Drafts and Replies

Everyday tasks like drafting documents or answering requests are handled directly by the AI, drawing on information from the web. An offer, a reply email or a first draft is ready in a few minutes, so nobody starts from a blank page. A person in the office still reviews and sends it.

Planning With Your Own Data

For deeper tasks, the AI reaches your company’s own data directly through MCPs, for example from scheduling or bookkeeping. A plan then comes from the real figures and appointments already sitting in the business, not from guesswork. The result is more accurate because it is built on your own data.

Routines That Free Up Your Head

Fixed routines run through your MCPs on a set schedule by themselves, without anyone triggering them by hand each time. A weekly report or a regular check then happens on its own in the background. That frees up your head and lets you do your job from anywhere, in the office or on the move.
Same access
the agent works inside the access you already have
IT plus you
an administrator and the account owner both switch it on
Stays in your tenant
content is read exactly where it already lives, in your own region
From real work
picked together with whoever already does it every week

Answering, Planning, Offloading

Drafts and Replies 3 steps
01

The AI Works Directly in Your Files

The AI gets access to the folder where your documents live and can draft texts or answer requests straight from there. Fixed rules for how a document should look are set once and then followed automatically. That is where the work starts: right where the files already are.
02

Finished Files Instead of Half-Done Work

You hand over the whole task: which documents to read, what to research and in which format the result comes back. Word, Excel and PowerPoint files come back in the layout you already use. Whoever owns the result reviews it at the end and sends it on.
03

Programs With No Direct Link Are Reachable Too

For a supplier portal, an older program or a screen with no direct connection, the AI works the way a person would: it reads the screen, clicks and types. Quirks of that particular screen are noted once and then handled automatically every time.
Planning With Your Own Data 2 steps
04

Access to Mail, Calendar and Files

Mailbox, files, calendar and Teams become available to the AI, under the same access the signed-in person already has. Two approvals switch it on: one from IT, one from the account owner. Real plans then come from this data instead of guesswork.
05

Your Own Programs Supply Data Too

An internal program, a database or industry software gets connected to the AI through an MCP, described in the terms the office already uses. The program itself stays unchanged, the AI just gets a way in to plan with the real figures.
Real Work Becomes a Fixed Routine 3 steps
06

Show It Once, Then It Runs on Its Own

A task is done by hand once together with the person who knows it. What happens is recorded in the right order, including everything that otherwise costs time. That becomes a fixed routine that runs by itself from then on.
07

Readable by Anyone on the Team

Every routine exists as plain text: when it starts, which steps run in which order, and which details only get pulled in when needed. That way anyone on the team can read what the AI does in each case, not only the person who set it up.
08

The First Routine Is a Monday Task

The task chosen is one that already comes up every week, picked together with the person who does it today. A weekly check built this way then runs on schedule by itself and only speaks up when something actually needs attention.
Routines That Free Up Your Head 2 steps
09

A Person Keeps the Overview

Anything that changes a live system happens deliberately and is checked afterwards. Pure reports stay pure reports, and access details never sit openly in the text. Over time, checking eases up routine by routine, based on how reliably each one runs.
10

Ready to Go for Whoever Owns the Task

A finished routine sits ready in the company's own storage, so the task owner starts their own run, whether in the office or on the move. For a whole team, several of these can be bundled into one single, ready-made package.

Frequently Asked Questions

Do we have to replace the software we already run?
No. Microsoft 365 is reached through Graph inside your own tenant with delegated permissions, so Outlook, SharePoint, OneDrive, the calendar and Teams keep their licensing and their permission model. Two approvals switch it on: an administrator in Microsoft Entra, and the account owner in Claude. Systems outside the tenant get a connector, and applications with no API are driven through the interface a person uses.
Do we need a developer to run this day to day?
Not for running it. Editing a procedure is editing a text file, which puts it inside reach of a technically curious business user, and the desktop packaging lets the owner of a task start their own runs. A developer is needed where an internal system ships no connector of its own, which is a bounded piece of work.
Is our data leaving the country?
Content stays in the region your tenant is provisioned in, and the agent sees what the account it acts for may see. The integration reads in place, so no second copy is made. For services outside the tenant, and for the model provider, residency is settled per service in the compliance pass.
What is left if the person who set this up leaves?
The skills, the connectors and the rules files sit in your own storage under your own accounts, as text your people can open and change. Every integration signs in through the directory, and everything runs on company machines and company accounts. The handover test is whether somebody else runs Monday's job on Monday from the file alone.
What is MCP, exactly?
MCP stands for Model Context Protocol, an open standard that describes an existing program or data source to the AI as a tool. An internal program or industry software is connected to the AI through it, in the terms the office already uses. The program itself stays unchanged, the AI just gets an additional way in.
Is an MCP connection difficult to set up?
The first connection typically costs more time than expected, because the questions are settled for the first time. Every further app connects faster afterwards, because the same approach gets reused. Each connection is also fully built and tested before actual access to the system is granted, and the existing program stays unchanged throughout.