isarcode

AI

Agentic AI: the quiet sort that actually does the work

At a glance

  • Agentic AI does not merely answer; given an objective, it works through the steps itself, plugged into the systems you already run.
  • The firms that gain most are not the giants but lean teams, where an hour rescued from admin is felt the same afternoon.
  • isarcode builds it on n8n, hosted in the EU, GDPR-sound, versioned and reassuringly dull: it does not fall over.
  • The only real prerequisite is a system with an interface (an API). Most have one, and where one was missing, we have added it in the course of a project.

A few weeks ago, over a flat white that had travelled rather further than the person who served it, a managing director from a manufacturing firm outside Rosenheim asked the question we now field most weeks. “Everyone says ‘agentic AI’. Is it something a company our size can actually use, or is it simply another word for the conference circuit?”

A fair question, and one that deserves an answer free of the usual fog.

Ordinary AI answers. Agentic AI acts.

The AI most people have met is a courteous assistant. You ask, it replies, the exchange ends there. Useful, occasionally charming, fundamentally passive.

Agentic AI is the version that gets up from the desk. You hand it a goal rather than a question, and it works through the necessary steps on its own: pulling data from your systems, deciding by rules you have set, triggering whatever comes next, then reporting back. Less “how much of part X is in stock?” and more “watch the stock, reorder when it runs low, and tell purchasing you have done so.”

The difference is not raw cleverness, then. It is agency, exercised inside boundaries you draw.

Why this lands harder in the Mittelstand than in the boardroom

Large corporations have departments for this sort of thing. Mid-sized firms tend to have the opposite: small teams in which one capable person wears three hats before lunch. Which is precisely where an agent earns its keep.

  • The clicking-about disappears. Shuttling data from one system to the next, assembling the Monday report, fielding the same five enquiries: the work that ties people up and produces nothing.
  • A small team starts to punch above its weight. You need not hire to handle more volume. You automate the routine and keep your good people for the work that genuinely wants a human.
  • Knowledge stops walking out of the door. The process that today lives only in one colleague’s head sits, documented, inside the agent. Holidays become less alarming.

None of this is speculative. It is doable the moment a system exposes an interface, and most do.

The honest part: where agentic AI has no business

Trust is not built on overstatement, so here is the unglamorous half. An agent is not a self-driving miracle. It wants clear rules, firm boundaries and a human to nod through the consequential decisions (the polite term is human-in-the-loop). For judgement calls with no rule beneath them, it is the wrong tool, confidently wrong, which is worse.

The right place to start is the dull, repetitive, well-defined task. You have more of those than you would care to admit.

How we build it

We do not buy a black box and drape it over your business. We build on n8n, an established open-source automation tool, and bring AI in only where it changes the outcome, rather than wherever the fashion points.

Three things we insist on:

  1. GDPR-sound. We host in the EU, usually in Germany. Your data does not wander. AI and data protection are not at odds when the architecture is set up properly from the first line.
  2. In your infrastructure or ours. No operations team in-house? We run it for you. Prefer everything on your own servers? We set it up there.
  3. Versioned and resilient. Everything moves through a proper pipeline (development, staging, production), tested before it goes live. No house of cards.

Our own monitoring system, Hawkeye, is built exactly this way. We sell nothing we do not run ourselves.

A real example, beans included

For a coffee roastery we automated the production planning: which beans, in which blend, in what quantity, and when they go into the roaster, all derived from the actual orders. It draws on the online shop, the inventory system and the B2B sales channel at once. Data that used to sit on three separate islands became one continuous, automated flow.

The pattern is not specific to coffee. It transfers to almost any sector the moment the systems can talk to one another.

Local enough to drop by, seasoned enough to play with the big names

isarcode is a Munich agency with an office in the Chiemgau. For you that means a partner within reach, who nonetheless works on international projects and holds its own against the large agencies. Local proximity, international standards. We did not grow up on the meadow, but we are close enough to come round for a coffee, whether you sit by the Chiemsee, in Rosenheim or in Munich.

Frequently asked

Do we need our own IT department for this? No. Where it is wanted, we run the operations side for you.

Isn’t this only for large companies? Quite the reverse. Lean teams from around ten people gain the most, because every hour saved is felt immediately.

What happens to our data? Hosted in the EU, usually in Germany, on a GDPR-sound architecture. You keep control.

Where does one begin? With a discovery workshop. We look together at which processes are worth automating, and turn that into a concrete plan with clear stages. No blanket promises.

And then?

Have a process in mind that quietly eats your team’s week? Let us talk for half an hour. We will tell you honestly where automation pays off, and where it does not.