Skip to content

How we test

How we test the AI software we build, before you rely on it.

For a growth-stage hardware or deep-tech team, the question is whether AI-built software holds up on your real systems. We agree on acceptance criteria with you before we build, then test every build against them.

What you can check

  • Acceptance criteria, agreed before the build
  • Tests at both ends of each connector, set in the acceptance criteria
  • You set the tools, execution limits and human review points
  • Your team can inspect the task, workflow, files, checks and execution history of a Zerg agent run together.

Before the build

We agree what passing means first.

The acceptance criteria say which tests, on which systems, and what counts as passing. You see them before work starts.

  • For example: an invoice that doesn’t match its purchase order waits for review, and doesn’t post to the ERP until someone on your team approves it.

During the build

What we test, and against what.

  • Workflows, on your real systems

    Our engineers build workflows, integrations and custom apps, then test them against your real systems.

    Each build is tested on the systems or hardware agreed in its scope, and against its acceptance criteria.

  • Connectors, at both ends

    A connector can break quietly when either system changes.

    A first connector build includes tests at both ends, and alerts for the failure conditions the agreed checks cover.

  • Kernels and low-level code, against benchmarks

    Kernels, ports or low-level code for the hardware you named, accepted against agreed benchmarks.

    Verified against tests we agree with you and run on your target platforms.

  • Simulations and digital twins, against the plan

    A digital twin starts from a scope we set with you: the operation, the plan to test on it, and what counts as a faithful model.

    Before the build, we agree which observations or operating data will test the model, and the limits of what it can predict.

After launch

Support after launch.

We monitor results, fix failures and adapt the workflows as your rules and requirements change.

Free consultation

Start with one process. We’ll show you how we would test it.

In a free consultation we recommend a first build, and how we would test it before you rely on it.