Why a chat UI is the wrong wedge for enterprise AI
by David Bunting, Founder
The chat UI is the easy answer
When a company wants to “do something with AI” on its own documents, the obvious-looking product is a branded ChatGPT clone — the company name above the logo, the company documents underneath, a familiar chat surface in the middle. It demos well. The board approves it. The procurement form is signed.
Six months later, nobody uses it. The team kept using Claude or Cursor or whichever AI client they already had open. The branded chat sits unloved on an internal subdomain because asking your team to switch tools is harder than building the tool was. The wedge was wrong from the start.
The real wedge is underneath
The thing your team is missing is not another chat surface. It’s a knowledge layer that knows about your documents and that any AI client can plug into. The wedge isn’t the UI — it’s the retrieval underneath, plus the open standard that connects it.
That’s why Laminae doesn’t ship a chat surface. Buckets manage the documents, MCP servers expose them, the AI client the team already uses handles the conversation. Three concerns, three pieces, none of them trying to be the others.
What the open standard buys you
When the interface to your knowledge layer is an open standard rather than a vendor SDK, the question of “Which AI client?” stops mattering. Your team can switch from Claude Desktop to Cursor to an internal Copilot without you rebuilding anything. If a new client shows up tomorrow, it works on day one as long as it speaks MCP.
That decoupling is the part you should actually be paying for. Not the chat skin on top, which any junior engineer can ship in a weekend — the careful, boring layer underneath, the one that turns your documents into source-attributed evidence and keeps doing that quietly in the background for the next five years.
The branded chat clone is a tempting first step. It’s also a layer competing with tools the team already prefers. Build the knowledge layer underneath and let the AI client they already use do the talking — that’s the wedge that survives the next model launch, and the one after that.