We build AI agents that run in your daily work: preparing invoices, setting up systems, creating content. Not a classic agency: no retainer, no lock-in, you own the code.
An AI agency, also called an AI service provider, plans and builds AI applications for businesses. For us, that mostly means AI agents.
An AI agent is a program that uses a language model to decide its next step on its own. It uses tools to do so: it reads emails, queries data and writes documents.
The difference from AI automation: automation follows fixed rules, an agent handles exceptions. That is why we build automation wherever a rule is enough. It is cheaper and more stable.
| Classic AI agency | MacNorris |
|---|---|
| Starts with a workshop and a concept | Starts with a prototype for your real process |
| Retainer with a minimum term | No retainer, no minimum term |
| Code stays with the agency | You own the code, including prompts and access |
| Tied to one model vendor | Swappable model, local too |
| Runs on the agency's platform | Runs in your cloud or on your server |
| After launch you stay dependent | Your team takes over, or we stay if you want |
We show you the details of individual agents in a call.
Reads invoices from the inbox, checks them and prepares the booking. The accounting team only approves.
Run on your own hardware instead of the cloud. The data stays in-house.
Walks you through a question catalogue, then sets up the tool following best practices. Instead of docs nobody reads.
Turns an idea into a finished comic with images and voice.
Six statements, answered honestly with yes or no. Takes a minute, and nothing is stored.
We go through one real case with you, including every exception.
The agent works with real examples. You see the result before you invest further.
Test cases, human approval, a log of every decision.
We set up operation on your side, in your cloud or locally. Plus docs, access and prompts. Your team takes over.
Five points that show you after the first call whether it fits.
Good agency: It asks about your process before naming a tool.
Warning sign: It starts with a platform demo.
Good agency: It calculates the monthly running costs for you.
Warning sign: "Tokens cost next to nothing."
Good agency: It builds in human approval.
Warning sign: "The AI doesn't make mistakes."
Good agency: You get the code, prompts and access.
Warning sign: The agent only runs on their platform.
Good agency: It tells you when automation is enough.
Warning sign: Every problem is an agent problem.
Ten questions for any AI agency, each with a good answer and a warning sign. Plus a running-cost calculation, the legal side on one page and a list to tick off.
We show our projects on MacNorris TV, mistakes included.
AI comic generator built from scratch: Notion as input, FastAPI as the hub, Higgsfield for video, ElevenLabs for voice. Stack, workflow, costs and what broke.
Higgsfield: AI Video, Soul ID and What It Costs
ElevenLabs: AI Voices and Your Own Voice via API
Notion: What It Does, Costs and When It Fits19 min video
Not legal advice, but the current state with sources.
Anyone who interacts directly with an AI system must be told so, unless it is obvious. This applies to agents with customer contact, for example in chat or by email.
AI Act Service Desk, Article 50 (opens in new tab)Anyone who uses AI supports the AI literacy of their own people. Since the Digital Omnibus, this means measures, not a set level for every person.
AI Act Service Desk, Article 4 (opens in new tab)If a provider processes personal data on your behalf, you need a data processing agreement. That includes the model provider. If the model runs locally on your side, that agreement is not needed.
EUR-Lex, GDPR Article 28 (opens in new tab)An AI agent has two cost blocks: building it once and running it every month. Running costs cover the model, hosting and maintenance and depend on the number of cases. We do this calculation with your real numbers before we start. You get a range for the build after the first conversation, not before.
AI consulting clarifies where AI helps your business and where it does not. The AI agency builds the agent once it is clear which process it is for. The two often overlap. If you already know your process, we start building straight away and skip the analysis.
Automation is enough when the process follows fixed rules, such as the same form every time. An agent pays off when inputs vary and someone decides based on text today. If automation is enough, we tell you. It is cheaper and more stable than any agent, and easier to maintain.
With you. We set up the agent in your environment: in the cloud you already use, such as Azure, AWS or Google, or with a local model on your own hardware. Locally, the data never leaves your building. Which option fits depends on your stack and on the data the agent processes.
You. You get the code, the prompts, the test cases and all access credentials. The agent runs on your accounts, not on a platform of ours. Your team can take it over. If you prefer, we keep handling maintenance and new features, with no minimum term.
Yes, MacNorris is based in Hamburg. We do most of the work on an agent remotely, with clients across Germany. For the start, for example to look at a process on site, we are happy to come by in Hamburg. More on that on the AI consulting Hamburg page.
Show us one real case. You get an honest answer: agent, automation or nothing at all.
