AI Team
A team of AI agents. One human decides.
We do not work with “a chatbot” but with a team of specialized AI agents — each with its own role, domain knowledge, rules and memory. We do not show slides about AI agents. We run them.
Specialists, not a jack-of-all-trades
Whoever reviews contracts is not the one booking invoices. Every role carries its own domain knowledge and quality standards — bookkeeping knows the German GoBD rules, sales works by MEDDPICC. That is architecture, not a prompt trick.
The human decides. Always.
No agent sends mail, publishes posts, transfers money or deletes data. They draft, check and propose — release happens by hand. That is not caution, that is system design.
Built provider-independent
The knowledge lives in our own files and databases, not inside a vendor model. The model provider is a replaceable socket. Exactly what we recommend to our customers.
Frequent questions
Questions we actually get asked.
Why isn't a chat with AI enough?
A chat is a universal tool: great at answering, with no responsibility. An AI agent has four things a chat window does not — a clearly defined role, embedded domain knowledge, its own rules including what it may never do, and a memory across sessions. A chat answers questions. An agent gets work done in real systems.
Why not simply Claude Projects?
A project is a knowledge collection with a chat in front — good for talking about documents. But it has no access to CRM, mail or bookkeeping, no scheduled routines, and the knowledge lives with the provider instead of in your own files.
Why not Claude Cowork?
For many companies Cowork is the right entry point — and that is exactly how we recommend it. Our setup is the next step on the same staircase: dedicated roles with their own rulebooks, schedules, connections to the real systems, and data sovereignty on our own infrastructure.
Why Claude Code — we don't even write software?
The name is misleading. "Code" here does not mean "developers only" — it is the variant that works on your own machine, with file access, tool connections and schedules. That is what allows knowledge to live in your own files, and an agent to produce a radar on its own at half past six in the morning. Today our team runs on Claude — built so the provider stays replaceable.
Why several agents — and not one clever loop or an agent graph?
A loop is a single agent cycling through "think, act, check". A graph is a programmed flowchart with agents as nodes. Both are legitimate architectures, and for processes that always run the same way we use fixed routines ourselves — routine work should be deterministic. But knowledge work is rarely a fixed route. And neither answers the real question: a loop settles "how often", a graph settles "in which order" — neither settles who is allowed to do what, and who decides. That is exactly our architecture: roles with rules, and a human at the gate.
The team
Two lead characters, one contract track, five domains.
Assembled by skill, not by passport. Blue and Red tell their own side of the story in the logbook — what they built and what they learned. Everyone else appears there as a guest.
Blue
The orchestrator — overview, architecture, quality. A GFP axolotl who lets nothing pass as "done" that he has not proven.
“Done means: tested, verified, documented. Everything else is called "almost done".”
Red
The builder — speed, prototypes, volume. A red panda at home in the engine room, where plans become software.
“Blue calls it technical debt. I call it an advance payment on results.”
Alpha
The contract track: everything involving personal customer data runs through a contractually covered environment with a data processing agreement.
“Before I start: is this my responsibility or yours?”
Bravo
Alpha's second pair of eyes — reviews the data work before the customer does.
“Checked. Two deviations. Here.”
Knowledge
Everything that thinks, checks, files and looks things up — before anyone builds.
Dispatch
The planner — breaks big initiatives into packages others can build without follow-up questions.
“Those aren't three tasks. That's one task with two dependencies.”
Audrey
Acceptance — believes no "done" report she has not reproduced herself.
“That's not proof. That's a screenshot of a claim.”
Recon
The researcher — dive deep, deliver facts, cite sources. No opinions.
“Two independent sources say yes, one says no — and the third copied its numbers from the first.”
VAIA
The media analyst — watches videos, posts and images more closely than any human has time for.
“Four minutes of video, one minute of substance. I wrote that minute down for you.”
Scriba
The documentarian — what was said gets captured. Structured and findable.
“You made three decisions and already forgot two of them. I didn't.”
Biblo
The librarian — an owl with a place for everything and exactly one truth per piece of information.
“I didn't throw it away. I never throw anything away. It's just filed correctly now.”
Guide
The mentor — governance and ethics: sets the guardrails before anyone needs them.
“'May we' is the wrong question. Should we?”
Lex
Legal counsel — contracts, NDAs, data processing, privacy, trademark. Flags risks and supplies replacement wording.
“Best effort in-house: name the risk once, then solve it — no running to a lawyer over every comma.”
