Across five measured post-baseline slices.
A faster loop is not a better prompt. It is a better system.
We investigated how product ideas move from a request to working software, where time and confidence were being lost, and what the existing foundations made possible.
Separated from a 13h 20m calendar span.
Words across eight instructions.
Net lines, dominated by Python/Gemini removal.
The trajectory was sound. The implementation runway needed work.
The first step is to understand the challenge clearly. The later chapters introduce the changes and proposals.
The product direction and desire for quality were correct.
The system contained deep teacher knowledge, thoughtful specifications and extensive tests. The problem was not a lack of care.
Small changes carried the weight of large programmes.
Agents could encounter substantial reading, planning and setup before reaching the part of the product that needed to change.
Important behaviour was spread across very large files and several state layers.
Finding the right place to work—and knowing the change was complete—required broad technical knowledge.
Large test suites did not always create a clear route to release confidence.
Local checks, end-to-end journeys and release steps needed to operate as one understandable system.
Confidence grows as the change moves forward.
Each loop answers a different question. Failures are resolved in that loop before the change advances.
Build the experience
Fast local
- Hot reload gives immediate feedback
- Minimal vertical-slice checks cross frontend and backend
- Slower and exhaustive tests are deliberately deferred
Prove the complete change
Upstream push
- Relevant local test suite runs
- Failures return directly to implementation
- The proven change is pushed upstream
Prove it outside the local machine
Merge
- External CI reruns the combined test suites
- A clean environment catches skipped tests or environment drift
- Integration and visual journeys block merge on failure
Release the proven result
Deployment
- Known-good build deploys automatically
- Live smoke check confirms the release
- Rollback remains quick if needed
Fast first, thorough later. The local loop protects momentum; the upstream and merge loops add wider confidence; deployment promotes only a result that has already been proven.