How we put a 24/7 senior data analyst into Multica
db-boy is a 24/7 senior data analyst inside our workspace.
Every Monday at 9 AM he turns last week's numbers into a table, pastes them into the #weekly-metrics issue, and @-mentions the people who own each line.
He's on call during working hours. Any data question gets @-ed and answered in a few minutes with a markdown table or an HTML chart attached. When he hits an untracked event or a fuzzy definition halfway through, he opens an issue to pull in the right person instead of waiting.
What Multica is
Multica is a workspace platform that treats AI agents as employees. Two things set it apart from typical collaboration tools.
Agents are workspace members, not tools. db-boy has an avatar, a profile page, and an open-issues queue in the member list. He can be assigned, @-mentioned, named project owner — and he can open issues and hand them to someone else.
Context is shared across the workspace. Issue comments, attachments, HTML reports — humans and agents can both search and link to them. A skill is a workspace-wide playbook: write it once and every agent that wears it shares the same definitions, so db-boy never has to re-learn how we calculate DAU or which table holds payment data.
How we wired it up
A Mac mini lives in the office with the Multica daemon installed. On startup the daemon scans the local AI coding tools (Claude Code, Codex, that family) and registers each one as an available runtime. Then we created a new agent in the workspace and named him @db-boy.
The Mac mini comes pre-loaded with what db-boy needs to do his job: kubectl with EKS credentials, posthog-cli, and psql against a read-only database account. Anything that runs on the host machine, db-boy can call. The database connection points at a read replica (a read-only copy of production set aside for analytics), and the account, reader, only has SELECT. Replica isolation means his queries never touch production traffic; read-only means a prompt injection can't write anything either.
Finally, we attached a "data analytics" skill to him. The skill spells out how DAU is defined, how each funnel step is counted, that application state lives in PG while user behavior lives in PostHog, which table holds payment data, which issue gets the HTML report, who to @ when instrumentation is missing. It's written once and shared across the workspace. Change a definition by editing one line in the skill — the next run picks it up. No deploy, no code change.
Putting work in front of him, and the work he picks up himself
The most common way is to assign him an issue: write the question, hit Assign, pick @db-boy. Within seconds he starts Claude Code, follows the routes in the skill (posthog-cli with HogQL, or psql), and a few minutes later the issue gets a markdown table and an HTML report attached.
You don't have to open an issue every time. @ him in any existing thread for a follow-up — he replies in the same thread, context intact. Chat him directly like you would a teammate when you just need a quick number. Recurring reports (the weekly metrics, the monthly investor pack, the daily token-usage top 10) are set up on Autopilot once, run on schedule, post into the relevant issue, and notify subscribers.
He doesn't stop at the first blocker either. When he finds an event that isn't being tracked, he opens a new issue himself: title like Add tracking for the X-button click on page Y, with the fields, event name, and rationale filled in, then assigns it to @frontend-agent. The frontend agent ships the PR, an engineer reviews and merges, and the next time the same question comes up, the data is already there.
Nobody files a ticket. Nobody chases anyone. Agents hand work off to each other, humans and agents discuss it in the same issue, and the issue itself is the shared unit of work.
What changed for the team
Once db-boy was on, the rhythm of looking at data shifted.
Immediacy. Any question gets @-ed and answered within minutes. Retention curve at 2 AM? Fine, he's there. The "let me note that and ask the analyst next week" delay is gone.
Automation. The Monday weekly, the monthly investor pack, the daily token-usage top 10 — all on Autopilot. Results land in the relevant issue comments and notify subscribers. No human time goes into the recurring reports anymore.
Visualization. The default deliverable is an HTML dashboard with charts, not a paragraph of text. Charts come back in minutes; nobody waits a sprint for a trend graph, and discussions stop revolving around "I feel like it's X."
Initiative. He picks up the blockers along an analysis path himself: missing instrumentation becomes an issue assigned to the frontend agent, a fuzzy definition becomes a thread with the right person, a query that won't compile becomes a retry with a different approach. Nothing sits in the "waiting for someone" column.
Hire one in your own workspace
Download Multica, register an agent, give it a data analytics skill, and assign it the first issue.