We take on software projects and carry them out with our own Agentic Factory, an AI production line in which different models cross-check every step. From the first question to acceptance, Andreas Korol looks after you personally.
You commission a feature, a refactor or a fix. Each task is clearly defined, commissioned individually and has an unambiguous definition of acceptance. We deliver a pull request to your target branch; you do the merge. After acceptance you are free to decide whether to continue.
You have an idea, but no code and no development team. We turn it into an MVP with a fixed scope at a fixed price, set after we scope it together, from €3,000. You accept it on the running product; after payment we hand over the repository and the rights. For companies we also offer operation and hosting on request, as a separate contract with no SLA and no 24/7 commitment.
You describe the idea. In a short scoping step we put the features, the acceptance criteria and the fixed price in writing. You get an app for iPhone and Android, submitted for review to the App Store and Google Play, or a web app for the browser, in each case with source code, usage rights and setup on your own accounts.
One core flow with up to five screens, as a mobile app for iOS and Android with submission to the App Store and Google Play, or as a web app. Plus simple design, tests, repository and evidence pack.
Total price including 19% VAT.
Several flows scoped to fit, as a mobile app, a web app or both, with user accounts and integration with your existing systems on request. With tests, store submission, repository and evidence pack.
Plus 19% VAT (€5,950 gross).
The developer accounts with Apple (USD 99 per year) and Google (USD 25 one-time) are in your name; Apple and Google decide on approval in their stores. For a personal Google account, Google requires a test with at least 12 people over 14 days before the first release; you provide the testers. Not included are payment features, hosting and operation, and further development after acceptance. Updates to maintain security and functionality are included for 12 months with the App Starter; adaptations to new major versions of iOS or Android are not.
You describe what should come out. That becomes a machine-readable plan with goals, boundaries and acceptance criteria. Building happens in small steps through fixed checkpoints. Every critical stage is checked by a different model before anything is accepted.
Even before building, a deterministic code gate checks the plan: every building block has to resolve its inputs and outputs cleanly, with no room for a model's judgement. After that, every build step passes through several roles: a quality scan, several adversarial review rounds by different models and, at the end, an integration architect. No model reviews its own work. Every change is classified by type and impact: a small text change takes the short path, a security-critical one the long path with additional review rounds.
Every run leaves friction behind: a finding that came too late, a review step that found nothing, an instruction that was misunderstood. These become machine-readable operating rules that steer every following run; today there are just under 900. Which model takes which role we decide based on our own measurements per role and task. That way the factory gets better with every project, even without a new model.
You don't have to read any code, but you can. Taking over the result remains your step: the merge for codebase projects, acceptance for the MVP.
Review is part of production. We do not guarantee that it finds every defect. For authentication, payments and personal data we also recommend an independent review.
We work with AI services from several providers, currently including Anthropic, OpenAI and GitHub. Which of them see excerpts of your code on your project is set out before the start in our confidentiality agreement. As a rule we do not process personal data in the course of this; if any turns up, we stop. Automated production runs with no human at the controls are only available with a data processing agreement.
We use these services' subscriptions, with model training switched off in the accounts. There is no separate data processing agreement with the providers in this setup. For most projects without personal data that is sufficient. Factor 2.2 on the measured effort.
If your policy requires a data processing agreement or a contractual exclusion from training, we work through the providers' APIs under their business terms and data processing agreement. Factor 3 on the measured effort.
A codebase project costs the verified production effort × 2.2, or × 3 with a data processing agreement. What we measure is the actual token consumption per model, priced at current API rates. Scoping, review rounds and approval are covered by the factor; there is no second invoice for management or orchestration. You are buying a result; consumption is how it is measured. The method is fixed before the start; the amount follows the measured effort.
Self-registration, email verification and password reset — including session invalidation on reset, rate-limited requests and emails in two languages.
~€1,000 · ~2 days
Subscriptions, seat limits and webhooks with the payment provider; VAT ID matched against the country of taxation; booking and accounting export through to a verified invoice run.
~€2,400 · ~3 days
A clearly defined fix that cleans up the cause instead of papering over it — with tests, through the same checkpoints as everything else.
~€210 · same day
Example amounts in the standard setup (× 2.2), plus VAT for companies. An MVP built from an idea gets a fixed price after joint scoping, from €3,000 (Your own app).
Your project is in my hands: I am a senior developer and managing director of BeyondWega UG, I steer the factory and I am responsible for the result. The factory is our internal tool; you do not operate it yourself.
The code lives in your repository and builds in your CI, without anything from us. Any other developer can carry on, and your IT can verify that themselves. For the MVP this applies from the handover of the repository and the rights.
You describe your codebase or your idea, and we clear up the open questions.
Before the start we agree on scope, definition of acceptance and pricing basis together.
We produce in our factory and deliver a pull request or a running MVP, each with an evidence pack.
You check the result against the evidence pack and take it over by merge or product acceptance.
Write to us about what you have in mind: a change to your codebase or an idea for a first product. Andreas Korol replies personally and tells you openly whether and how we can do it.
Fang!
HomepageLotse
faktor400
gantt400
looplocks