How Tech Mahindra is applying AWS's AI-DLC methodology across its developer teams

As enterprise AI moves beyond chatbots toward autonomous, multi-step systems, large software firms are under growing pressure to rethink not just their tools but the entire structure of how software is built. For Tech Mahindra, one of India's largest IT services companies, that rethink has centred on a methodology called AI-DLC, and a partnership with AWS that has put it into practice across developer teams at scale.

From Augmentation to Autonomy

The story of how AI entered enterprise software development, according to Adrian De Luca, Director of Developer Experience at AWS, is a story of two distinct phases. The first, beginning around 2023, was augmentation: using AI for narrow, well-defined coding tasks such as auto-completion, documentation, and test generation. Developers gained speed on specific activities, but the broader delivery process did not fundamentally change.

"What used to be very slow in Agile became very fast," De Luca said. "The sprints that used to take two weeks, these can now be done in hours. And so what was fast, which were the rituals, have now become the slowest part."

Those early productivity gains, in his estimate, around 10 to 20 percent, were real but insufficient given what the underlying technology had become capable of. The bottleneck had simply shifted.

What changed that calculation, De Luca explained, was the arrival of reasoning capabilities in AI models. Unlike generation-focused tools, reasoning-capable systems can interpret intent, ask clarifying questions, decompose requirements into components, and move across multiple phases of a development process without constant human direction. This is what has given rise to agentic AI (autonomous systems capable of multi-step, independent decision-making) and, in AWS's framing, to a concept called Frontier Agents: long-running autonomous agents that operate in the background, enforcing norms around security, compliance, and quality as development progresses.

The Scale Problem

For Subhash Yadav, Global Technology Learning Head at Tech Mahindra, the challenge was less about the technology itself and more about deploying it consistently across large, distributed developer teams. The risk, as he described it, was that AI would continue to be treated with suspicion, "hiding behind the word hallucination," in his phrasing, rather than integrated into actual delivery workflows.

"We always keep on hiding behind the word hallucination and just not trusting what basically AI is giving to us," Yadav said.

The goal Tech Mahindra set itself was to move from probabilistic AI use, where developers accepted results with a degree of uncertainty, toward something more deterministic: a system where each step in the software development lifecycle could trigger the next only once defined conditions were met. That shift, Yadav argued, was what made it safe to allow agents to operate with greater autonomy.

"What is happening is it is getting more and more safe for us to make these agents work a little bit more autonomously than basically the case earlier," he said.

The Methodology

The framework that structured both companies' work together is the AI-DLC (AI-Driven Development Lifecycle), an open-source methodology that AWS published as a white paper in August 2025. The methodology is not a rigid prescription. De Luca was explicit that, like Agile before it, AI-DLC is designed to be adapted to each organisation's own rituals, norms, and dependencies.

What it does provide is a way to align tools, roles, and ceremonies around AI-driven flow, rather than the linear stage-by-stage progression of traditional Agile. In a flow model, as De Luca described it, AI systems interact continuously with humans, starting from a business intent, extrapolating requirements, breaking those into units of work, and carrying the process through to deployment, with human approval loops at key decision points rather than at every step.

"It takes that business intent and extrapolates the requirements, those requirements into units of work," De Luca said, describing a process that encompasses product managers and QA testers, not only software engineers.

What Changed

Yadav shared the outcomes with some caution, noting that specific figures are still being validated internally, but he indicated that the overall improvements have been substantial. Across use cases, Tech Mahindra has seen between 50 and 60 percent reductions in timelines and productivity improvements. A manufacturing sector case study he referenced saw estimation work that previously took months completed within a week.

The more immediate commercial effect has been on customer engagement. Yadav described a shift in how Tech Mahindra now responds to customer requests for proposals: rather than lengthy scoping cycles, the team can build a working prototype in approximately a week, validate requirements with the customer in real time, and refine the brief before significant engineering effort has been committed.

"Something which used to take from ideation to basically prototype, those timelines used to be in months earlier," Yadav said. "Today, we basically can go and talk to customers, understanding what they want by quickly creating a prototype, and that doesn't require more than a week for us."

The productivity gains, he added, extend across testing and DevOps roles, not only development, with all three functions seeing material impact.

Lowering the Access Cost

Both speakers pointed to a broader ambition beyond the immediate partnership. De Luca said AWS has recently released self-paced learning courses built around the AI-DLC methodology, available to individual developers, students, and professionals, alongside structured workshops. The aim, as he described it, is to democratise access to a methodology that was developed through large enterprise engagements, making it available to smaller teams at lower cost.

"AWS's philosophy of working backwards from a real, meaningful problem and then democratising that access, lowering the access cost, the consumption cost of those things, means that those things are going to spread a lot faster," De Luca said.

Yadav, for his part, described the value of the partnership not only in tools but in cadence: AWS experts have been present at each stage of Tech Mahindra's product development process, and solutions validated in one vertical have proved adaptable to others.

Both spokespersons avoided making long-term predictions, noting how quickly the technology is evolving. What De Luca did say was that the next phase of learning would come from continuing to work in the most complex environments, with the largest teams, and feeding those lessons back into both methodology and product. The reasoning, as he put it, is straightforward: "If you can solve for the big teams, the small teams are very simple."