The Agentic Engineering Company
AI agents in everyday work, with every task measured for cost and quality
We build AI agents and models with your teams, for software development and for the processes behind your business. They run on infrastructure you control. Each change is tested on work you have already done, and we keep it only when it is cheaper and at least as good. When we are done, your team owns it and can run it without us.
What we work on
Coding agents
We set up coding agents that know your repositories and stay within budget, and test each change on code you have already merged.
Help with Claude Code, Codex, Cursor or CopilotAI agents and process automation
Agents take on the routine part of complex cases, such as claims, invoices or service requests, and pass anything uncertain to a person.
Custom model training
On a single, well-defined task, a small open-weight model trained on your data can outperform a large general model and cost far less to run.
AI engineering and architecture
We add AI to the products you sell and build the serving, retrieval and connections to your systems that it relies on.
You can start with any of them.
Before and alongside Cloudsail, our team has built software and AI systems for
Find your starting point
Each situation links to the service where the work begins.
What changes for you
The cost of each task, in euros
You see what an accepted change, a handled case or a model request costs, by team and task, with the acceptance or error rate beside each figure.
Decisions backed by your own data
Every option is tried on your past work first, and you get a plain recommendation, including when to stop.
A setup your team owns
Configuration, test sets and runbooks stay with you, written so your team can run them without us.
How we test each change
Each kind of work is tested differently, always on your own data.
Coding agents
We replay changes your team has already merged, against tests the agent cannot see.
Process agents
We run past cases with known outcomes and compare the agent's answer with the real one.
Custom models
We compare the new model with your current one on cases it was not trained on.
Coding agents: cost per finished task
Before and after a change to the setup
€62000 to €49000 a month
Rework 11% to 10%
Example with synthetic data, not a client result.
Working with us
A first project with one clear question
A single process, team or model, with the measures agreed before we begin.
Built together with your people
Our engineers work in your teams on real tasks, and your colleagues learn the setup while it is being built.
Yours to run
You get the code, test sets and runbooks, and we step back once your team is comfortable running them.
You can bring us in for a defined project, have our engineers join your teams part-time or for longer placements, or book training on your own work. Training runs remotely or on site in Germany and Poland, in English, German or Polish.
The people behind the work
The engineers who scope your project also build it with your team and hand it over.
Mickey Graf
Founder and AI systems engineer
Mickey Graf founded Cloudsail and still writes code himself. He has run machine learning in production since 2017, starting with image classification at trivago, and went on to build cloud platforms, voice AI and enterprise assistants. He speaks German, English and Polish.
How we handle data
Work happens in your environment, and every data flow is agreed with you first.
No model or tool vendor pays us, so our advice follows what works on your data.
Each contract includes a data processing agreement.
Reports cover teams and tasks. We never report on individual employees.
We handle your works council's requirements before anything goes live, and prepare the documentation your data protection team needs.
Recent writing
All postsThe latest posts, each with the date it was last checked against the vendors' own documentation.
Common questions
Is the first conversation free?
Yes, there is no charge.
Which models and tools do you use?
Whichever suit the task: commercial models such as Claude, GPT or Gemini, open-weight models on your own hardware, and the coding agents your teams already have.
Do your agents decide on their own?
Only within limits you set. Each agent starts by preparing work for a person to decide, and takes on more only after you have reviewed its results with us.
Can everything run on our servers?
Agents, test sets and open-weight models can. Commercial models are used only through data flows you have agreed to.
How do you report results?
By team, task and model, with cost and quality side by side.
What if a change doesn't pay off?
Then we leave it out. You keep the test set and the measurements either way.
Can you promise a saving?
No. We report what the tests show and how certain each figure is, and sometimes they show no saving.
Talk to an engineer
Bring one piece of work you would like to improve. The call takes 30 minutes and needs no access to your systems or your data.
Look for the draft in your email program
Your email program should now show a draft addressed to us. It stays a draft until you send it, and this page cannot tell whether the draft opened or your email arrived. You can still edit the form, or write to miki@cloudsail.com yourself.