You have two ways to create an agent: You start with a Pickware template, for example the AI Warehouse Manager, or you describe the task in your own words. Both work when you create the agent in the chat. To learn how creating an agent works, read the article What are Pickware AI Agents?

When you create an agent, the chat shows these templates under Or start with a template:

AI Warehouse Manager: prepares the working day every morning, covering shipping work, bottlenecks and open decisions.

AI Revenue Controller: reports the previous day's revenue and orders per Sales Channel every morning, compared to the same weekday of the previous week. After a weekend, it reports each day individually.

AI Sales Analyst: evaluates the previous week every Monday for the weekly meeting, covering Sales Channels, top sellers and risers.

Expected Goods Receipts: announces every working day which supplier orders are expected today and which are overdue.

Master data check: EAN and weight: reports every week which active products have no EAN or no weight, so that scanning and shipping labels work. Amounts that a variant inherits from the main product count as present.

During setup, the chat only asks for what it needs to know about your shop, usually the time, and suggests a value for it. You can customize the agent afterwards like any other.

Every morning, the AI Warehouse Manager gets an overview of the situation in your warehouse: Where is the flow of goods stalling, what has to ship today, where are there bottlenecks? From this it decides what matters today and puts a deliberately short warehouse briefing in your inbox. So it doesn't just repeat key figures, it assesses the situation.

Because the briefing follows the data, it doesn't look the same every day. Use it as your start into the day: In the chat attached to the message, you can ask follow-up questions about individual points.

For the briefing, the Warehouse Manager evaluates a lot of data, so a run often takes 20 minutes or longer. During setup, the chat suggests 6:30 a.m. on working days, so that the briefing is ready before work starts.

Many checks in day-to-day business are quickly explained but tedious: "Is my margin slipping somewhere?", "Has someone forgotten a purchase price at goods receipt again?" That's why most custom agents work according to the same pattern: You define a rule that should apply in your company, and the agent checks it regularly.

The most important sentence in these tasks is often the last one. If your task asks for a list or a report, the agent reports back on every run, even if everything is fine. If you want it to report only when it finds something, state that explicitly.

You can take over the following examples and adapt them to your numbers.

New purchase prices constantly enter the system via goods receipts. Your sales prices stay where they are, and at some point a product falls below the margin you had planned, without anyone noticing.

Check the margin of all my products and product sets every day at 7 am. Calculate using the net sales price in the Sales Channel Hauptshop. List all products and product sets with a margin below 25%. If all of them meet the condition, don't message me.

Margin can be calculated in several ways. So state in the task which sales price and which Sales Channel you use for your internal calculation. For a one-off evaluation instead of ongoing monitoring, the chat is the faster way: Calculate margins and simulate price changes with Pickware AI

The purchase prices on your goods receipts determine the valued stock. But this very field is often forgotten in daily work. An agent that starts with every completed goods receipt catches this:

On every completed goods receipt, check whether a purchase price is stored for all line items. Only message me if the purchase price is missing for at least one line item.

If a supplier order arrives late, customer orders are often waiting for exactly this merchandise. An agent shows you every morning where this is the case:

Every weekday at 6:30 am, check which supplier orders have passed their expected delivery date. For each supplier order, list the open customer orders waiting for these goods, oldest first. Only message me if there are such supplier orders.

The agent can also prepare a text for you to inform affected customers. You have to send it yourself.

Some details only become apparent once the parcel is already packed, such as a missing phone number that the freight carrier needs:

Every weekday at 7 am, check all open orders with freight shipping to see whether a phone number is stored. List the orders without a phone number and draft a short message for each order to ask the customer for it. Only message me if you find such an order.

A DATEV payment should exist for every paid order. If one is missing, this often only becomes apparent during the export.

Every day at 7 am, check whether a DATEV payment exists for all orders with the payment state "paid". Only report the orders for which no DATEV payment exists.

When a remaining shelf life becomes critical is something each product range decides for itself. Instead of a fixed traffic light, you define your own limits and have them presented to you sorted in the morning:

Every morning at 7 am, check the remaining shelf life of my batches and sort them into three categories: critical with less than 20 days remaining, to watch with 20 to 40 days, and non-critical beyond that. Only message me if at least one batch falls into the critical or to watch category.

If you offer the same product through several Sales Channels, the prices drift apart over time, usually unnoticed, because nobody compares them regularly.

Every Monday at 7 am, check whether products that are available in multiple Sales Channels have the same sales price in each of them. Only report the products whose sales prices differ from one another.

This agent goes one step further than pure monitoring: It calculates the order quantities for you.

Every weekday at 6:30 am, check my stock and suggest order quantities for products below the reorder point, grouped by supplier. Take open supplier orders into account.

The agent doesn't create the supplier orders itself. If a suggestion works for you, write in the chat attached to the message, for example, "Create the supplier order for supplier X like this." The chat then creates it with your user permissions, if your plan includes write actions in the chat.

Phrase restrictions as clearly as possible. A requirement like "maximum ten line items per suggestion" can also be understood to mean that a large order is split into several orders of ten line items each. So look at the result of the first run and refine the task if it was implemented differently than intended.

Start small: Start with a rule you already check in your head, and only expand the agent once the basic function works.

Put everything in the task: An agent can't ask questions during a run. State thresholds specifically: "under 25% margin" is a usable requirement, "too low a margin" is not.

Start before work begins: A run can take 20 minutes or longer. Set the schedule so that the result is ready when you need it.

Refine: Via Customize on the detail page, you can change the task at any time via chat. This isn't a stopgap, it's the intended way to get the right result.

Cross-check results: Treat the reports as a starting point and check critical amounts in the Admin before you act on them.