The Agentic Engineering Company
Coding agents are only as good as the setup around them
Cost per accepted change and review load depend largely on the harness: repository instructions, tool access, permissions, verification and model routing. We build and run it with your platform team and evaluate every change against replays of your merged work. A change stays only if cost falls and acceptance holds.
What we do
Three service lines, all measured on your own data and run inside your infrastructure.
Agentic engineering
Harness, model routing and cost control for coding agents such as Claude Code, Codex, Cursor and Copilot, measured on replays of your merged work.
All coding agent servicesCustom Model Training
Open-source models retrained on your data for a specific task, more accurate than frontier models at a fraction of the cost. Measured on your own test set before anything ships.
AI engineering and architecture
Custom agents, MCP servers and multi-agent systems, from engineers who have also built the infrastructure and trained the models underneath.
Our team has built software and production AI for
Start from your situation
The situations teams most often bring to us, and where the work starts in each case.
Everything around the agent
Our engineers design, build and run it inside your infrastructure, with the models and tools you already use.
What could you save?
Agent spend today
€600,000 a year
At an assumed saving of 19 percent
An illustrative figure. The actual saving is determined from replays of your merged work.
€114,000 a year
An estimate. On a call, we go through which savings apply to your setup.
Measured on your code, not on benchmarks
We replay merged changes against the current setup and each alternative, with held-out tests. Savings are reported only where acceptance and rework hold.
Spend per finished task
Sample result, synthetic data
€62000 to €49000 a month
Rework 11% to 10%
Who you work with
Engagements are run by engineers who work part-time inside your team, from scoping to handover.
Mickey Graf
Founder, AI systems engineer
Building production machine learning systems since 2017, starting with image classification at trivago. Since then: cloud platforms, voice AI, enterprise AI assistants and open-source tooling for coding agents.
Individual Claude Partner for Claude Code. Works in German, English and Polish.
Your code stays yours
We are independent of model and tool vendors. No vendor pays us, so our recommendations follow what works for you.
We work inside your environment.
A data processing agreement comes with every contract.
Reporting by team, never by individual developer.
Works council requirements handled early.
From the blog
All postsWritten by our engineers and checked against each tool's own documentation on the date shown.
Questions
What does the first call cost?
Nothing. It is a 30-minute conversation with one of our engineers, with no access to your systems and no data shared.
We cannot share code with an outside party.
The first call needs no data and no source code. For work that touches repositories or infrastructure, we agree the data flow with you before it starts, including which model providers are involved and what your contracts with them allow.
We already work with our vendor's engineers.
They help with their own tool. We compare tools and models on your code, and route work to whatever performs best for the money.
How much of our team's time does this take?
Thirty minutes for the first call. For consulting work it depends on the scope and on who is available on your side, so we agree both with you before work starts. There is no fixed timeline.
Do you guarantee savings?
No. We measure savings on your own work and say how certain each figure is.
Do you report on individual developers?
No. Reporting is aggregated at team level. Individual usage data says little about the quality of the work, and in Germany it falls under works council co-determination.
Do you only work on coding agents?
Coding agents are most of our work. We also retrain open-source models on client data for specific tasks, and design and build custom agents and MCP servers. The approach is the same: measured on your own work, inside your infrastructure, and you keep what we build.
Talk to an engineer
A 30-minute call with one of our engineers about your coding agent setup, what it costs and where it can improve. No access to your systems and no data shared.
Check your email program
We tried to open a draft in your email program. Nothing is sent until you send it, and this page cannot tell whether the draft opened or the email was delivered. The form stays editable, and you can also write to miki@cloudsail.com directly.