When Coding Agents Started Calling the Shots

New Delhi has become an unlikely venue for corporate software drama. According to a McKinsey report, AI-powered coding agents are beginning to flip the traditional IT budget script. Instead of buying polished vendor software, many companies are asking a blunt question: Why rent a machine when the machine can now write itself? The sales teams are scrambling.
Nearly one-third of respondents said their organisations had declined to purchase at least one software product or feature because agentic coding tools could build the functionality internally. In other words, the SaaS aisle is suddenly feeling less like a buffet and more like a suggestion box. The trend is strongest in technology and healthcare, with professional services, energy and materials not far behind.
This shift is happening as agentic AI scales fast. Forty percent of respondents at organisations with more than $1 billion in annual revenue said they were scaling AI agents in at least one function, up from 27 percent a year earlier. Adoption among smaller organisations, however, remained broadly unchanged at 22 percent. AI high performers are moving even faster. They are twice as likely as other organisations to report scaling software coding agents and 2.7 times more likely to scale other agentic AI. Nearly half of high performers said they had forgone software purchases in favour of building internally, compared with 31 percent of other respondents.
One anecdote from the enterprise world feels neat. Imagine a healthcare IT team facing a licensing quote so thick it needed its own binder. Rather than sign, they let a coding agent draft a narrow reporting dashboard. It was imperfect, certainly, but it solved the immediate problem, avoided a sprawling contract, and made procurement sigh with relief. That is the kind of moment vendors dread: not because customers hate software, but because they now hate paying for software they can coax into existence.
I have a more modest personal experience of this new economy. Once, I tried to automate my inbox with an AI helper. It seemed magical until it spent so many tokens on summarising one confused meeting request that I felt like paying a butler to argue with a doorbell. The lesson is simple: AI can save money in one room and set it on fire in the compute closet.
That is why the economics matter. Around one in five organisations said AI-related operating costs, including token costs, had constrained their use of the technology. Yet investment appetite remains strong, with 60 percent of respondents expecting their organisations to increase AI investment over the next year. The trend could increasingly force technology companies to rethink the traditional software buy model, particularly as coding agents become more capable.
The report suggests leading organisations are becoming deliberate about where to buy, build and develop capabilities. As agentic systems handle complex tasks, firms must balance savings from replacing purchased software against compute, tokens and development costs. Coding agents are not merely typing faster. They make price tags and the chance to say no more interesting.