Long-form essays from the engineers shipping AI inside payers, hospitals, energy operators and proptech platforms. Written for technology leaders who care more about what runs in production than what trended last week.
In an AI system the model, the prompt, and the data can all change independently, and any one can cause a regression. How to catch it with one quality baseline and change attribution.
AI applications do not give the same output twice, so pass-or-fail testing breaks. How to test non-deterministic systems with evaluations, baselines, and quality distributions.
Canary releases turn every deploy into a small, measured experiment with automatic rollback. How to use them as a quality instrument, with the right metrics, gates, and abort criteria.
Testing across every device, OS, and screen size is impossible to do exhaustively. How to automate mobile testing against a risk-based device matrix that covers real users without the madness.
Automated accessibility checks catch a fraction of real barriers. How to combine automated and manual testing into accessibility coverage that actually passes an audit.
Performance testing done once before launch catches problems too late. How to make it continuous, with baselines, budgets, and regression detection that catch slowdowns as they happen.
Test data is the quiet bottleneck behind slow, flaky, non-compliant testing. How to make test data realistic, compliant, and available on demand, with a clear owner.
Fault injection testing deliberately breaks things to see how the system responds, so you rehearse failure before customers experience it. How to run it safely with hypotheses and blast-radius…
End-to-end tests are valuable and easy to overdo. Why fewer, deeper, more stable E2E tests beat a sprawling flaky suite, and how to build the ones that earn their cost.
API tests are the fastest, most stable quality signal you can build: below the flaky UI, above the isolated unit. How to automate schema, contract, functional, and boundary checks.
Exploratory testing is where human curiosity and judgment find the bugs scripts and agents miss. Why it stays human, how to run it with charters, and where AI helps around it.
Visual regression testing catches UI breakage that functional tests miss. From brittle pixel diffs to perceptual AI checks, how to catch real visual bugs without drowning in noise.
Flaky tests have specific, diagnosable root causes: timing, shared state, environment, and nondeterminism. A field guide to finding the real cause and the fixes that stick.
Test maintenance is the hidden cost that quietly eats your QA budget. Where it comes from, why it stays invisible, and how agentic tools and better design cut it.
Most QA metrics measure activity, not quality. How to pick metrics that predict escaped defects, like defect escape rate and coverage of risk, instead of test counts and cases run.
TestOps runs quality like an operations discipline: test infrastructure, data, environments, and flakiness managed as a system, so tests stay fast, reliable, and trusted at scale.
Shift-right testing uses real production signals to catch what pre-release testing cannot, without gambling in production. How to test in prod safely with observability and guardrails.
Shift-left testing moves quality earlier, but only works if the gates help developers instead of blocking them. How to design fast, trusted gates engineers do not resent.
The test pyramid still holds, but AI changed the economics of each layer. What generation, agents, and cheap tests mean for how you balance unit, integration, and end-to-end.
Risk-based testing spends your limited quality budget where failure costs most. How to weigh impact and likelihood so testing effort follows risk instead of habit.
Mutation testing checks whether your tests actually catch bugs by introducing deliberate faults and seeing if tests fail. How it exposes weak tests coverage never could.
Self-healing tests repair broken selectors when the UI changes, killing the maintenance that makes UI tests hateful, as long as healing does not quietly mask real bugs.
AI can generate tests fast, but coverage is not the same as protection. How to tell real coverage from confident noise by judging assertion quality, not test count.
A maturity model for autonomous testing, from scripted automation to self-healing to AI-generated to fully agentic, with the guardrails each level needs in a production team.
One long-form essay every other Wednesday. Written by the engineers shipping production AI for our clients, not by a content team. No promotional emails. Unsubscribe in one click.
