# Agentic back office: how AI agents take over your admin

> An agentic back office is one where AI agents do the routine admin and you approve the result. The agent logs hours, drafts invoices, spots overdue payments and finds missing receipts; you decide what is sent, issued or paid. It already works today for drafting, sorting and checking. AI prepares, you approve.

Source: https://back-bone.io/en/blog/ki-backoffice  
Published: 2026-10-16 · Updated: 2026-10-21  
Category: Guides

An agentic back office is one where AI agents do the routine admin and you approve the result. The agent logs hours, drafts invoices, spots overdue payments and finds missing receipts; you decide what is sent, issued or paid. It already works today for drafting, sorting and checking. AI prepares, you approve.

**TL;DR · Key facts**

- **An agent works in steps:** it uses tools and your data to work through a task.
- **You approve:** the AI drafts freely; anything that leaves your business waits for your OK.
- **Start small:** hand over one routine and review it.
- **DE and AT:** AI fits the rules you already know. GDPR, retention and correct invoices apply as usual.

## What is an agentic back office?

It is a back office where AI writes the first draft of every routine task. "Agentic" comes from "agent". Anthropic, the maker of Claude, draws the line like this (as of 08/10/2026): in a workflow, a fixed script steers the AI; in an agent, the AI itself decides which steps and tools it uses to reach a goal.

A chatbot answers "How do I write a payment reminder?" An agent finds the overdue invoice, writes the payment reminder in your tone and puts it in front of you with one button: send or not now.

## Which admin work does AI handle today, and what do you decide?

AI handles repetitive text and routines you can check quickly. You make the calls on law, money and client relationships.

| AI handles it | You decide |
| --- | --- |
| Log time from a sentence | Deciding to send an invoice |
| Draft invoice lines from your hours | Tax treatment of an unusual case |
| Write friendly payment reminders and quotes | Calling a client who is late for the third time |
| Find receipts missing for a payment | Approving payments and bank details |
| Summarise open items for the week | Choosing what a client is charged |

AI prepares, you approve. Whether a client is in default is for the law to decide, see [AI in dunning](https://back-bone.io/en/blog/ki-mahnwesen).

## What are the steps from chat to agent?

Most people go step by step. Stop wherever it's enough.

1. **Chat.** You paste a text into ChatGPT or Claude and ask for help. Good for wording, no setup. For client data, have a quick look at the rules below first.
2. **Instructions and skills.** You teach the AI your way of working once. In Claude, a skill is a folder with a `SKILL.md` file; custom skills are available on Free, Pro, Max, Team and Enterprise plans and need code execution switched on (Anthropic help, as of 08/10/2026).
3. **Connector.** You connect the AI to your real data through MCP, the open standard for this (see MCP in our [glossary](https://back-bone.io/en/blog/ki-begriffe)). The AI now works on your actual hours and invoices.
4. **Agent.** You hand over a whole routine ("prepare September's invoices") and review the result.
5. **One place.** Data, rules and approvals sit in one back office, so the agent never works from stale copies.

Each step has its own guide in this series.

## What does it take: data, rules, approvals?

What makes an agent really pay off:

- **Data in one place.** The fewer places an agent has to search, the more accurate it gets. Keep hours, clients, invoices and receipts in one workspace.
- **Rules written down.** Rates, payment terms, tone of voice, who may approve what. What you wrote down once, the agent applies every time.
- **Approvals built in.** Drafts and sending are different steps. Sending and deleting wait for your approval; bank details change only in the app.

Teams also set **roles**. The AI works with exactly the rights of the person it works for.

## What about GDPR, GoBD and the AI Act in DE and AT?

AI fits the rules you already know. For a small business, these apply:

- **GDPR (both countries).** Client data you put into an AI tool is personal data. If the provider processes it for you, GDPR Art. 28 requires a data processing agreement. Which plans offer one differs by provider; read their terms. Art. 22 limits decisions made only by automated means that significantly affect a person; your approval keeps a human in the loop. Austria adds the national DSG on top of the GDPR.
- **GoBD (DE).** The Federal Ministry of Finance's rules (letter of 28/11/2019, amended 14/07/2025) require that issued invoices stay unchanged (§ 146 (4) AO); the AO keeps invoices for eight years and books for ten (§ 147 AO). An AI-drafted invoice follows the same rules once issued. Mistakes are fixed with a cancellation, see GoBD.
- **BAO (AT).** Austria has no GoBD. The comparable duties are in § 131 BAO (proper records) and § 132 BAO (seven years' retention).
- **AI Act (EU).** Art. 4 has applied since 02/02/2025: organisations that use AI must take measures to support AI literacy among the people who work with it. The amendment Regulation (EU) 2026/1744, in force since 27/07/2026, keeps the duty but no longer requires a guaranteed competence level. For a small team, a short written rule and a brief introduction are a sensible response.

## How do you start this week?

One routine, five days:

1. **Monday:** choose the admin task that annoys you most and happens weekly, for example payment reminders or logging hours.
2. **Tuesday:** write down the rule in three lines: what, when, in what tone.
3. **Wednesday:** try it in a chat with made-up or anonymised data.
4. **Thursday:** connect your tool, or move the routine to a back office where the AI can reach real data.
5. **Friday:** review what the AI drafted. If you would have changed it, add that to the rule.

> My goal is to optimise and automate back-office tasks so that we founders can focus on what matters most.
>
> Automating back-office work with AI agents and AI integrations is the future – that's why I built Backbone!

## How Backbone does it

Your back office in the chat. In the app, or in a chat with ChatGPT, Claude and many more.

1. **Create the invoice** Say what goes on it. You approve.
2. **Who's overdue?** One question, every open invoice.
3. **Receipts matched automatically** Upload the statement. Clear matches assign themselves.
4. **Track your time** One sentence in the chat. The hours are in your week.
5. **Bill your hours** What you tracked goes on the invoice.
6. **Send the reminder** You approve. Only then does it go out.
7. **Review, then approve** The AI drafts. You check it and issue it.
8. **Permissions per person** Everyone gets what they need. Their AI gets no more.
9. **Ask the company wiki** Ask instead of searching. The answer comes from your wiki.

## Frequently asked questions

### What is the difference between an AI assistant and an AI agent?

An assistant answers. An agent acts: it uses tools and your data to work through several steps. A good agent asks you before anything leaves your business.

### Will AI replace my bookkeeper or tax adviser?

It makes working together faster. In Germany and Austria tax advice for others is reserved for qualified people; AI speeds up the preparation and sharpens your questions.

### Is it safe to put client data into ChatGPT or Claude?

It depends on the plan, the contract and what you enter. Check for a data processing agreement and the training settings, or work through a connector so the data stays in your back office.

### Do I need to code?

No. Chat and connectors need no code. Skills are text files.

### Does the AI Act apply to a one-person business?

Art. 4 applies to organisations that use AI professionally, whatever the size. The duty is to support AI literacy. No certificate needed.

**AI prepares. You approve.**

Backbone drafts, you approve. Try it for 7 days, card required.

- [Try for 7 days](https://back-bone.io/en/sign-up)

---

General information, not legal or tax advice and no review of your case. Details on providers and laws can change. For your case, follow your tax adviser.

Checked on 08/10/2026