TestMu ships AI agent that finds duplicate test cases

25 Sep 2026 (ET): TestMu AI (formerly LambdaTest) launched a Test Deduplication Agent inside Test Manager that reads titles, steps, and descriptions by meaning, scores matches 0–100, and refuses to delete anything without a human confirmation.

SOFTWARE desk — an AI that writes test cases can fill a suite with copies of the same check, and a coverage number then counts those copies as if they were different tests. A second agent that scores near-matches from 0 to 100 only helps if a person still has to confirm the delete, so the tool that made the mess is not also the tool that erases the one copy the team meant to keep. TestMu’s agent does the matching. The delete stays a human step, and the old test runs stay on the record.

What the wire says the product is. TestMu AI, called the Full Stack Agentic AI Quality Engineering platform, announced the Test Deduplication Agent in Test Manager. An AI agent, here, is software that does a job, not only a chat reply. That gloss is this desk’s. The agent finds duplicate and near-duplicate test cases across a project and hands teams a scored review queue to clear them. A duplicate is a second copy of the same check. A near-duplicate is a check that means the same thing in different words. A review queue is the list a person still has to read. Those glosses are this desk’s. The subhead says the new Test Manager capability reads test cases semantically, scores every match, and puts a human in front of every deletion. It ships alongside archiving, cross-project sharing, and bulk test run actions. Semantically, in that subhead, means by meaning. That gloss is this desk’s. Full Stack, and Agentic, are the company’s words.

The mess the wire says the agent is for. AI has made test cases fast and easy to produce. The result, for many teams, is a repository that has grown faster than anyone can maintain it. The same login flow gets written three times. A regression suite gets copied for a release and never merged back. A regression suite is the set of checks a team re-runs to see if a new change broke an old behavior. That gloss is this desk’s. The cost shows up later: slower runs, fixes that miss two of the three copies, and coverage numbers that claim more than the team tests. Coverage, here, is the count of how much of the product the suite claims to check. If three copies of one login test are counted as three tests, the number is larger than the work. That gloss is this desk’s. Past a few thousand cases, the wire says, nobody finds them manually. Three copies, and two of the three, are this paragraph’s. The quote below says “the other three.” Do not collapse those counts.

How the agent matches, in the wire’s words. The Test Deduplication Agent reads test cases the way a reviewer would. It analyses each case’s title, steps, and description and matches on meaning rather than wording, so cases that describe the same behaviour in different words are caught, not just exact copies. Every match gets a score from 0 to 100, grouped into bands, so teams clear near-certain duplicates first and work down to cases that overlap in intent but diverge in wording. A band is a bucket of scores, so the obvious copies can be cleared before the fuzzy ones. That gloss is this desk’s. The wire does not name the bands. The docs do. That list is in the docs paragraph below. This desk did not run a scan.

Nothing is deleted automatically. That sentence is the wire’s. Each duplicate is reviewed side by side against a base test case, with an AI-written summary of how the two relate. A base test case, here, is the copy the team keeps. The others are the ones under review. That gloss is this desk’s. Reviewers can ignore a match, promote a better-written duplicate to become the base, or mark it for deletion. A running tally shows what has been marked, broken down by confidence band, and the final deletion requires explicit confirmation. Test runs and their results are retained. A test run is one time the team actually executed the check, and the result is what happened. Those glosses are this desk’s. Explicit confirmation is the wire’s phrase. It does not print the word DELETE. The docs and the card do. Do not move that word onto the wire.

Mudit Singh, as a quote, not as a suite this desk timed. Singh, Co-Founder and Head of Growth at TestMu AI, said: “AI has removed the friction from writing tests, and that has quietly created a new kind of debt inside test repositories.” He said: “Duplicates don't look like a problem until a suite takes twice as long to run and a fix lands in one copy but not the other three.” He said: “We built the Test Deduplication Agent to understand what a test case is actually checking, surface the overlap with a score, and then get out of the way so a person makes the final call.” Twice as long, and the other three, are his. They are not a stopwatch this desk held. The earlier paragraph’s three copies and this quote’s other three are both on the wire. Do not collapse them. Get out of the way, and a person makes the final call, are his limit on the agent.

