The future of the web... again.
I built my first commercial website in 1999. That feels like a very, very, very long time ago.
Envelope - my core product - has now existed for almost 15 years, which is long enough to have lived through several supposed endings of the web as we knew it.
Mobile was going to make conventional websites irrelevant. Social media was going to replace them. Apps were going to replace them. Website builders were going to replace agencies. Headless systems were going to replace traditional content management.
Some of those predictions contained a useful point. None quite removed the need for businesses to explain what they do, publish reliable information and give customers a sensible way to deal with them.
AI is a bigger change.
It also raises a slightly uncomfortable question:
Have I spent 15 years building a much better version of something that is about to become entirely… redundant?
Fair to say it has caused a few sleepless nights.
I have landed on:
Drum roll.
I don’t think so.
Although I do think the role of the website is likely to change.
If someone wants to know a company’s opening hours, there is no particular reason they should have to visit its website, dismiss a cookie notice, find the contact page and scroll to the bottom. Their AI assistant can probably answer the question.
The same applies to more complicated requests.
A customer might ask an AI to find products matching several requirements, compare options, check availability or assemble the beginnings of a quotation. They may only need to see the website when there is something worth reviewing or approving.
That probably means fewer interactions will begin with someone typing a web address and browsing through a navigation menu.
The information still has to come from somewhere.
An AI cannot reliably know whether a product is in stock, whether a service is available in a particular area or which terms apply to an order unless there is an authoritative system behind it. If it is going to prepare an enquiry, create a booking or assemble a quotation, it also needs rules, permissions and current business data.
That is where I think the future of Envelope becomes more interesting.
A page is increasingly one way of presenting the information held by a business. The same underlying content might also be supplied as structured data, a clean Markdown document or a response to an AI assistant.
The more significant question is what an AI should be allowed to do.
Agents can already use websites by reading the page and clicking buttons in roughly the same way a person would. That is clever, but it is a rather indirect way for one computer to ask another computer to do something.
A site could instead expose a small number of clearly defined capabilities.
Search the catalogue. Compare these products. Check availability for this date. Prepare an enquiry using these details.
Each capability would use the same business logic as the human interface. The same permissions, validation and limits around who can access what would apply. The AI would not receive a magical administrative pass simply because it described itself as an agent.
I have spent a lot of time over the years, going right back to my days in the University of Sussex web team, advocating for the importance of user stories.
People arrive with intent. Working out what that intent is and how to serve it has always been the part of this job that makes my brain happy.
And now, in the middle of this enormous Copernican revolution, that way of thinking matters more than ever.
An AI visitor is not really one type of visitor.
A crawler collecting public information is different from an assistant working inside someone’s signed-in browser. An agent acting for a customer is different again. Codex helping us amend a website has a very different job from an anonymous system asking what time a shop closes.
They require different levels of access, identity and accountability.
That also makes Sonder’s role clearer. Envelope can hold the information and enforce the rules. Sonder can keep doing the bit I have always enjoyed: understanding what someone is trying to achieve and working out how the system should help.
It has made the overall direction clear:
Build for humans. Make the result easy for machines to understand. Give agents useful things they can do. Keep people in control of what matters.