Krug Studio — Podgorica, Montenegro

We build AI systems
that do real work*

Not demos. Systems that draft the quote, chase the invoice, read the contract and hand a person the decision — built by a studio that designs as carefully as it engineers.

Scroll ↓

Identities we designed — Cuka · Norréa · SESA · GOAT Barber Shop · Slatka Planeta · Disciplina Kičme · Legacy

Client logos designed by Krug Studio
The position [KS® — 01]

Everyone will sell you AI.
Almost nobody will tell you
what not to automate.

We are a design studio that learned to build AI systems by shipping one — a real product, with a delivery pipeline where independent agents argue with each other before a line of code is written.

That process has caught things a solo developer would have shipped: a draft feature that would have silently spent real user currency, a stored credential that should never have existed, a launch metric under-reporting by a factor of four. We would rather find those in a review than in production — and we would rather your system found them too.

What we build [KS® — 02]

01

AI connectors for the product you already have

Your users already ask an assistant to do this work, then copy-paste the answer into your product. We remove the copy-paste — and keep you in control of exactly what the assistant may touch.

  • Scoped tool surface
  • OAuth 2.1 + PKCE, no stored credentials
  • Submitted to the assistant directory

Fixed scope, from €4k

02

Agentic automation for back-office work

We find the two or three processes eating a week of someone’s month, automate the boring 80 %, and hand the judgement call back to a human. Always an audit first — we will tell you if it is not worth building.

  • Half-day process audit
  • Human approves, machine drafts
  • Care plan once it runs

Audit €500, builds from €2k

03

AI-native product, designed and shipped

The whole thing — research, interface, engineering, store release — with AI in the core loop instead of bolted on the side. This is what we did for ourselves, and it is the reason the rest of this list exists.

  • One team, sketch to store
  • Agentic delivery pipeline
  • Design you would actually use

Project based

How we work [KS® — 03]
01 Shrink the blast radius first

Most AI risk is not the model — it is the size of the credential you hand it and the number of tools it can reach. We cut both before writing a feature.

02 Augment, do not replace

A system that drafts and a human who approves survives contact with reality. A fully unattended one gets switched off after its first bad day.

03 The review is adversarial

Plans are attacked by independent agents with different lenses before they are built, and every finding is then challenged again. It is slower per change and much cheaper per mistake.

04 Say what it cannot do

Every system we ship comes with an honest list of its limits. If a thing should not be automated, that is the deliverable.

Proof, not adjectives [KS® — 04]
0

territories drawn for Agresor

0

brand identities shipped

0

% original artwork, zero stock

Our own product

Agresor

A strategy board game about Europe in 1940 — forty-two territories, every one drawn by hand, every card written and typeset here. Nobody commissioned it and nobody art-directed it but us, which makes it the honest measure of what this studio does when left alone.

  • Full art direction, in-house
  • Historical research behind every card
  • In development — playtesters welcome
See the game →
Agresor board — illustrated map of Europe in 1940

What do you dread
every month?

That is usually the thing worth automating. Half a day, we walk the real process with you and come back with an honest ranked list — including the parts you should leave alone.

Book a process audit