knows what to do about AI.
In the Webby Awards session on AI and trust, Nick Borenstein shared a stat that felt obvious but also kind of profound to see on a slide: comments asking “Is this AI?” were up 1400% year over year. They also said 77% of industry leaders trusted their own AI outputs, while only 16% trusted brand outputs.
I don’t know enough about the methodology to build a cathedral on those numbers. But it certainly matches my own experience.
I went to a lot of talks basically around the same theme: “Sure AI is disrupting stuff but [X will always be human] or [here’s what to do].” And we all love someone who knows what to do. I didn’t leave one of those talks convinced anyone actually had a good answer.
As the Risky Business put it: “No one knows what to do and everything is up in the air.”
Useful Signals
Amy Webb and the Future Today Strategy Group framed trends as weather data and convergences as the actual storm system. I always love Amy’s talks and I thought this one was particularly astute. In an age of exponential change we have to be ready — even keen — to destroy what got us here in order to create something new. Creative destruction. Sure there are a ton of trends, but what makes things even harder is that the trends are all accelerating and combining and merging and interacting in truly mindblowing ways.
The notes I came away with were basically these:
- Human augmentation: glasses, exoskeletons, biology, brain-computer interfaces, all slowly moving the boundary of the body.
- Automation: agentic systems and robotics changing who does the work.
- Emotional outsourcing: people using AI for friendship, romance, therapy, religion, reassurance. This isn’t theoretical—it’s happening now.
The phrase from Sam Jordan’s later FTSG talk that I wrote down was: automation follows rules; agentic systems pursue goals. The distinction matters. A workflow speeds up a known path. An agent can decide which path to try.
So yes, there’s a huge productivity story here. Promptable simulations can let people test before building. AI-assisted biology can search through options evolution never tried. Agents can turn a rough goal into a ranked set of possible actions.
But the old slow bits did some work for us. Building by hand gave people intuition. Wandering around dead ends gave researchers curiosity. Junior work gave future leaders the chance to be wrong, corrected, embarrassed, and better. If we remove the friction, we probably need to put some of the learning back deliberately.
This captures it so well.
The Trust Bottleneck
You can’t verify trustworthiness just by looking at a piece of content. You need its connections: the author, the publication, the date, the funding body, the edit history, the whole network around it. As AI compounds what we can do, the constraint isn’t capability anymore — it’s whether humans can actually grasp what they’re signing off on.
Receipts
For publishing, brands, commerce, and probably most of the web, the practical lesson is much less glamorous than the talks.
Show the receipts.
Where did the claim come from? What changed since the first version? Who’s accountable for the work? Was AI used, and where? Can I get back to the original source instead of another flattened summary of a summary?
The AI trust conversation often gets pulled toward labels and badges. Some of that will matter. Content Credentials could be useful. So could regulation, disclosure norms, and better platform tooling. Fine.
But a lot of this is ordinary publishing discipline wearing a futuristic hat. Link properly. Keep a canonical copy. Make corrections visible. Put your name on the work. Let people inspect the chain of custody. Do fewer things that look slick but make people wonder what was hidden.
This is where I think owned websites get more interesting again. Not because everyone should retreat from social platforms or marketplaces. Use them. They’re where attention lives. But the original version of important work should live somewhere you control: with the source links, the update history, the context, the disclosures, the feed, the boring metadata that lets both humans and machines understand what they’re looking at.
Boring, but boring in the way foundations are boring. Sometimes boring is good. I don’t want a house with “interesting” foundations and you probably don’t either.
Agents Change the Surface Area
The commerce version of this is already visible. If a customer asks an agent to compare products, summarise reviews, check return policies, or buy something, the agent needs usable material to work with. Catalogue data. Availability. Shipping rules. Support docs. Reviews. Guarantees. A checkout path it can understand.
A merchant who only exists inside someone else’s closed surface is making a bet that the gatekeeper will keep passing the relationship through. Sometimes that will be true. Often enough, maybe. But I wouldn’t want it to be my only plan.
This is one of the reasons I keep coming back to the open web. Not as ideology. More as practical resilience. Publish the thing. Own the record. Make it legible. Syndicate outwards. Let platforms amplify you, but don’t let them become the only place the truth of your work exists.
What We’re Losing
The strongest SXSW talks for me weren’t really about AI replacing work. They were about AI removing the conditions that used to produce judgement.
That’s harder to talk about because it’s less measurable. We can count time saved. We can count output. We can count cost reduction. It’s much harder to count the intuition someone never built because the simulation gave them the answer first. Or the character someone never developed because the assistant kept agreeing with them. Or the curiosity that died because the system got very good at giving plausible next steps.
I don’t want to be melodramatic about it. These tools are extraordinary. I want more of them, not less.
But I also want better receipts, more visible human judgement, and a bit less polish where polish makes things harder to believe. That may be the practical line for me after SXSW: use the machines, don’t be used by them.