Blog
Notes on proving software works
False greens, the agent dev loop, and what it takes to trust a green check.

Latest
mabl alternatives, and where mabl is the right platform
mabl is an enterprise platform with governance and mobile coverage. Six mabl alternatives compared on verdict source, plus cases where mabl is the better buy.
Comparison11 min read
Divyanshu Shekhar ·

Comparison11 min read
Functionize alternatives, and when an autonomous platform is the right buy
Functionize sells autonomous QA at enterprise scale. Six Functionize alternatives compared on how each decides pass or fail, including where Functionize wins.
Divyanshu Shekhar ·

Comparison11 min read
Selenium alternatives, and when Selenium is still the right answer
Selenium still wins on breadth and cost. Six Selenium alternatives compared on what each one verifies, and the cases where replacing it would be a mistake.
Divyanshu Shekhar ·

Comparison11 min read
Momentic alternatives, and where Momentic is the better choice
Momentic is strong at test generation and mobile coverage. Six alternatives compared on what each uses to decide pass or fail, including where Momentic wins.
Divyanshu Shekhar ·

Comparison12 min read
BrowserStack alternatives, compared on what they actually verify
BrowserStack is a device cloud, not a verification layer. Six alternatives compared on what each verifies, plus the case for keeping it and adding a check.
Divyanshu Shekhar ·

Guide9 min read
Verification vs automation: the distinction that decides your CI bill
Automation answers whether the steps ran. Verification answers whether the thing worked. The difference, and what each one actually costs you per commit.
Divyanshu Shekhar ·

Definition11 min read
What is a verification layer, and what does it have to prove?
A verification layer decides whether a change worked by reading the running system, not the test output. The definition, four criteria, and how to test a claim.
Divyanshu Shekhar ·

Tutorial10 min read
Build a verification harness in 20 minutes
From an empty repo to a recorded flow, a verdict, and a failing run you can hand back to your coding agent. A 20-minute tutorial with real terminal output.
Divyanshu Shekhar ·

Comparison12 min read
Test automation frameworks, compared on what they can assert
Jest, Vitest, pytest, Playwright and Cypress compared on the axis that decides year two: what each can assert on, and what that costs you to maintain.
Divyanshu Shekhar ·

Guide14 min read
MCP testing: the tools nobody tests
MCP gave agents tools. Testing them means asserting on the tool-call trace, not the chat transcript. The contract, the failure modes, and a minimal harness.
Divyanshu Shekhar ·

Comparison14 min read
The best AI testing tools, compared on what they actually verify
Twelve AI testing tools compared on the one column nobody publishes: what each uses to decide pass or fail. Plus which architecture fits agent-written code.
Divyanshu Shekhar ·

Guide15 min read
Automated testing: a field guide for agent-written code
Automated testing decides only what it was told to decide. A field guide to the four layers, the nine test types, the judgement gap, and the real cost curve.
Divyanshu Shekhar ·

Comparison12 min read
Regression testing tools: what each one actually verifies
Nine regression testing tools compared on one axis: what they use to decide pass or fail. Plus the maintenance tax nobody quotes and a two-week pilot plan.
Divyanshu Shekhar ·

Guide12 min read
AI QA tools: buy, build, or neither
An AI QA tool is only as good as what it uses to decide pass or fail. Buy versus build economics, an 11-question RFP, and a two-week pilot design.
Divyanshu Shekhar ·

Guide10 min read
Self-healing tests fix the red. They do not tell you what broke.
Self-healing repairs selectors, waits and sometimes asserts. What it fixes, what it quietly hides, and the three cases where healing is the right call.
Divyanshu Shekhar ·

Comparison11 min read
Playwright alternatives: what each one is actually for
Six Playwright alternatives compared on what they actually verify, including the case where the right answer is to keep Playwright and add a verification layer.
Divyanshu Shekhar ·

Guide11 min read
AI test case generation, without the slop
Most AI test case generators produce plausible cases, not risky ones. How to generate cases from recorded flows and failure history instead of from a prompt.
Divyanshu Shekhar ·

Guide11 min read
What smoke testing is, and what should run on every commit
Smoke testing is a fast check that the build is worth testing further. What belongs in a commit-time smoke suite when an agent is writing the commits.
Divyanshu Shekhar ·

Guide12 min read
Exploratory testing, rebuilt for agents
Exploratory testing is the discipline where the tester learns while testing. What changes when an agent does the exploring, and what has to stay human.
Divyanshu Shekhar ·

