The idea is not new. Plenty of startups have built into Slack, because that is where corporate attention is and it is convenient to reach people there.

OpenTag is after something bigger, and it is right about it. Slack already contains everything.

What they built

OpenTag is an AI coworker that sits in your Slack with full context on your company. It answers questions, takes real work off people's plates, and keeps the wiki up to date automatically so the knowledge never goes stale.

The founders have done this together before. Tony Kam, chief executive, was previously at Tesla and founded Lilac Labs. Shelden Shi, chief technology officer, co-founded Lilac Labs too. Wilson Nguyen was a founding engineer there. All three studied at Berkeley. This is the same team on their second company.

Plenty of people want to be the assistant in your Slack. Glean built a large business searching across everything a company uses. Slack ships its own AI, and Microsoft Copilot has the strongest position of all because it already sits where the work happens. Notion and Confluence are where the wiki rots today, with Guru and Slab built specifically to stop it rotting. Dropbox Dash comes at it from the file layer, and Mem0 sells memory to developers building agents. Most of them treat Slack as a place to reach people. Very few treat it as the place the knowledge already is.

The point: all the company knowledge is already in the conversation

Start with how knowledge actually comes out of people.

Written down as documentation, it is at its best: structured, detailed, ordered, complete. That is why every company wants a wiki.

The trouble is that documentation is unnatural. It is not how anybody thinks or speaks. It takes effort, it happens after the fact, and it helps somebody else later while costing you time now. So it does not get done, and what does get done goes stale.

The same knowledge comes out of the same people all day long in conversation, and it comes out easily. Somebody asks why the billing job runs twice, and a colleague explains it properly in four messages. That explanation is real knowledge, given freely, with no effort anybody resented.

It is just badly structured. Scattered across a thread, missing context, never titled, impossible to find in six months.

So human knowledge arrives in two forms. Documentation has the structure and never gets written. Conversation gets written constantly and has no structure. The gap between them is not a people problem, it is a format problem.

AI cannot make people be structured, and does not have to

This is the part worth taking out of the file.

Every previous attempt at this tried to change the human. Write it down. Use the template. Tag it properly. Update the page when things change. All of it is asking people to behave in a way they do not naturally behave, which is why thirty years of knowledge management tools have all ended the same way.

AI cannot force anyone either. What it can do is supply exactly the thing people are bad at.

People are excellent at producing knowledge naturally and poor at structuring it. AI is the reverse. So stop demanding that humans supply the structure, and let the machine add it to what they were going to say anyway.

That is a proper division of labour rather than a workaround. The human does the part only a human can do, which is knowing the answer. The machine does the part humans hate and machines find easy, which is organising, titling, cross-referencing and keeping it current.

Read that way, OpenTag is not a Slack bot. It is a system that listens to people talking and extracts what is useful for the people paying for it.

Expect a lot more of these

This is my opinion, not something the company has said.

Building AI on top of conversation is going to produce a run of good surprises as these teams put more products in front of real users, and OpenTag is an early one rather than a special case.

The reason is that conversation is the largest and most neglected source of knowledge in any organisation. Every company has years of it. Nobody has ever been able to use it, because unstructured text at that volume was worthless before machines could read it properly. That constraint has just gone.

Look at what else lives in conversation and is currently thrown away. Why a decision was made and what was rejected. Which customer complaints keep recurring. Who actually knows what, as opposed to what the org chart claims. How a process really runs, rather than how it is documented. All of it is said out loud, in writing, every day, and none of it survives the week.

We should expect more startups built on conversation rather than on documents, and they are worth watching closely.

Why it is bigger than it sounds

The short description is an AI coworker in Slack. The idea underneath is about who supplies what.

  • The knowledge is already there. Slack holds it because people explain things to each other all day. It is not missing, it is unstructured.
  • Documentation is structured and unnatural. Conversation is natural and unstructured. That gap is a format problem, not a discipline problem.
  • AI should fill the gap rather than close it. People cannot be made to be structured. The machine adds the structure to what they were going to say anyway.

What to watch

The number that matters is how often people ask OpenTag something instead of asking a colleague. That is the only proof it has become where the knowledge lives rather than another tab.

The second is whether the structure it adds is actually right. Turning a scattered thread into a clear, correct, findable page is the whole product, and a confidently wrong page is worse than no page at all.

The third is what they build once they have been listening for a year. If the conversation really is the company's knowledge, then answering questions is the smallest thing you could do with it.