You open Slack. Three things have already happened. Your stockout Handler flagged two SKUs overnight and paused the ads burning money on them. Your reporting Handler dropped the weekly trade summary into #ops — formatted, linked, ready to forward. Your sample Handler noticed a replenishment threshold and raised a PO draft for review.
None of this required you to prompt anything. You briefed these Handlers once. They learned the job. They show up every day.
The team treats them like junior staff: hand them work in Slack, get it back done. When something's ambiguous, the Handler asks a clarifying question before acting. When something's outside its scope, it escalates.
Day-to-day
You don't configure Handlers in a dashboard. You brief them the way you'd brief a new hire — in Slack, in plain language, with the context they need.
"Hey, every Monday morning pull last week's trade orders from Shopify, cross-reference against Cin7 stock levels, and flag anything we can't fulfil. Drop the list in #ops."
That's it. The Handler confirms what it understood, asks if anything is unclear, and starts. Next Monday, the list is there.
Using Handle 6 months in
Every card below is a Handler doing a job that used to be someone's Thursday afternoon. Pick a card to see the before, what they own, and how they're set up.
Answers the trade floor's stock and pricing questions, drafts quotes, and watches reorder levels — the desk the Trade and Supply Chain teams turn to before they interrupt each other.
Stock and pricing answers on the spot — not a colleague interrupted
The Trade and Supply Chain teams used to break each other's focus for every lookup. Now they ask Trey.
Tells the warehouse floor where stock lives and updates the books when it moves — the pick-bay replenishment the team used to queue for a team lead to handle.
13 interruptions a day, off the team leads
Replenishing a pick bay used to mean grabbing a team lead to read a long sheet. Now the floor asks Sammy — around 13 times a day, almost no errors.
Frank watches every open spare-parts PO against supplier SLAs and flags a breach before it becomes a shortage — so machines get repaired, not written off.
Machines repaired, not replaced
When a part can't be sourced in time, the whole machine gets swapped on out-of-warranty repairs. Frank catches the shortage while there's still time to order.
Owns daily stock ordering across all 16 stores — the forecasting Stocky does today, with a purchase-order suggestion for every store waiting before you open.
40–60 daily reports, gone
The per-store, per-SKU forecasting that ran by hand off Stocky — now suggestions are ready before 8am, no report runs.
Owns the inbound delivery timetable — keeps a live schedule of what's arriving and when, and reslots it the moment a driver or supplier calls a change.
A daily job — off one person, out of the spreadsheet
Supplier deliveries were booked by hand over email and phone and tracked in one person's Google Sheet — now they sit on a live timetable the whole warehouse can see.
Under the hood
Every ops tool promises integration. Most bolt on brittle, point-to-point connections that break the moment something upstream changes — and fail silently until a customer notices.
A Handler works across the stack you already run — Shopify, Cin7, Xero, your data warehouse — orchestrating a job end-to-end instead of stitching together one-off webhooks. Fewer moving parts to break, less manual troubleshooting, less operational risk as you grow.
For the steps that really matter, a Handler doesn't improvise. It runs deterministic code tools — with no language model in the execution path. The model decides what to do; reviewed code does the doing. So the judgment is AI, but moving funds from Xero to Airwallex is the same audited connector every time — no hallucination in the money path.
Your Handler doesn't guess its way through your Xero API. It runs the same reviewed connector every time — the way you'd want a careful new hire to.
Reasons
The Handler works out what needs to happen and which step comes next.
Selects a saved tool
It picks a reviewed code tool you've already approved, and the inputs to run it with.
Code executes
Deterministic code runs against your systems. No model in the loop.
The judgment is AI. The execution is deterministic.
Safe by design
The team treats Handlers like junior staff, and the same instinct applies to trust: you scope what a new hire can touch before you hand them the keys. With a Handler those boundaries aren't an HR policy — they're built into the system.
A Handler never holds your logins. A policy-based firewall sits between the code it runs and your third-party systems, so it can use a connection without ever possessing the credential behind it.
Each Handler is scoped to the tools its role needs, split by permission per team and per user. No single omnipotent agent holding every key — you decide which Handler the right people can reach.
Routine work runs on tools you've reviewed. Anything new needs your sign-off first — the same way you'd give a staff member a new login. A human stays in the loop where it matters.
A Handler can only reach the systems you've given its role access to — and only through tools you've approved.
See the full security & governance modelWhy this is different
The AI most companies have today is one or two people with good prompts. It saves those people time. It doesn't change how the business runs.
AI labour is different. It's part of how the company operates, not how individuals operate. The whole team can see it, brief it, hand work to it, hold it accountable.
AI Tools
Faster individuals
AI Labour
A named role ✦
What it is
A subscription each person uses on their own work.
A persistent role anyone in the company can hand work to.
Who sees it
One person at a time. Lives in their tabs.
The whole team. Lives in Slack, in the workflow.
When it's on leave
When that person is on leave.
It doesn't go on leave.
What changes
How individuals operate.
How the company operates.
Access to systems
Each person wires up their own. Credentials scattered.
One permissioned role. No stored credentials, approved tools only.
When it fails
Fails silently in someone's tab.
Escalates, asks, or flags in Slack — visible to the team.
Shared or solo
One seat, one person.
Multiplayer — one Handler the whole team briefs and hands work to.
Under exceptions
Breaks. Needs a human to notice.
Learns the edge cases. Handles what breaks static workflows.
Multiplayer
It's one shared teammate. Everyone briefs the same Handler in the same channel, so context isn't trapped in one person's chat history — unlike a Claude or agent account wired up per person.
When someone's away, the Handler still knows the job. When someone new joins, they inherit a teammate that's already up to speed.
First Hire gets you from "I get it" to Handlers running real work in 60 days.