Why now
Luna ships a small device and a simple routine that helps people on long-acting insulin wake up in range. The mission is calm, careful, and clinical — and the work behind it is anything but: regulatory submissions that span thousands of pages, clinical-trial deviations that need same-day judgment, hardware iterations on a tight schedule, a customer support queue we never want to be slow on.
The principle here is the same one we hold for our patients: the system should do the careful repetitive parts so the human can focus on the parts that matter. For our team, that means giving everyone on the team the AI support they need so the work that requires judgment, taste, or human contact is the work they actually do.
This site is the workplace where that happens.
What lives here
01 · Chat
Full-page conversations with any Luna agent you have access to. Threads persist, attachments live in Luna-owned storage, identity is verified.
02 · My Agents
Each person on the team controls their own agents — toggles, custom instructions, what's remembered. You see exactly what each agent retains about you.
03 · Workflows
Multi-step tasks Luna runs over and over — morning briefing, support triage, fleet briefing. Designed, observed, and tuned in one place.
04 · Fleet
What every Luna agent is doing right now, plus a daily and weekly briefing — written report and audio podcast — of what the fleet has been working on.
What's changing
What's not
The honest part
We're standing up a small team — a skunkworks — to work through this with the rest of the company. The fears below are ones we expect to hear, ones we share, and ones we want to be honest about up front. Each is paired with the design choice we're aiming for. If a row reads as marketing rather than as a design choice we'll actually build, tell us.
The fear
Am I being replaced?
What we built
No. Luna's headcount plan grows alongside this — agents are a force multiplier, not a substitute. The work AI removes is the work nobody wanted in the first place: meeting-note typing, triage, status rollups, format-converting the same data three times. The work it can't do is what we hire for.
Enforced by: Hiring plan owned by leadership; no agent has 'replace headcount' as a goal.
The fear
If the AI gets it wrong, who takes the hit?
What we built
The human whose name is on the work. Agents are tools your judgment uses, not co-authors. Anything that goes to a regulator, a patient, an investor, or a clinician carries your sign-off — and the audit log shows you reviewed it. If you spot a consistent failure mode, tell us; that's how we improve the agent or pull it.
Enforced by: Audit log records the reviewer for every regulated artifact; agent outputs are flagged 'unreviewed' until then.
The fear
What about my data — my words, my conversations?
What we built
Your conversations are yours. Each person at Luna has a private Durable Object instance with their own memory. Anthropic's Zero Data Retention is contractually on. Audit logs capture metadata (who asked what agent, when, how much it cost) — never the contents. You can see and delete what each agent remembers about you on My Agents.
Enforced by: ZDR is a contract clause, not a setting; per-user Durable Objects are how Cloudflare physically partitions state.
The fear
Is this going to feel like another tool to keep up with?
What we built
The places you already work — Slack, your inbox, this intranet — are where the agents live. We're not adding a new place to check. Onboarding is a 20-minute self-serve flow keyed to your role. If a workflow doesn't make your day better in the first week, we want to hear it.
Enforced by: Slack-first delivery for personal agents; /onboarding tracks are role-scoped so you only see what's yours.
The fear
Will I lose the parts of my job I actually like?
What we built
The opposite is the goal. Luna people stayed because the work matters and the craft is real. We're not automating the craft — we're trying to clear the bureaucratic and repetitive layer around it so there's more time for craft, not less. If a workflow we ship feels like the wrong call here, that's a signal to revise the workflow, not the person.
Enforced by: Workflow ownership lives with the team that does the work — not a central 'automation team.'
How decisions get made
What to do next
Onboarding is short, role-specific, and ends in a real task you do with an agent. The first module on every track is the same: what we're not asking you to do.