Artificial intelligence (AI) is changing what cars can do, such as recognising a pedestrian or responding to a spoken command. It is also changing how cars are engineered, long before a vehicle reaches the road. That second shift is less visible to a consumer, but it is a critical change that shapes how quickly new vehicles reach the market.
Gartner predicts that by 2028, 90% of enterprise software engineers will use AI code assistants, up from less than 14% in early 2024. For carmakers, BCG estimates that generative AI could cut the time needed to bring a new vehicle to market by up to 40%. But a car cannot reach customers until its software has been tested and proven safe. That is where the real value of AI in automotive engineering will be decided.
The real bottleneck is proving code safe
Every requirement in a vehicle programme has to be analysed, built, tested and documented. As more vehicle functions move into software, the number of requirements and the links between them keep growing.
Automotive software also faces stricter safety rules than most enterprise software. Engineers must show that each requirement has been built and tested correctly, and that a change in one area has not caused a fault in another. The challenge is therefore less about writing code quickly and more about proving that it works as intended.
This is the work AI is starting to take on. It can draft test cases straight from requirements and flag the ones that have not yet been tested. When a requirement changes, AI can show which tests and documents are affected, so engineers do not have to trace every link by hand. It can also help them work through defects faster. BCG estimates that generative AI can reduce effort in automotive software and feature development by roughly 25% to 30%. It sees efficiency gains of close to 30% as increasingly realistic in integration and testing.
From projects to platforms
Many engineering teams began using AI one project at a time. The next step is to build it into shared platforms that serve several vehicle programmes at once.
The value of such a platform lies in the engineering knowledge it holds, including approved test cases, past requirements, defect histories and proven verification steps. When AI can draw on this knowledge, a new programme can start from work that has already been validated.
AI agents, which can plan and carry out multi-step tasks on their own, are starting to show what this looks like. McKinsey describes a leading automotive supplier that trained a team of AI agents on its database of past requirements and test descriptions. The agents now write first drafts of test descriptions for new requirements, and some requirement types take 50% less time. The gains were strongest among less experienced engineers. The system worked best when a similar case existed in its past data. For entirely new scenarios, the company added more human review.
The same idea applies to day-to-day verification work. On over-the-air software update programmes, some teams now pair AI tools that read system logs with AI agents trained on technical documentation. Feasibility checks that engineers once worked through manually can then be completed much faster, leaving more time for harder problems.
From implementation to orchestration
AI will not remove the need for engineers, least of all in safety-critical development where a person must still review and sign off the work. Their role is moving from writing code towards directing AI tools and verifying the output. More of their time will go into system design and the decisions that need engineering judgement. In automotive software development, that means AI prepares the initial code, test cases and documentation, and engineers decide what is fit for use.
Human oversight matters even more as the adoption of AI increases. Gartner predicts that by 2027, 40% of enterprises will scale back or switch off autonomous AI agents because of gaps in governance that surface only after something goes wrong in live use. It recommends matching the level of control to how much independence each agent has. It also warns that human approval works only while it stays meaningful. Under time pressure, approvals can become a formality that gives a false sense of safety. In automotive engineering, safety sign-off must remain a real review by named engineers, whatever the agent’s level of autonomy.
The skills engineers need will change as well. They will need to know where AI adds value and, in safety-critical software, where it should not be relied on at all. India has a particular stake in this shift, since its engineering teams support vehicle programmes around the world. As AI takes on more routine work, safety expertise and sound judgement of AI output will set these teams apart.
Re-Engineering engineering
Vehicle engineering no longer ends when production starts. Updates and new features now arrive over-the-air throughout a vehicle’s life. Each one has to be tested and proven safe before release, and that workload grows faster than any engineering team can grow by hiring alone. That calls for a change in how engineering itself is done.
Over the next few years, the edge will move from what a vehicle offers at launch to how safely it keeps improving. Features can be copied. An engineering system that uses AI to test and release updates quickly and safely is far harder to replicate, and it takes years to build. The companies that start building it now will set the pace for the rest of this decade.
Authored by Ganesh S Rao, VP, Technologies at CEO Office, Embitel Technologies
(The views expressed are solely those of the author(s) and do not necessarily reflect those of ET Edge CIO&Leader, its management, or members. The author(s) are solely responsible for the originality, accuracy, and authenticity of the content, including proper attribution of third-party material and compliance with applicable copyright and intellectual property laws.)