Posted on Aug 27
Six bets on web development in the era of AI
Is it safe to assume that AI is not coming for our jobs after all? What about SEO experts and old-school webmasters? I wouldn’t claim to know exactly what the future holds, but I’d like to share with my thoughts on the matter and hear yours.
1. SEO would still matter
Currently applications like Claude or ChatGPT rely on three sources of data: their internal training data, contextual information (system and user prompts, RAGs, “memory” etc) and old-school web search APIs.
Depending on a particular prompt the system decides if it needs to fetch new information from the web or not. And for e-commerce requests like “I want to book a hotel” or “buy a bag of cat food” AI would need to go and fetch fresh data from actual sellers - they simply won’t have reliable information available elsewhere.
That’s why I believe that structured and clear content, semantic markup and engagement signals would still matter just as much, if not more than today.
In fact, I think that AI would have to learn to measure the quality of the content and its reliability. It’s all fun and games until someone is legally liable for producing false information, as they say. Cleaning up data - AI slop in this case - is a necessity for the next generation of AI tools to emerge.
2. Progressive Embedded media would emerge
I don’t think users would ever stop visiting the websites directly. It might have been the case if we’d only rely on text and basic media, but it’s not - interactive web components are what makes user stick longer on the pages and give those good engagement signals for the search engines as a side effect.
I believe that we will continue evolving technology to allow machines enjoy clear and fast well-structured content and for humans to enjoy accessible, rich, interactive widgets that can be accessed on the source website or directly in your AI chat.
3. CSS would be used less for Design, more for Design Systems
If we are to explore the idea of chat embeds further, I would bet that AI chat applications would still want to have a modicum of control over how the interface would look like. At the very least it would be great if an embed wouldn’t stick as a sore thumb and instead would fit into UI of the tool seamlessly.
We have done something similar with dark and light modes, but I think we would move it even further - towards customisable designs. CSS already has variables that we could use to make embedded applications look great in ChatGPT, Claude, Gemini and our own apps/websites.
4. Accessibility would matter for AI too
Let’s say I want to schedule an appointment with my hairdresser. I can ask my AI assistant to schedule it for me over voice or in a chat.
If I use a device without a display, e.g. a home hub, AI assistant would have to present me with information verbally, akin to a screen reader.
And to make an actual appointment AI would definitely prefer to use text-based information rather than a screenshot of the page as it’s much less energy consuming.
5. Wider adoption of MCP & LLMs Docs
Model context protocol is already quite widespread and I think it would make its way into pretty much any e-commerce website.
I’m not sure what we would settle on, but at some point we would most likely adopt something like /llms.txt as we need a way to share the information about capabilities like MCP support with the machines.
6. Single Sign On & Passwordless authentication adoption
The future where one would trust AI to perform operations on its behalf on a wide variety of platforms demands a secure way to authenticate AI agents on external platforms.
I think the most likely winner would be a combination of password managers and single sign on (SSO) from them or other trusted providers.
That also means that MFA codes sent by email and SMS would have to retire and make way for passkeys or at the very least switch to digital MFA codes that one can share with the password manager.
Final thoughts
Call me a hopeless romantic, but I’d love to see AI tools giving rise to genuine, unique, human made high quality content.
Yes, we took a bit of a detour with image and video generation and Suno is making quite decent music nowadays, but ultimately I would like to hope that there soon come a time when AI would need creators to feed them new content. To be fair, the time is almost there, I just hope that writers, musicians, researchers, poets, artists, photographers, painters and, yes, us, developers would be valued for our skills and contributions.
What do you think about the future?