Why we built an AI employee, not a chatbot

Ever hired someone who answers questions beautifully but never actually does the work? Neither have we. Nobody would. Yet that’s exactly what teams have been sold for years under the name “AI assistant”: you ask a question, it answers. You close the tab, it forgets. Next morning you explain everything again, to the same tool, with the same patience.

When we started Luge, we asked ourselves one simple question: what would an AI look like if you hired it the way you hire a person? Not one more tool in the pile — a colleague. Someone who shows up in the morning, knows where the team left off, and moves the work forward. The answer eventually fit in one sentence, and it became our manifesto: a chatbot answers, an employee does the job.

A chatbot answers. An employee does the job.

The difference shows up on an ordinary day. Ask a chatbot to follow up on a client file: it writes you a lovely paragraph about the importance of client follow-ups. Ask your AI colleague in Luge: it opens the file, checks where things stand, moves the card on the board, and drafts the follow-up message.

The difference isn’t intelligence — both run on the same kind of models. The difference is responsibility: an employee takes a task and carries it through. It reads the team’s notes, updates the boards, writes what needs to go out, and reports back on what it did.

And when it gets stuck, it does what any good employee does: it asks a question instead of making something up. The question lands in your office, you answer when you have two minutes, and the work picks back up. A chatbot invents. It has no choice — it has to answer on the first try, because it may never see you again.

A memory with names on it

You train a new employee once. Not at every conversation. That’s so obvious nobody says it — except with AI tools, the opposite happens: every session starts from zero, and you end up carrying the memory on its behalf.

In Luge, your AI colleague remembers. But we went one step further, because a team memory without provenance is just a rumour: its memory is attributed. Everything it learns carries the name of the person who taught it.

Concretely: when your AI colleague states that quotes always go out on Thursdays, you can see it was Julie who taught it that back in March. If it’s no longer true, you correct it once — and it’s corrected for the whole team, not just for your next chat. You always know who taught what to whom, and when. That’s what makes trust possible: you don’t trust a memory you can’t check, or a colleague who never cites their sources.

The draft before the send

You shouldn’t trust an AI that presses “send” on its own. Nobody should.

So we built Luge around a gesture everyone already knows: the draft. When your AI colleague prepares a reply to a client, you get it before it leaves. Not a curt yes/no in front of take-it-or-leave-it text — an actual draft, which you correct right in the text before approving. You change two sentences, you approve, it goes out — in your words, not the model’s.

It’s the same deal you’d make with a new hire: at first, everything goes through you. They write, you review, you refine. Over time the corrections get shorter — because what you change, it notices. But the approval gate stays as long as you want it there. You decide what leaves the office. Not the model.

A shared office, not a personal tab

The big players’ AI assistants live in a tab: yours. What happens there stays between you and it. Whatever it learns with you, your teammate will never benefit from. That architectural choice gives the vision away: those tools were designed for one person alone at a screen.

Work happens in an office. Notes the team shares. Boards where cards move forward. Meetings — where the notetaker sits in as a visible participant, listens, and attributes to each person what they said. Your AI colleague works there, in the same office as everyone else. When you teach it something, the team inherits it. When it finishes a task, it shows on the board everyone looks at. When the meeting happens in a voice room, it was in the room — and the summary names who committed to what.

An employee working in a locked room where a single colleague holds the key — that doesn’t exist. We couldn’t see why an AI should get to.

A flat price, because the AI runs on your side

Why are AI tools sold by the token? Because every question your team asks passes through somebody else’s servers, and that somebody keeps a meter running.

We took the opposite road: Luge’s AI runs on your machines, or in a private cloud that stays in Canada. Your data stays on your side — and we have no meter to spin. That’s why the price is flat: you pay for an employee, not for every sentence it speaks.

That changes more than you’d think. When every question costs something, teams start rationing — they ask less, delegate less, and the tool ends up gathering dust. On a fixed salary, nobody checks a meter before talking to a colleague. That’s how a team ends up actually adopting one.

Come meet it

A chatbot answers. An employee learns, asks, works in the same office as you, and leaves you the last word. That’s the difference we built — and the one we’ll keep digging into, version after version.

If you’re at the point of wondering what this looks like in practice, we’ve described what the AI employee does with its days. And the pricing fits on one page — flat prices, as promised.