Guide15 min read
AI test automation that does not rot
Most AI test automation writes tests faster and leaves judgement to the model. That is why suites rot. Here are the four layers, and the one nobody automates.
Divyanshu Shekhar ·

Guide12 min read
AI QA: what it is, what it is not
AI QA is used for three different things: generating tests, judging results, and replacing QA. Only one of them holds up. Here is how to tell them apart.
Divyanshu Shekhar ·

Guide10 min read
Agentic testing: two meanings, one that survives
Agentic testing describes both agents that write tests and tests that check agents. Defining the second one, and why the distinction decides what you build.
Divyanshu Shekhar ·

Guide12 min read
What is a software test harness?
A test harness is the scaffold that runs a test and reports the result. Here are the five parts, a working example, and why most harnesses verify nothing.
Divyanshu Shekhar ·

Guide12 min read
Flaky tests: root causes, not retries
A flaky test is asserting on something that is not the program's truth. Here are six root causes, how to detect them, and why retries only make it worse.
Divyanshu Shekhar ·

Guide11 min read
Testing AI-generated code: seven failure modes
AI-generated code fails in seven repeatable ways that unit tests miss and end-to-end tests catch late. The list, the harness, and what still escapes.
Divyanshu Shekhar ·

Guide13 min read
Verification vs validation, and why AI broke the distinction
Verification asks if you built it right; validation asks if you built the right thing. AI-authored code breaks that split: here's how to tell them apart.
Divyanshu Shekhar ·

Guide15 min read
AI Verification: the step your agent skips
AI verification checks what your agent actually shipped, not what it claimed. How to read network calls, app state and console to get a verdict that holds.
Divyanshu Shekhar ·

Guide10 min read
The Jev playbook: where it wins, where it lags, and the number behind each call
Eleven workloads, and whether TypeSafe AI's Jev should take them, with the evidence behind each call: our own harness runs and published measurements we could link.
Divyanshu Shekhar ·

Guide14 min read
Which decisions in your testing agent should be Jev calls
A practical guide to wiring TypeSafe AI's Jev into a testing agent: Choice, Score and Noul question design, option sets, batching, and what must stay in code.
Divyanshu Shekhar ·

Engineering12 min read
Six things TypeSafe AI's Jev cannot do, and the one that will cost you a release
TypeSafe AI's Jev cannot abstain, explain itself or read your field names, and holds 32K. A field guide to the System One model's limits and which ones hurt.
Divyanshu Shekhar ·

Benchmarks12 min read
Jev works. The question is whether the giants let it live.
TypeSafe AI's Jev drove our verification harness for $0.006 and found a 100x under-refund that a green Playwright suite walked past. It is a real breakthrough. Netscape was a real breakthrough too.
Divyanshu Shekhar ·

Benchmarks11 min read
Jev vs LLM-as-judge: what $160 per million graded answers actually buys
An LLM-as-judge costs $33,000 per million graded answers; Jev matched one 91.5% of the time for $160. What that changes about evals, and what it can't.
Divyanshu Shekhar ·

Story10 min read
Reticle started because I was too lazy to stay awake
Half a Claude Code quota, one night of sleep, and a morning that looked like a car crash. How a batch of wasted tokens turned into a product people queued up to install.
Divyanshu Shekhar ·

Guide10 min read
How to use Claude Code and Codex for testing
Claude Code and Codex for testing, set up so it holds: what to hand them, what to never let them decide, and how to stop believing a summary that says everything passes.
The Reticle team ·

Guide8 min read
How to use Claude Code for QA automation
Using Claude Code for QA automation that holds: what it can verify on its own, where it quietly fails, and how to give it a check it cannot fake.
The Reticle team ·

Engineering5 min read
Why a green check isn't proof
A green check tells you the page rendered, not that the app works. The gap between the two is where false greens live, and what a passing test cannot see.
The Reticle team ·

Benchmarks19 min read
I gave my coding agent eyes three different ways. Here's the honest scorecard.
Playwright MCP, Chrome DevTools MCP, and Reticle all let an AI coding agent see a running web app. I built Reticle, so I ran a committed benchmark across all three and wrote down where each one wins and where mine loses.
Divyanshu Shekhar ·

Workflow5 min read
The bottleneck moved, and nobody updated the org chart
Agents write code in minutes, then a human spends days checking it. The verification bottleneck is the slow step now, and a faster agent only makes it worse.
The Reticle team ·