Give a coding agent a request for discount codes and it may return something that looks finished. The checkout works. A code reduces the price. Everyone can move on. Rukmangada Kandyala would like someone to interrupt the applause. Can two discounts be combined? What happens after a code expires? Could the total slip below zero? The machine has been answering product questions while appearing to do programming.
This is the example Kandyala used in September 2026 to explain a change in software development. A human developer could take a vague ticket to a colleague and ask for clarification. An agent can fill the gaps without a conversation. Its confidence makes the missing decisions harder to spot. A working checkout is a surprisingly convincing disguise for an unfinished specification.
Kandyala, the Bengaluru-based co-founder and CEO of Testsigma, has built a business around giving more people access to automated testing. His recent writing follows that ambition into less comfortable territory. Once software and tests can both be produced quickly, the difficult work moves toward deciding what they were supposed to establish in the first place. The green light still needs an explanation.
02An engineer before the pitch
His route to that question passed through years of engineering jobs. Kandyala studied mechanical engineering at NBKR Institute of Science & Technology, affiliated with S.V. University, from 2003 to 2007. His professional history then moves into software: Zoho from 2007 to 2011, Avenda Systems in 2011, and Aruba Networks later that year. He became a technical lead at Aruba, a Hewlett Packard Enterprise company, in 2016.
The sequence matters because Testsigma grew around a problem its founding team had encountered inside product organizations. Automated testing could turn into another software project, with its own frameworks, scripts, expertise and maintenance. The machinery assembled to save time began requiring a considerable amount of it. Anyone who has bought a labor-saving appliance and acquired a new weekend hobby will recognize the arrangement.
Kandyala co-founded Testsigma with Pratheep Velicherla, Vikram Chaitanya and Rajesh Reddy in 2019. The early proposition was unusually readable for an engineering product: let people express automated tests in plain English. Manual testers and colleagues with product knowledge could participate without first becoming specialists in a particular scripting language. Technical fluency would no longer determine whose knowledge could enter the testing process.
03Opening the workshop door
By March 2022, Kandyala was discussing another consequential choice: making Testsigma open source. In a TestGuild conversation with Joe Colantonio, he explained the decision and the needs of development and quality assurance teams. He also announced the open-source framework on DEV Community. His public shorthand there was KR, with a profile description devoted to building Testsigma.
Open source gave the early project a public workshop as well as a sales counter. The accompanying argument concerned flexibility: teams needed room to extend a system, accommodate unusual applications and work together. A tool that removed scripting requirements still had to leave space for technical work when a particular situation demanded it. Accessibility and extensibility had to fit in the same product.
The company announced a $4.6 million financing round in 2022, led by Accel and STRIVE, with BoldCap participating. Entrepreneurs and operators also joined, including Freshworks co-founder Shanmugam Krishnasamy and Ally.io CEO Vetri Vellore. Those connections placed Kandyala among people who had already spent years building software businesses. The stated use of the capital was to strengthen engineering and product work and support a global testing community.
04Capital follows the English lesson
In June 2024, Testsigma announced another $8.2 million, led by MassMutual Ventures with Accel, STRIVE and BoldCap participating. The announcement paired financing with new generative AI capabilities and a product for the Salesforce ecosystem. The natural-language approach now covered automated tests for web applications, mobile applications and APIs. Kandyala described an ambition to build an operating system for quality engineering teams.
The phrase suggests how far the scope had moved beyond composing a first test. A quality team has to organize work, understand changes, execute checks and explain failures. Each handoff can lose information. Putting more of that activity together was a larger undertaking than supplying an easier editor. For Kandyala, the expanding product offered a way to pursue the same participation problem across more of a team's working day.
The business also extended through partnerships. In April 2025, Umpisa announced a relationship with Testsigma to bring its testing tools to businesses in the Philippines and Southeast Asia. Kandyala pointed to Umpisa's local presence and technical expertise. The development supplied a practical picture of global reach: software distributed across borders still benefits from people who understand the market on the other side.