What keeps new copies from being written, kept apart from the scan. The wire says the agent clears duplication that already exists. Keeping new duplicates out is handled at the point of creation by Smart Context, which checks the existing repository before AI generates any test case. The docs use a longer name, Smart Context: Memory Layer, and say that layer prevents duplicates from being created during AI generation, while the duplicate scan finds duplicates that already exist. Smart Context is the wire’s name. Smart Context: Memory Layer is the docs’ name. Do not collapse them into a second product. This desk did not watch a generation run.

Three other Test Manager jobs that shipped in the same announcement, still the wire’s. Archive test cases: retire a test case without destroying it. Archived cases drop out of lists, runs, reports, and duplicate scans but keep their ID, version history, and past results, and can be restored unchanged. Up to 500 can be archived in one action. Share test cases across projects: maintain a test case once in a source project and share it, with the same ID, to other projects instead of copying it. Every change reaches every receiving project. Each project runs it and keeps its own results. Bulk test run actions: move, copy, or delete up to 1,000 test runs in one action. Moved runs keep their ID, URL, and history. Copies start fresh for the next regression pass. Five hundred is cases archived. One thousand is runs moved, copied, or deleted. Those are different counts on different objects. Do not add them.

Where it is, and what a scan costs. The Test Deduplication Agent is available now in Test Manager. Each scan costs a flat five credits, charged only when it finds at least one duplicate group. Five credits is the wire’s price. The docs say the same price as a flat 5 credits, deducted only when the scan produces at least one duplicate group. A scan that finds no duplicates costs nothing. The cost does not change with the size of the scan. Scanning 100 test cases and scanning 5,000 both cost 5 credits, and finding one duplicate group costs the same as finding fifty. Those size examples are the docs’. Five, on the wire, and 5, in the docs, are the same claimed price. This desk did not see an invoice, and it does not print a dollar amount the pages do not print. The credits number in the top bar of the card is the screenshot’s chrome. It is not this price.

Who the about box says TestMu AI is, and who to call. TestMu AI, formerly LambdaTest, is a full-stack agentic AI quality engineering platform. The about box says it is engineered for scale and AI-native by design, and that it offers end-to-end AI agents to plan, author, execute, and analyze software quality across web, mobile, and enterprise applications on real devices, real browsers, and custom real-world environments. AI-native, and end-to-end, are the company’s words. This desk did not audit the device lab. The media contact is Nikhil Saxena, Corporate Communication Manager, TestMu AI, nikhils@testmuai.com, +91 98709 81968. The about box does not print a headcount, a price list, or a street address. The dateline names San Francisco and Noida. Do not turn the dateline into a street.

What the docs add about the review, and which clock is theirs. The page is “Find and Remove Duplicate Test Cases in Test Manager.” It says detection is by meaning, not by matching the same letters. Each test case becomes a numerical fingerprint of its title, steps, and description, and the scan looks for the fingerprints that sit closest, so two cases can score as near even when the sentences differ. That gloss is this desk’s. The docs’ own words for the fingerprint and the search are in the source note. Every result is a % Match between 0 and 100. The bands the docs name are Above 80%, 50-80%, 20-50%, and Below 20%. The base test case is the one the rest of a group is compared against, and the one that survives when its duplicates are deleted. A duplicate is a case that scored above the minimum match score the team set. That minimum is a number from 1 to 100. Scope is the whole project, or one folder and its sub-folders. A title-only case is still scanned, and the docs say its score is far less reliable. Deletion is a confirmed step: mark the cases, then type DELETE in all caps. The docs say deletion is permanent, there is no archive step in this flow, and removed cases cannot be restored. That limit is the delete path. It is not the Archive action on the wire, which keeps the case and can restore it. Do not collapse them. The docs also say the scan covers manual test cases, including ones it calls KaneAI-authored and stored in Test Manager. Automation scripts are not analyzed. This page does not define KaneAI. Do not invent a definition. One scan at a time. A scope needs at least two test cases. Duplicates are detected within one project. Changing the base starts that group’s review again and does not cost additional credits. schema.org dateModified on this page is 2026-09-09T20:28:09+05:30, which is 9 Sep 2026, 2:58:09 p.m. UTC and 10:58:09 a.m. Eastern. The HTML last-modified header was Fri, 25 Sep 2026 18:44:32 GMT, which is 2:44:32 p.m. Eastern and 6:44:32 p.m. UTC. Do not collapse 9 Sep and 25 Sep. This desk read the docs page. It did not click Delete.

