Field notes on testing, AI and agents
How enterprise teams keep tests readable, runs predictable and AI accountable.
The runtime AI tax
Runtime-AI test tools call a model on every step of every run. How that cost compounds, what it hides, and where AI pays for itself.
Plain English needs a grammar
Natural-language tests drift into ungoverned prose without a schema. Twelve verbs, quoted targets and a parser keep them readable and exact.
Quality has a tool-sprawl problem
Enterprises buy functional, performance, security, test data and other quality tools separately, each with its own model of the same app. The case for one.
The desktop gap in enterprise testing
Packaged apps and custom Windows clients run critical work, yet desktop tooling is thinning out. What the 2027 deadlines mean and how to plan.
Coding agents need a test gate
Coding agents can now run, edit and heal tests through MCP. Why the test suite needs a policy gate, and why an agent must never approve its own heal.
Earned autonomy
How much should AI be allowed to change your tests? A six-level autonomy ladder, why trust has to be earned per scope, and the rules that never relax.