TestMu AI Blogs

Tanvi Mittal on why generating tests got cheap, and what a closed loop running from a production signal back into the test suite actually has to do.

Pooja Oza of Disney on replacing the exact-match oracle with a bounded behaviour envelope, why fluency is not correctness, and the false green in self-healing.

Meenakshi Singh on why AI sales pilots stall - 85% launch, under 18% scale - and the start, test, evolve, adopt sequence that moves them into practice.

Andrew Duncan on why the AI trust gap is widening, what evidence keeps a system deployed rather than merely deployed, and the wine agent that explains drift.

Neelmani Verma of Infosys on why a code-generated test is a mirror not an oracle, and five pillars that make trust a credential with an expiry date.

Three BFSI quality leaders on where agents already act alone, why autonomy is earned through evidence, and the audit trail a regulator will ask for.

Chinmay Kothari on retiring pass and fail for a trust score, the four-step RMTE loop, and why one passing trust attribute is still not a passing system.

Harinee Muralinath of Thoughtworks on why AI made the forward-deployed engineer affordable, the judgement gap it left behind, and how to close it.

Gaurav Khurana of Microsoft on why assertEquals proves nothing for an AI agent, and what to assert instead: reasoning, tools, memory, orchestration, outcomes.

Zebra put QA at the front of product discovery - the nurse documentation problem, the pivot to dictation, and the internal AI tools behind it.

Agentic QA tools compared: 15 agentic AI testing tools ranked on autonomy, test ownership and execution breadth, plus what enterprise teams should check.

Agentic test management tools that write cases from requirements, refine what exists and keep traceability current. 10 platforms compared on what their AI owns.

Nilesh Dalvi of Glean on why 5x faster coding cut his cycle by only 16 percent, and what turns an AI assistant into a coworker that advances the work.

AI powered testing tools that generate cases, heal them and judge what broke. 15 platforms compared on the AI capability each brings and where the tests end up.

Harleen Bedi, CTO of Quality Engineering at Infosys, on why coverage stopped being the benchmark and who is accountable when AI writes and passes the tests.

Lisha Rakesh of Capgemini on the assisted, augmented and autonomous modes of AI in QE, why autonomous is not unsupervised, and the 80/20 shift to strategy.

Rushabh Mehta of Meta on agent evals: idempotency keys, checkpointing, memory TTLs, the three grader types, and why GAIA 2 shows temporal awareness still fails.

Magnus Hillestad of Sanity on why agents are not digital employees, what Myriad cost at $10,000 a day, and the missing compiler for organising intelligence.

Jaydeep Chakrabarty of Piramal Finance on why retrieval alone caps your AI stack, and how a knowledge graph can derive context nobody ever wrote down.

Viktoria Semaan of Databricks on running 16 models through one eval set, why Gemma 12B matched Sonnet at a fraction of the cost, and when fine-tuning pays.