We're Not Another AI Assistant. We're the Knowledge Substrate Underneath Your Stack.

The people on my team don’t really trust Copilot.

They’ve all tried it. They use it occasionally. But when it gives them an answer, they can’t tell whether it’s pulled from our policies, an old draft someone never deleted, or something the model made up that sounds right.

So they ignore most of what it tells them. They go back to asking each other or digging through Teams.

The problem isn’t the AI. The problem is what’s underneath it.

That’s the gap KnowledgeScout is built for. Not another AI assistant on top of the mess. A substrate underneath. The canonical layer of truth that humans and AI agents both call into, with a clear trail of where every answer comes from.

I’m going to spend this post explaining what that means, why it matters, and why we picked this position deliberately.

Two ways to play in the AI era

Most products in the knowledge space right now are betting on the same thing. Be the AI assistant on top.

Microsoft Copilot is an assistant on top of Microsoft 365. Glean is an assistant on top of your enterprise tools. ChatGPT Enterprise is an assistant on top of whatever you upload. The pattern is: take the existing chaos of where knowledge lives, layer AI on top, hope inference does the work.

The other position is the opposite. Be the foundation underneath. Build the canonical layer of truth that everything else (humans, AI agents, the AI assistants themselves) reads from.

We chose the second one.

It’s a quieter position commercially. “AI assistant” is what gets demoed in keynotes. “Substrate” doesn’t trend on Twitter. But we think it’s the position that actually solves the problem.

What we mean by “substrate”

A substrate is what something sits on. In biology, it’s the surface a fungus grows on. In manufacturing, it’s the base layer of a circuit board. In software, it’s becoming the word for the foundational data layer underneath an AI stack.

For knowledge management, the substrate is the canonical answer to a question. The version of “what’s true” that’s been written by a human, owned by someone, reviewed on a schedule, and made available to anything that needs to read it.

A substrate is different from a tool in three ways.

A tool is a destination. A substrate is a foundation.

A tool is something you log into. A substrate is something other things call into.

A tool serves one type of user. A substrate serves whoever needs it. Humans through search. Customers through widgets. AI agents through MCP.

KnowledgeScout was built to be that foundation. Every product decision flows from it.

Substrate to humans

Most knowledge bases assume a human in a browser, sitting at a desk, typing a question into a search bar. KnowledgeScout does that, of course, but we don’t assume it.

Your team mostly doesn’t have time to leave whatever they’re doing to go find an answer. They’re on a call. They’re in a CRM. They’re in their email. They’re in Microsoft Teams. The substrate has to come to them.

That’s why we built widgets. The same canonical content can be searched from a chatbot embedded in your CRM, a help button on your website, or a panel inside your internal portal. Same source of truth, three different surfaces. One curation effort, three places it shows up.

The point is: a substrate doesn’t make humans come to it. It shows up where humans already are.

Substrate to AI agents

This is the bit people are starting to get excited about, and it’s where the substrate framing matters most.

AI agents aren’t a future thing. Your team is using them today. Someone is pasting articles into Claude. Someone is building an agent in OpenAI to draft customer responses. Someone is experimenting with Microsoft Foundry IQ to orchestrate a process across systems.

All of those agents need grounded knowledge to be useful. Right now they get it through prompts, dumps, ad-hoc retrieval, or whatever the developer cobbled together. None of that is a real foundation.

We built an MCP server with eighteen tools so AI agents have a proper way in. They can search articles. Read content. Check what’s overdue for review. Propose new content into a human review queue. They can do almost everything a human editor can do, scoped to permissions, with an audit trail of every action.

The agents don’t have to come from us. We’re not in the agent business. Claude does that better than we will. So does OpenAI. So does Foundry IQ. Our job is to be what they call into.

There are two flavours of this. Internal agents (your team building automations on Claude, OpenAI, or in Foundry IQ) call into the substrate to act on behalf of humans inside your business, with the full surface of articles and documents the signed-in user is allowed to see. Customer-facing AI tools (Fin, Copilot Studio, Agentforce, your own bot) call into the substrate to serve your customers, scoped to public articles only, no surprises about what gets exposed.

Same substrate. Different MCP endpoint. Different permissions. One curation job that feeds both.

Substrate underneath Foundry IQ

Microsoft’s Foundry IQ is an agent orchestration platform. It’s built to coordinate AI agents across processes in a business. It’s a serious piece of work and a lot of enterprises are going to use it.

Foundry IQ is excellent at orchestration. It’s not built to be the canonical knowledge layer of a business. It assumes the knowledge layer exists somewhere and that its agents can call into it.

That’s where KnowledgeScout fits. We’re the substrate Foundry IQ calls into. Same for Claude-based agents. Same for OpenAI’s agent framework. Same for any custom agent your team builds.

This isn’t a competitive position. It’s a coexistence one. Foundry IQ does what it does well. We do what we do well. Together you get a stack that works.

What we replace, and what we don’t

Substrate doesn’t mean we sit on top of everything. We do replace some systems. We coexist with others.

Replaced by KnowledgeScout (typically): the team wiki, the internal training platform, the basic customer help centre, the SharePoint knowledge folder, the LMS used for compliance training. If you have these and they’re not really working, KnowledgeScout absorbs them.

Coexists with KnowledgeScout: Slack and Teams (for chat, not for canonical knowledge), email, your CRM, your project management tool, Microsoft 365, Google Workspace. These tools generate content and conversations. KnowledgeScout is where the canonical version of that content lives once it’s been turned into a real article.

We’re not trying to be your everything tool. We’re trying to be the one place where “what’s true” lives.

The honest admission

Substrate is a less exciting positioning than “AI assistant.” It doesn’t demo as well. The product team can’t say “watch the AI write a sales email.” We can say “watch the AI propose an article for human review,” which is true and useful but doesn’t have the same wow factor.

We’re betting the wow factor isn’t what wins. The companies that figure out knowledge management in the AI era will be the ones that built the cleanest foundation. Inference is amplification. A clean foundation amplifies into useful answers. A messy foundation amplifies into confident-sounding nonsense.

Substrate is the unsexy, slow, deliberate position. It’s what we built. It’s what we believe in.

Where this is going

Within five years, every business will have a stack that looks something like this. Multiple AI agents from different vendors, each handling a different process. Multiple AI assistants for different teams. Customer-facing chatbots. Internal copilots. Some will be Microsoft. Some will be OpenAI. Some will be custom builds.

All of them will need to read from one canonical layer of truth. The business that has that layer will move fast and get accurate answers. The business that doesn’t will spend the next decade chasing contradictions across forty different inference engines.

KnowledgeScout is the layer. That’s what we’re building. It’s why we exist.

The KnowledgeScout Team