Classic delivery
A black box
- Every behaviour tweak is a ticket, a wait and an invoice.
- Domain know-how gets translated through meetings and specs.
- Feedback gets lost in emails without context.
- The know-how and the system stay with the vendor.
Methodology
Autonomy is climbed in steps.
We do not drop AI into a company all at once. We start with an MCP server over your tools, continue with an agent that has a clear job, and end with a system that runs on its own. You can stop on any step, and each next one has to be earned by the previous.
This is how all three of our products were built, and the systems we run for clients.
The process
Three steps to autonomy
MCP server
We connect AI to your tools: email, project tool, data, calendar. The Claude you already use suddenly sees your company and can work inside it. A small project with value from week one.
Living example: AI assistantAgent
The AI gets a job and responsibility: it holds conversations, prepares paperwork, watches a process. Guardrails define what it may do alone and where it waits for a person. Outputs are measured, not guessed.
Living example: Video AI agentAutonomous system
The system runs on a schedule without a person. It handles the routine itself; decisions with consequences wait for your approval. Every run has a log, versioning and costs to the cent.
Living example: Autonomous website
You can stop on any step. The next one makes sense only once the previous has proven itself in production.
The difference
Between turnkey delivery and touching the code
Our model
Your expert trains it
- Your person tunes criteria and rules right in the chat, no code.
- Changes are tried in a sandbox, go live with a reason and can be rolled back.
- Feedback attaches its own context, so the developer knows exactly what it refers to.
- The code, the documentation and the trained knowledge are yours.
Division of labour
We build the system, your expert teaches it
Development and training are two different jobs done by two different people. That is why the system improves every week, and developer hours go into development instead of rewriting rules.
Us
- Architecture, connectors and integrations
- Guardrails: sandbox, approvals, rollback
- Measurement: logs, costs, evaluation
- New capabilities, driven by the data
Your expert
- Criteria and evaluation rules in their own words
- Trying changes in a sandbox, promoting them with a reason
- Feedback straight from the work, with automatic context
- The decisions that need domain knowledge
Questions
What others ask
No, but it is the cheapest way to learn where AI actually helps in your company. Connecting to existing tools is a small project, and everything built for it gets reused in the later steps. If you already work with AI, we can join at a higher step.
Less than you expect, but not zero. A few hours at the start for the process audit and access. After that the main role belongs to the domain expert who teaches the agent alongside their normal work: tuning criteria and leaving feedback right in the chat. Not extra meetings; work they do anyway, which the agent now learns from.
They will not. Changes are tried in a sandbox that never touches production. A change goes live only after a promote with a reason, and every one can be reverted in a single step. Everything is versioned, so you can see who changed what and why.
Each step is a project with a price agreed up front; operations is a monthly subscription. The point of the model is that you do not buy judgement tuning from a developer by the hour; your own person does it in the chat. You will get a quote with an exact number within two business days.
Yes. The code, the documentation, the access and the trained rules and knowledge base are yours. No vendor lock-in: if we parted ways tomorrow, the system stays working and understandable.
Claude and MCP (Model Context Protocol) as the primary building blocks, plus your existing systems: email, project tools, data, git. You are not buying a new platform to migrate onto; the AI comes to your tools.
Let's look at where you are.
Forty minutes looking at how things stand. It tells us how far we can take you towards being an AI-native company, and what that will cost you at the end of the day.
- Where AI makes sense in your processes, and where it does not
- An estimate of scope, time and cost
- A first step you can start on right away
- No presentation, no commitment
Jaro Rais
AI consultant and developer
Reach out:
jaro.rais@itashu.coGet in touch and we will come back to you within one working day.
By submitting you agree to your data being processed as described in the privacy policy.