What the card shows. The card is the docs screenshot of the Delete Duplicates confirmation, the step where a person has to confirm. A modal titled Delete Duplicates sits over Review & Resolve Duplicates. The picture’s question is about permanently deleting 6 duplicate test cases. It says the action cannot be reversed later, that associated test runs will continue to exist, and that the person is then sent back to the project’s test cases list. The field asks for DELETE in all caps. The button reads Delete Duplicates with a 6. Behind the modal is a sample queue: a whole-project scope, match scores in the 90s, and the same four bands the docs name, Above 80% down to Below 20%. The 6, the sample IDs, and the credits readout in the top bar are the picture’s. They are not a census and not the five-credit price. No desk date is painted on the card. The catalog chip is SOFTWARE.

Plain English for the rest of the card. TestMu, the company that used to be LambdaTest, shipped an agent inside Test Manager that looks for test cases that mean the same thing, even when the words differ. Each match gets a score from 0 to 100. A person can ignore it, keep a better-written copy as the one that survives, or mark the extra for deletion. Nothing is deleted until that person confirms. Old test runs stay. A scan is a flat five credits, and only if it actually finds a duplicate group. A separate switch, Smart Context, is supposed to check the library before the AI writes a new test. Archive, sharing one test into other projects under the same ID, and moving up to 1,000 test runs are the other upkeep tools in the same announcement. The card is the confirm screen, with a sample of 6 cases waiting on the word DELETE.

PRIMARY here: TestMu AI’s 25 Sep 2026 GlobeNewswire release 3369152, datelined San Francisco and Noida — Tier A PRIMARY, the company’s own wire, read on the Financial Post mirror and the Markets Insider reprint because the direct GlobeNewswire page timed out. STATUS PRIMARY. The desk label on the chip is CONFIRMED because that is the catalog word for a primary page we can stand on. The agent, the meaning match, the 0 to 100 score, the bands as a word, the human review, the ignore-promote-mark choices, the tally, the explicit confirmation, the retained test runs, the Mudit Singh quote, Smart Context, archive up to 500, share with the same ID, bulk actions up to 1,000, available now, and the flat five credits are that wire’s. The four named bands, the DELETE field, the flat 5 credits with the 100-versus-5,000 and one-versus-fifty examples, the permanent-delete limit, the manual-only limit, KaneAI as an undefined name on that page, one scan at a time, one project, and the two docs clocks are the docs’. The 09:30 AM stamp is Markets Insider’s. The 13:34:07 UTC stamp is the Financial Post schema. NOT claimed: a GlobeNewswire header this desk did not load, a dollar price, that 09:30 AM and 9:34 a.m. Eastern are the same clock, that the three copies and the quote’s other three are the same count, that this desk ran a scan or typed DELETE, that the sample 6 or the credits readout on the card is a customer figure, a definition of KaneAI the docs page does not print, a stock tip, or investment advice. The card is the docs confirmation screenshot. Distinct from the already-filed langsmith-engine-v2, jobber-mcp-chatgpt-claude, and gen-ai-data-checker. The catalog chip is SOFTWARE, not SIGNAL.