The first challenge wasn’t building the simulators. It was discovering exactly what they needed to simulate. Much of the expected behavior of the physical hardware, including error conditions and protocol interactions, wasn’t fully documented. Critical knowledge lived across existing documentation, source code, device protocols, and the experience of a small number of specialists.
Gorilla Logic started by running Chronicle, its structured and repeatable workflow for reconstructing how a system behaves. Chronicle first mapped the available information, then analyzed the system in greater depth, synthesized the findings, and finally validated them with the people who understood the system best. When expected behavior wasn’t clearly documented, the team captured it.
That process transformed knowledge that had previously lived in people’s heads into documented and validated requirements for all three simulators: the dispensing hardware simulator, web application simulator, and Canister Station Simulator.
Once the team knew what the hardware needed to do, the next challenge was turning that knowledge into working software efficiently without giving up engineering control.
Instead of treating Claude as a coding assistant working task by task, Gorilla Logic built an agentic SDLC around the requirements produced by Chronicle. The system operated more like a small development team than a single AI assistant.
A Planner agent handled intake and scoping, proposed the technical approach, and transformed the validated requirements into structured work items small enough to deliver in roughly a day. Each work item was tied to acceptance criteria before implementation began.
From there, a Generator agent created the development workflow needed to execute the plan. An orchestrator routed the work to developer agents that followed strict test-driven development practices. When repeated failures occurred, agents were required to stop and escalate rather than continue generating changes indefinitely. A reviewer operated in a separate context and independently re-ran the checks instead of simply accepting the developer agent’s results.
That agentic workflow accelerated implementation, but AI never controlled the final engineering decision.
Developers committed their work locally, and the orchestrator pushed changes only at the end of a phase and only after Gorilla Logic reviewed and approved them. Nothing merged into the main branch without a person signing off. AI-generated code was therefore treated the same way the team would treat code produced by any other contributor: it had to satisfy the requirements, pass its checks, survive independent review, and clear a human gate before becoming permanent.
That combination of Chronicle, agentic development, and human oversight created a repeatable path from undocumented system behavior to validated software.
The approach also helped the team make pragmatic implementation decisions along the way. For the dispensing hardware simulator, Gorilla Logic selected .NET 8 instead of .NET 6, enabling the lightweight background server that allows automated tests to drive the simulator over the network rather than through its desktop interface.
For the web application simulator, the team avoided an unnecessary rewrite. Instead, Gorilla Logic extended the existing simulator with a new CLI and REST layer, preserving previously validated behavior while making the simulator accessible to automated workflows.
The same approach produced supporting capabilities around the simulators. A Simulator Configuration Tool replaced manual, error-prone configuration with a desktop workflow supporting snapshots and rollback, while the Release Tag Diff Analyzer introduced another agentic workflow that compares release tags or pull requests and drafts corresponding test proposals.
The result wasn’t simply three simulators. Gorilla Logic created a repeatable way to capture system knowledge, translate it into software requirements, accelerate implementation with AI, and still keep engineers responsible for what ultimately reaches production.