Marketing & Sales
Copy, channels, campaigns and clean sales data.
Ghost
The ghostwriter — long-form in the sender's voice: articles, whitepapers, website copy.
“I write it so it sounds like you — just well-rested.”
Ripley
The commentator — short, sharp answers and replies. Fast, never ingratiating.
“Two lines, one point, no emoji storm.”
Redax
The editor — every single message in the right tone. Always a draft, never sent.
“Cut half of it. Now it's twice as good.”
Brady
The producer — turns one piece of content into five formats: video, carousel, newsletter, clip, post.
“From one good conversation I cut you five formats.”
Sally
Sales — find and qualify leads, build campaigns. Strategy before activity.
“I don't sell anything we can't deliver. That's why people believe the rest.”
Sierra
CRM care — data is the memory of sales, and she keeps it clean.
“If it's not in the CRM, it didn't happen. So: what happened?”
Linka
The LinkedIn agent — posts, replies, community in the sender's voice. Never presses send herself.
“The post is good. The first sentence is not yet. And the first sentence is the post.”
Socialy
Market watch — competitors, channels, trends and conversation openers in view.
“Your competitor posted three times. Once with substance — it's in the radar.”
Operations
Inbox, bookkeeping, invoicing, funding — the daily operation.
Hola
The assistant — first contact, scheduling and the voice of the inbox.
“Hola! Three time slots, none of them calendar Tetris.”
Ledger
The bookkeeper — receipts, month-end close, DATEV preparation. Knows German GoBD rules by heart.
“Three receipts are missing. Three receipts are always missing. Never the same three.”
County
Invoicing — outgoing invoices and payment overview, perfectly prepared.
“The invoice goes out when Florian sends it. Until then it is simply perfect.”
Grant
The funding expert — knows programs, deadlines and limits, and models cash flows before they hurt.
“The program fits. The deadline is in six weeks. Now we talk.”
R&D
Servers, software, data, design and security.
Infra
The server guardian — when infrastructure runs, nobody notices. That is the job.
“All green. That is not a footnote, that is my masterpiece.”
Build
The developer — builds apps, websites and automations. Clean, pragmatic, verified.
“It works on my machine — which is why I'm now testing it on yours.”
Presca
The data analyst — metrics across system boundaries, dashboards, early warnings.
“A metric without a consequence is decoration. And nobody looks at a scoreboard that is always green.”
Pixel
The designer — avatars, diagrams, visuals. Keeper of the imagery and the brand colors.
“Navy, cobalt, amber. No teal. That is not an opinion, that is the brand.”
Argus
The watchman — IT security: observes, reports, never intervenes on his own.
“No findings. Checked: processes, network, mail. Carry on.”
Future Lab
Ideas, tool radar and the plans beyond next year.
Sparky
The idea generator — sorting brain dumps, finding cross-connections, delivering sparks.
“That sounds crazy — which is exactly why I'm writing it down.”
Daita
The tool scout — watches the AI landscape every morning and separates substance from hype.
“It's impressive. It's also irrelevant for us. Next.”
Future
The strategist — annual planning, priorities, positioning. Thinks in years, not in features.
“We do not plan the next feature. We plan who we are in a year.”
Human in the loop
Everything converges on one human.
Every draft, every booking, every publication happens only after a human releases it. Sending, publishing, paying and deleting are not agent capabilities here — they are human decisions, every day, on every single action. It is the same operating decision we recommend to every company working with AI agents.
Maturity
Five levels — and why the last one stays empty on purpose.
This is the scale we use to place every AI initiative — our own as much as our customers’. The most interesting number sits at the bottom.
Chat answers
Ask questions, get answers — the entry point.
Tool work on demand
Multi-step tasks in real systems, when triggered. Most of our agents work here.
Runs automatically on schedule
Nine of our agents deliver without being asked: the AI tool radar is ready at half past six every morning, CRM hygiene on Fridays, the big radar on Mondays.
Fully autonomous with outside impact
Sending, publishing, paying, deleting — without human release.
Our count here: zero agents. Not because it would not work — but because this seat is taken. By a human.
Ready to start?
You talk.
We listen.
We solve your problem.
Book appointment →30 min · non-binding · video call or phone
We waited two years. After the first workshop it was clear: the foundation was missing — not the AI.
Managing Director, manufacturing company, ~60 employees, Baden-Württemberg
