Recent release of OpenAI Atlas browser is a huge game-changer. Similar to ChatGPT initial release. Love it or hate it, here's what's going to happen:
- Chrome will have competition.
- AI-friendly UI will be the new expectation for modern web apps.
- Microsoft and Apple will launch similar agents on OS-level.
- Over time human interaction with apps will reduce to bare minimum - reactions, micro-blogging, comments, messages. Most of the complex flows will be handled by agents.
Even now, observing Atlas interacting with our new client UI (NEO) I can see the adjustments we'll have to make. Good apps will need thorough element labelling (alt, aria, tooltips, captions), keyboard controls and most likely some form of standardised API with flow instructions - like a JSON file that guides the agent. Much like the old "SEO archive" pages for search-engines.
The dark side of it, of course, is that all the anti-spam tools and settings will have to be dialled in constantly and with much more sophistication.
Comments
In fact, the Edge browser has had all of this for quite some time — ever since the Copilot button appeared on the toolbar. In any case, there’s nothing good about this for website owners. When people first started talking about artificial intelligence and robots, we were told everything would be great — the robots would do all the work for us 😁 and we’d be free to focus on creativity. But in the end, even now, musicians, programmers, photographers, artists, copywriters, content site owners, and even video bloggers are starting to suffer from all these innovations.
Nobody can really know what happens next - it’s an unprecedented technology leap. Yet there are many way how it can pan out as a good thing. AI is not creative. It can do things that we perceive and consume as something creative, but the nature if this technology is tied to inference and transformation. It needs original data to improve.
What it means is that it will eventually consume any (maybe all) jobs, tools and processes that are derivative and that can be assessed under a known criteria. Perhaps not a bad thing, and may be the end if various forms, reports, docs, manuals, processes and protocols. Or rather, the end of those things for humans.
If your app/site is a resource library for accountants, or an ad-ridden blog with recipes, or a doom-scroll feed of nudes - you’re in trouble.
If you are building a platform for real people to connect and collaborate on topics that require human touch - you’re likely to be in control of the hottest resource of the next decade.
Consider this - modern AI models are running out of original training data. Historical data is already absorbed. Publicly available mainstream data (news, blogs, tweets) is synced in almost instantaneously. Shared personal data is linked in too. Models “intelligence” growth already starting to slow down - most will be about the same in the next 2-3 years. Much of the new training is infested with slop. Training models on AI-generated data is akin to inbreeding - it’s a sure fire way to introduce more hallucinations.
The world will need original data with deep context, peer review, multiple participants, innovative ideas - the kind of data that is only available at scale from group conversations of real people, in specialised topics, current events, relatable trends, real world observations. In other words I believe that you can bet on communities more than ever before.
This is an interesting angle Andrey. Niche communities may rise again. But then, AI chat bots are so addictive. It is hard to take your eyes away, once you are hooked in.