05A coworker called Atto
On May 7, 2025, Testsigma introduced autonomous testing capabilities and a new test management product. The central character in the announcement was Atto, described as an AI coworker for quality assurance teams. Its proposed responsibilities stretched across planning, creating tests, executing them and reporting what happened. Kandyala's launch essay connected the change to conversations with quality leaders, automation engineers and manual testers.
Their complaints included scattered tools, repeated setup work and testing processes struggling to follow faster software development. The product's intended answer included agents that could respond to development activity, such as a new sprint or a pull request. The test management component brought manual testing into the same broader effort. These were product promises about reducing coordination work, as well as generating more test steps.
Atto 2.0 followed in November 2025. Testsigma emphasized adaptation to product changes, attention to real user behavior and the discovery of untested scenarios. The direction is revealing: an initial emphasis on making automation accessible had widened into a concern with keeping tests relevant as applications changed. A test written yesterday can be perfectly executable and thoroughly out of date.
06Ninety people who had never raised a pull request
A glimpse of Kandyala's approach inside the company comes from marketing leader Sathyanarain Muralidharan. He described proposing a hackathon to get marketers building with Claude Code. Kandyala expanded the idea across the go-to-market organization. The resulting event reached further still, bringing together participants from sales, customer success, support, human resources, finance and other functions.
Muralidharan reported 35 teams, 90 people who had never raised a pull request, and 29 deployed applications. His account included 150 deployments in eight hours. These are the organizer's figures for one internal event, rather than a productivity benchmark. Even with that distinction, the anecdote makes Kandyala's enthusiasm for wider participation concrete. The people usually receiving a demonstration were being invited to build something themselves.
That experiment sits neatly beside the original Testsigma proposition. Colleagues can understand a problem well before they acquire the vocabulary of its implementation. Letting them participate changes who gets to contribute. It also creates a new obligation: when more people can produce working software, the organization needs ways to examine what that software assumes. An open workshop still needs someone checking the joints.
07The missing stretch before release
In October 2025, Kandyala began The Feedback Loop, a biweekly newsletter about AI, automation and systems thinking. Its opening concern was the time between producing code and learning whether it could be trusted. Faster creation could coexist with slow feedback. Waiting for tests, approvals and useful results still consumed the day. His writing gave that less photogenic part of software development sustained attention.
By December, he was examining what happened when software creation outpaced understanding. His year-end essay argued that execution results could arrive quickly while losing the context that made them useful. A binary outcome is easy to display. Understanding what a change means for a user takes more work. He framed contextual feedback as the lesson to carry into the following year.
In June 2026, he narrowed the problem to the interval between generated tests and a release. Someone must connect those tests to the sprint's requirements, find omissions, fill the gaps and interpret results. He wanted the output to reach developers where they already worked and to acknowledge uncertainty. His July essay made the human responsibility explicit: the developer pushes the button, and the engineering lead owns the outcome.
“The developer still pushes the button.”
Rukmangada Kandyala · July 2026
08Before the machine starts guessing
His September checkout example brings that argument to the beginning of the process. He advises teams to specify intended behavior, forbidden outcomes, known edge cases and runnable checks. Another habit is especially economical: ask the agent to list its assumptions before writing code. Reading those assumptions gives people a chance to make the product decisions they have accidentally delegated.
There is a place for testers here that precedes the finished application. Their practice of imagining exceptions can improve the initial request itself. The awkward question about an expired discount belongs in the specification while it is still cheap to answer. Kandyala's advice makes quality assurance part of defining the work, with a useful role for the person inclined to interrupt a smooth demonstration.
His path through engineering, plain-English automation and AI coworkers keeps returning to who can contribute and what a result means. The latest tools can make participation easier and output plentiful. His recent argument asks the people using them to preserve intent, expose assumptions and keep responsibility visible. Somewhere between a checkout that runs and a checkout a team is willing to release, the green light has to earn its keep.