Back to NeuralWorks
Layer 1 · Specification & Assurance
Tests as Spec.
The flight plan for agent delivery.
Test-driven development has been the right idea for twenty years and too expensive for most teams to sustain. Spec-driven development is the open problem in AI engineering right now. Tests as Spec sits between them.
If an agent’s failure mode is misunderstanding requirements, then the requirement — not the code — is the artefact that must be hardened. So every requirement ships with a test plan, and a person reviews that plan before a line of code exists.
Legend
Agent step
Developer step
★ Critical review point
Step 1
developer
New requirement arrives
Step 2
agent
Drafts test plan
Step 3
★developer
Reviews and approves test plan
Step 4
agent
Implements the tests
Step 5
agent
Implements logic; prepares test report
Step 6
agent
Runs unit / integration / E2E suites before raising PR
Step 7
agent
Fixes regression failures; compiles regression report
Step 8
★developer
Reviews test results and regression report; signs off
The load-bearing moment
Step three. A developer reads the AI-generated test plan and checks it against what was actually asked for, including what was implied and not written down. It is the one step we will not automate.
What it costs you
Existing features need unit, integration and E2E coverage before agents can safely change them. Business analysts have to validate the E2E tests, including edge cases. This is real work and it takes months on a large estate.
What you get
Developers stop writing code and start owning requirements. One engineer runs several features and fixes through agents at once, in the way a product owner runs a backlog. The regression net holds while they do it.
Start here
Start with one thing that works.
A 60-minute working session with two of our engineers. You bring a delivery problem. You leave with a written assessment of where AI would help, where it would not, and what it would take to find out.
No pitch deck. No obligation.