Separate concerns
Instructions, knowledge, conversation flow and calculation live in separate places. Once mixed, nothing can be changed safely.
01Who’s allowed to change what your AI says?
Putting AI into production creates questions the prototype never had to answer: who can change what the system says, who can see what it costs, and who notices when its quality gets worse. We build those controls in.
The question that can’t be asked early enough
Most AI projects don’t fail on the technology. They fail on what comes afterwards: who can change what the system says, and who sees what it costs.
Until that’s resolved, every AI system stays a special case only its builder can touch. We build steerability into the architecture, because retrofitting it later is a rebuild.
Why no client names2x10
Our clients hire us to build what gives them an edge. Confidentiality is part of the job. Publishing the details could give that edge away. We’ve delivered in regulated environments and on projects worth hundreds of millions.
What we've built
The control layer inside a client's product. Conversation flows are built visually by the subject-matter team rather than written in code, and behavior is changed in plain language.
Model choice, cost tracking and budgets are part of the interface, not of the invoice. New content and new behavior go live without a developer.
A content system where throughput is deliberately throttled, budgets are set per period and alerts fire before a limit is reached rather than after. Coverage is visible at a glance: what has been produced, what has not, what it cost.
Four properties
What control actually means
Evals first
Evaluation runs show before rollout whether a change made something worse. Without them, every improvement is a guess.
02Cost ceilings
Budgets, throttling and warnings inside the system, not a surprise in the monthly close.
03Swappable models
The model is a configuration, not a foundation. Providers have outages and prices move, and you inherit both.
04Production questions2x10
What makes them hold up
Governance decides who can change the system later. Three questions, all asked before the build.
- When does steerability get built?At the start, or it becomes a rebuild nobody has budget for.
- What does a frozen system cost?Over time it freezes the business that sits on top of it, and that never appears on an invoice.
- Is governance a document?It is a property of the software: what isn’t permitted has to be impossible, not forbidden.
What we could build for you
- We build AI systems whose behavior can be steered without a development team.
- We rebuild existing AI systems so your own specialists can change them without us.
- We make costs visible and cappable before they become a topic in a board meeting.
- We introduce evaluation runs that show degradation before your customers do.
- We remove provider lock-in by keeping models interchangeable.
- And whatever you have in mind that AI and solid software engineering could make possible.
Under the hood
Control · visual flow editors · plain-language instruction documents · knowledge management with staleness detection · Measurement · evaluation suites · regression testing · quality monitoring · Operations · cost tracking and budgets · model routing · audit logging · LLMOps
Most engagements start small: talking it through, a few weeks of engineering, something real you can judge. If it works, we take it further.If it doesn’t, you find out early
More work2x10
Novid20 · Our own product
Millions of users
View themeAI Agents
Beyond the demo
View themeChatbots
More than a chat window
View themeN° 06 Contact
Tell me what you’re planning
No form, no ticket system. Email me directly and you get me, not an inbox someone else manages.
Moritz Miedler · Founder, replies personally2x10 Technologies · Viennahello@2x10.com