How we keep the office kitchen stocked

Somebody always wants more of the good crackers, and nobody quite owns the job of buying them. Asks turn up in Slack threads, chats by the coffee machine and notes stuck to the fridge. The office manager pieces it all into an order, and by the time it goes in, half the asks have been forgotten. Nothing much rides on it, but it never stops.
These days an agent on Machine looks after it. People ask for snacks with a Slack shortcut. Once a week the agent pulls every ask together, builds the order and posts it for the office manager to approve. Nothing gets bought until a person says yes.
The problem
Snack asks trickle in one by one, from all over. One person posts in a channel, another mentions it walking past, and the office manager is left building a list from memory. Some asks get lost, some get counted twice, and the order goes out later than it should.
It's an easy job, just a never-ending one. What it needs is a single place to ask, a regular moment to gather everything up, and a human yes before money leaves the account. A shared spreadsheet doesn't give you that, and neither does a repeating calendar reminder.
What we built
There is now one way to ask: a Slack shortcut. Each week a cron trigger starts a fresh, isolated session, a clean cloud computer just for that run. It gathers the week's asks, builds an order in the ordering account, and posts the draft in Slack for the office manager to approve. It only places the order once that approval arrives.
How it works
Take asks through a Slack shortcut
We connect Slack as a channel, and people make their requests with a shortcut. Every ask is added to this week's batch. There's a single place to ask, and nobody has to keep a running list in their head.
Gather them once a week
A cron trigger is a timer that goes off weekly. Each time, Machine starts a new session, a clean sandbox for that run alone. It collects everything asked for since the last order. One run, one sandbox, thrown away afterwards.
Hand it the shopping rules
The vendor we prefer, the budget, the staples that always go in, and how to fold duplicate asks into one line are kept as skills and memory, the playbook the agent reads each run. When tastes or the budget change, we update it.
Connect Slack and the ordering account
The agent reaches each tool through a connector, a narrow link whose keys stay on our servers, so the model never handles a raw token. With those, it can:
- See the asks in Slack. Everything sent in through the shortcut this week.
- Fill a cart in the ordering account. It builds the order with the vendor but stops short of checkout.
- Post in Slack. The draft order, for the office manager to look over.
Approve before anything is bought
The draft order waits at a human approval gate in Slack. The office manager sees the whole cart and the total, and nothing is bought until they approve. The ordering account's credentials are stored encrypted in the Secrets Manager and only handed to the agents you grant them to.
The pattern
Take asks through a Slack channel shortcut, gather them with a cron trigger, keep the vendor and budget rules in skills and memory, and park the order at a human approval gate before any money moves. Asks go in one door, and one approved order comes out.
Guardrails
The agent can put together an order that costs real money, so even a small job like this one gets proper limits:
- Isolation. Every weekly run gets a sandbox to itself. The only things allowed out are the draft order and the Slack post it is cleared to send.
- Scoped secrets. The ordering account's key is stored encrypted in the Secrets Manager and only handed to the agents you grant it to, when the sandbox starts.
- Human approval gates. No order is placed until the office manager says yes to the draft in Slack.
- Everything is code. Vendor, budget, staples and the weekly schedule are versioned files in the repo, changed only through a reviewed change request, never a dashboard toggle.
The outcome
Asks no longer go missing, doubles get merged, and the office manager checks a ready-made order instead of building one from scratch. Spending still waits for a person, so the busywork is gone and the agent never buys anything on its own.
More use cases
How we answer GDPR data-subject requests before the deadline
Our Machine agent checks each GDPR access or deletion request, finds the person's data across our product database, builds the report well within the SLA and hands it to legal before anything is sent.

How we keep our docs matching what the code does
Once a day a Machine agent reads every commit merged since its last pass, finds the doc pages those changes made wrong, rewrites them and sends you one pull request to review.

How we test every pull request without lifting a finger
Our Machine QA agent works with GitHub and our test environment. For each pull request it checks out the branch, runs the tests, tries the change for real, and posts what it found.
Put your first agentto work today.
- SSO
- RBAC
- Governed agents
- Full audit trail