Your automated tests are green and you trust them, but they only ever check what you thought to check. A confusing label, a broken layout on a phone, a flow that technically works but feels wrong: a machine sails past all of it. This chapter gets you the human pass that catches what no assertion can.
8.8.1Automated tests miss the human bugs
A test only ever checks a case you thought of in advance. You wrote it, or your agent did, to confirm the thing you already pictured. It cannot flag the case that never crossed your mind.
It also never asks the question a person asks without trying: does this feel right? A form can submit, save, and pass every check while its button sits in a place no one would look. That whole class of bug, ugly, confusing, awkward, working but wrong, is human-only.
8.8.2Explore, don't just follow a script
Exploratory testing is poking at your app like a curious or careless user with no script in hand. You deliberately leave the path you designed and see what breaks.
The moves that find bugs are the ones a tidy user never makes:
- click things in the wrong order
- submit the empty form, then the same one twice
- hit the back button in the middle of a flow
- resize the window small, then tiny
- paste an essay into a field built for a name
Spend ten unhurried minutes doing the weird thing. That is where the bugs no test imagined are hiding.
8.8.3Hand the boring passes to an agent
Your agent can drive a real browser too, which means the repetitive half of this does not have to be yours. Agentic QA is writing the case in plain English and letting the agent walk it. It checks what actually landed in the database, then reports back or fixes what it finds:
That covers the passes you would skip on a Friday. It does not replace your own ten minutes, because an agent only checks what you thought to ask, and noticing that something feels wrong is still yours.
8.8.4Keep a release checklist
Before every release, every time you ship a new version to users, walk one short list by hand on a real device. Not the desktop browser shrunk narrow, an actual phone, because that is where layout and touch bugs live. The same list every time turns into muscle memory:
Keep it to what would embarrass you if it shipped broken. A list you actually finish beats a long one you skip.
Those last three are not polish. Safari behaves differently from Chrome often enough that it is the classic thing a human pass catches. And a flow you cannot finish with a keyboard is a flow some of your customers cannot use at all.
8.8.5Write a bug report your agent can fix
When the human pass finds something, how you write it up decides whether the agent fixes it in one round or three. A vague "the signup is broken" forces the agent to guess and re-ask. A reproducible report, the exact steps plus what should happen and what did, gets it fixed on the first try.
Fill in the same four parts every time:
Hand your app to a senior QA engineer and let them build both artifacts around your specifics:
Do this now: paste the prompt, then walk your app's release checklist by hand on a real phone before your next ship.
