Skip to content

Blog

Delegate, don't operate

The Luge team

Luge's founding principle: delegate, don't operate. Instead of driving AI gesture by gesture (isolated prompts, copy-paste), you hand objectives to AI colleagues that know how to act (tools), know your trade (skills), keep context (memory) and check your sources (knowledge base) — under three rules: every piece of information lives on the right surface, every AI message declares itself, and human decisions stay sovereign.

Look at how most teams use AI today: an open tab, a prompt, an answer, a copy-paste. Then another prompt, because the answer forgot the context of the first one. Every gesture goes through your hands. You didn’t hire a colleague — you’re holding a steering wheel. And a tool you drive gesture by gesture doesn’t take work off anyone’s plate: it moves the work around, it doesn’t delegate it.

Luge is built on the opposite principle: delegate, don’t operate. You don’t dictate gestures, you hand over objectives — the way you would with someone on the team. “Prepare this week’s quote follow-ups.” “Write up the meeting and open the tasks that come out of it.” The agent acts, collaborates, and executes in the background. You decide.

What it takes to delegate to someone

Think about what makes delegation possible with a human. It’s exactly the same four things for an AI colleague — and it’s Luge’s architecture.

Knowing how to act — tools. A colleague you can’t hand anything concrete to isn’t a colleague. Luge’s agents act on your systems: your calendars, your code repositories, your CRM, your boards and tables — through ready-made integrations or tool servers you plug in. A tool answers the question “what can I do?”.

Knowing the trade — skills. Knowing how to act isn’t enough: you have to know how it’s done here. A skill is a package of instructions loaded only when the task calls for it — your quoting procedure, your meeting-minutes template, your conventions. It answers “how do we do this here?”, without weighing every conversation down with the whole employee handbook.

Keeping context — memory. A colleague you have to re-brief every morning isn’t one. Luge’s memory holds durable facts — preferences, conventions, decisions — distinct from the raw conversation history, and every fact keeps the name of whoever taught it. History says what was said; memory says what remains true.

Checking the sources — the knowledge base. For questions whose answer lives in your documents, the agent searches by meaning through what you’ve indexed — contracts, guides, internal notes — and never beyond the access rights of the person talking to it. It cites where things came from.

Four pillars, one consequence: you can hand an objective to someone who knows how to act, knows your ways, remembers, and checks its sources. That’s the definition of a colleague.

Where the work lives afterwards

Delegation produces work — and that work has to live somewhere the team can see it. That’s why Luge is a full office rather than a dialog box: channels where an agent answers when named, versioned notes whose history names every author, boards where a card carries its own discussion thread, tables for structured data, meetings with a note taker that attributes every word, automations that fire on a schedule or an event — and a governance layer that runs through all of it.

The three golden rules

Delegation only holds if three rules are respected — and they’re wired into the platform, not left to goodwill.

Every piece of information lives on the right surface. A fleeting idea goes into the capture inbox. Shared work in progress becomes a card. A durable analysis becomes a note. Repetitive data goes into a table. You pick the surface by how long the information needs to live — and you never explain the same thing twice again.

Every AI message declares itself. Text written by an agent carries an explicit marker. Never any ambiguity about who wrote what — it’s the condition of trust in a mixed team, and a requirement AI governance frameworks spell out in black and white.

Human decisions stay sovereign. When an action matters — something leaving the company, an irreversible step, a judgment call — the automation stops and files a request in your inbox. You can correct the draft before approving, not just say yes or no. And no agent can answer its own question: autonomy is a privilege you grant, never a design default.

Delegating isn’t giving up control

It’s moving control to the right place: from gestures to decisions. You stop driving and start directing — and every one of the agent’s moves stays traced, attributed and reversible behind you.

To see what this office looks like, take the tour. To meet the employee itself, it’s here. And for straight answers, the FAQ doesn’t dodge.