By Raghavendra K.A, SVP and Global Head of Engineering – Integrated Product Development, Engineering Systems & IoT, Infosys
Three-quarters of large manufacturers have embedded AI into their enterprise strategy, as per a recent Infosys Knowledge Institute survey of 650 senior executives at organisations with annual revenues above one billion dollars. This shows a mindset shift is underway for AI in engineering.
However, the returns are still not fully visible. Forrester predicts that fewer than one-third of decision-makers can tie AI’s value to their organisation’s financial growth.
Better returns typically go to organisations that apply AI across the whole product lifecycle, where engineers set goals and constraints, and models evaluate thousands of potential solutions simultaneously.
Decluttering bottlenecks in knowledge management
A significant share of product development delays trace back to inaccessible knowledge, often lying in the minds of engineers or scattered across siloed documents. Instances include reasoning behind a tolerance, rationale around a design decision, a material substitution, or a rejected product form factor, all typically discussed in meetings and recorded nowhere.
Capturing design context and institutional memory inline is crucial. A model that provides clean specifications and design rationale can describe what a product is and why it evolved in that way. Any change then proposed by it will be reliable, resulting in potentially faster product iterations.
Gartner predicts that by 2027, more than half of the generative AI models that enterprises use will be domain-specific rather than general-purpose, up from 1 percent in 2024. A model trained on real material properties with heat and strength constraints and regulatory requirements is capable of producing designs that work in practice.
When models capture context as work happens, rather than reconstructing it afterward, they can generate technical documentation directly from design files, code, and simulation runs, preserving reasoning with results. That documentation becomes a searchable record for the whole product lifecycle, with every decision traceable to its source.
Keeping ops smooth: Requirements engineering, simulations, and digital twins
Requirements analysis accelerates with model capability to read across interviews, field reports, and regulatory documents to reveal patterns and contradictions that manual reviews miss.
Digital twins, working replicas of physical assets, return predictions in minutes rather than days. Testing and simulation run alongside development rather than after it, surfacing issues early and reducing cost of product changes. They continue keeping loops running after the product sale.
Sensors in products already in use warn of failures before they happen, guide software updates, and show engineers which parts wore out first and why. That informs the next designs on what to build into products that can be repaired instead of discarded.
NASSCOM projects India’s engineering research and development services revenue to grow from an estimated 63 billion dollars in FY26 to more than 100 billion dollars by 2030. Its study of Physical AI describes demand arriving in three stages: intelligent products and connected operations, simulation and virtual engineering as businesses move into production, and finally, autonomous systems over the longer term. The order of the sequence matters because the second stage depends on the data discipline established in the first.
Accelerating the product lifecycle shift
To enable the transition, the focus must be on key implementation pillars. First, models must be trained on clean, physics-grounded engineering data to minimize hallucinations, with governance for that domain.
Next, data from products in use continuously loops back to engineering where validation continues. Finally, models and digital twins are constantly re-tuned as operational conditions change.
Human judgment continues to decide which constraints hold, while exercising creativity in product design and evolution loops alongside questions of safety, ethics, and strategic intent.
Transition, ROI, and staying power
The destination is to self-optimize products and learn continuously from the data they generate in use, while people stay in charge of oversight, strategy and the trade-offs involving safety. Organisations that pair this judgment with product architectures built to learn, will see returns across better product quality, less capital locked in products that don’t ship, and knowledge that stays and evolves with them.
An engineering base, deep in embedded systems, simulation, and industrial domains, is positioned to own that knowledge rather than execute pieces of it.