4SO Lab · How we build products
20 years of hands-on experience × AI × engineering discipline
4SO Lab is how 4SO builds products: human experience defines the problem, automation and AI speed up analysis and implementation, and independent testing decides what can actually be relied on.

Practical experience defines the problem, its limits and the real effect of a decision.
AI comes after the problem, not before it.
The base path is repeatable and testable. AI enters the loop when it adds real value to analysis, proposing solutions or speeding up implementation, and its output is still checked independently.
- 01UnderstandReal system behaviour, operator needs, constraints and failure modes.
- 02DesignSource of state, state transitions, user experience and safety boundaries before execution.
- 03BuildAgents and AI speed up implementation, inside the product contract.
- 04ValidateRepeatable tests, negative scenarios and reading back the real state.
- 05LearnFailures and evidence feed back into the design instead of hiding behind a green status.
Every claim is checked at the level where it matters.
Correct code, a successful install, behaviour under failure and real-world operation are four different questions. For each one we use evidence from that level.
- 01
Behaviour and contract
Contracts, API, state model and expected behaviour are checked at code level.
- 02
Installed environment
The product package is built, installed in an independent environment, and the install result is checked again.
- 03
Failure scenarios
Service loss, retries, interruptions and recovery paths are tested under near-real conditions.
- 04
Real operation
Where behaviour depends on real infrastructure, the same product package is run and observed in a real environment.
AI is an engineering tool, not the decision authority.
How AI is used differs between 4SO products, but one principle holds: AI must not create an independent source of state, a hidden approval or a change path outside the Product API.
- 01
Minimal, necessary context
The model gets only the information the problem needs, not extra or unrelated data.
- 02
A healthy path without the model
When a testable path works without AI, the model should not become a permanent dependency.
- 03
No security shortcuts
AI does not replace authorization, human approval or the formal change path with direct access to a shell, SQL or the runtime.
- 04
AI output is tested again
A proposed change is not valid until it passes tests and a read-back of the real state.
We reduce technical complexity; we don't remove human judgement and control.
The operator should understand what changes, what effect it has, what blocks it and where the result was confirmed. AI is good when it makes that path clearer and faster.
- 01Humans decide
- 02Policy owned by the product
- 03Deterministic validation
- 04Evidence from the runtime
See the result of this method in the products.
Every product page has a technical reference section: architecture, workflow, failure behaviour and operations.