The future of intranets

I spent a good chunk of my professional life trying to convince companies they needed an intranet. During my time on the commercial side at Happeo, I spoke to hundreds of businesses about their internal communication. I sat in meeting rooms in Stockholm and London, listening to executives complain about how their teams were disconnected and their information was scattered. We sold them employee experience software. It was good software. But I will be completely honest with you: the vast majority of companies quietly despise their intranets. The failure rate of these platforms is staggering. We blamed adoption. We blamed company culture. We blamed a lack of executive support. The truth is simpler. The traditional intranet failed because the fundamental mechanism behind it was broken. We asked busy people to do work they hated, and we gave them a tool that made it harder to find the result.

The publishing problem

Think about how a standard company portal is built. Someone in human resources or operations is told they now own the internal wiki. They spend three weeks designing an elaborate structure of folders. They write an enthusiastic welcome post. They upload the travel policy, the brand guidelines, and the onboarding documents. For roughly a month, the system looks great. Then reality hits. The company acquires a small competitor, the travel policy changes, and the person who built the folder structure goes on parental leave. The platform decays. Publishing is a chore, and maintaining that published work is even worse. Nobody wants to log into a separate content management system to edit a text block about the new office alarm code.

We expected busy people to act like amateur journalists, meticulously updating pages nobody ever looked at.

The decay of content was bad, but the search was the final nail in the coffin. I remember sitting at my desk, trying to find a simple pitch deck from a colleague. I would type a specific file name into the search bar, and the system would confidently present me with an outdated holiday form from four years earlier. If the search is terrible, trust evaporates instantly. Once an employee searches for a specific expense policy, fails to find it, and has to ask a colleague on Slack anyway, they will never use the search bar again. They bypass the system entirely. The intranet becomes a ghost town of outdated executive updates and useless links. We tried to fix this by redesigning the homepage. We moved widgets around. It did not matter. The interface itself was the problem. We were forcing employees to browse through a strict hierarchy of categories when they just wanted an answer to a specific question.

Asking instead of clicking

The shift happening right now is profound. With AI retrieval, the interface moves from browsing to asking. You no longer need to know which department owns the parental leave policy or whether it lives under specific administrative folders. You just open a chat window and type a simple question. The system reads the messy, unorganised pile of underlying documents and generates a direct answer. It changes the entire dynamic. I am seeing this transition clearly now with Substrat, where we are building infrastructure for AI developers. The products succeeding are the ones removing the burden of navigation entirely. The intranet of the future is not a destination you visit. It is an invisible layer you interrogate. The software does the heavy lifting, not the employee.

The mess we still haven't fixed

You might assume this solves the entire internal communications problem. I definitely thought so at first. I assumed large language models would instantly fix the enterprise knowledge mess. I was wrong. The interface is fixed, but the backend is still largely a disaster. The hardest part of internal knowledge has nothing to do with reading or summarising. It is about permissions. An AI that reads everything is a massive liability if your permissions are sloppy. If you have an open Google Drive where the founders accidentally left a spreadsheet of executive salaries or redundancy plans accessible to anyone with a link, the AI will cheerfully fetch that data the second a curious intern asks about pay bands. Sorting out who is allowed to see what is a tedious, entirely human problem. I got this wrong for years. I thought technology would eventually auto-organise permissions based on user behaviour. It doesn't work that way.

There is also the issue of truth. AI retrieval is brilliant at reading a document, but it cannot tell you if that document is the correct one. If the system reads an outdated PDF from 2019 that says employees get a specific bonus, and the actual rule changed last month but nobody wrote it down, the AI will confidently give the wrong answer. Someone still has to own the content. Someone still has to delete the old files. You still need an owner for the truth. I haven't got this figured out entirely. I sit in Marbella today, looking at how fast the AI space is moving, and I am amazed by the progress in retrieval. We have finally killed the terrible browsing interface. But the messy reality of running a business remains. Deciding who knows what, and ensuring what they know is actually true, is still waiting for a real solution.