Companies considering whether to build their own AI software or buy established platforms should weigh long-term costs, security and corporate knowledge before cutting subscriptions, a software industry executive has warned.
The former chief information officer said the growing availability of tools such as Claude Code had encouraged businesses to create custom applications in-house. But he argued that the ability to build a system did not necessarily mean it was the right commercial decision.
He recalled a project earlier in his career, when a beverage company asked his technology team to develop an analytics dashboard. Despite having the necessary staff and resources, the project suffered repeated setbacks over four months before the team abandoned it and bought an external solution instead.
The decision freed the team to concentrate on other priorities, rather than continuing to devote time to a system that was not central to the business.
Build or buy: where AI development adds value
The first question for technology leaders, he said, should be whether building software supports the company’s core expertise. Even highly capable engineering teams may create greater value by focusing on innovation and their main business rather than developing internal systems for functions such as billing or human resources.
Potential savings must also be examined beyond the subscription fee. The cost of developing, operating, maintaining and securing an in-house replacement can outweigh the apparent reduction in annual software spending.
The executive said companies developing their own large language model-based software had typically spent five to 10 times more than they would have done using his company’s workflow automation platform, depending on the complexity of their operations.
He also described a financial services company where a chief information security officer proposed using a large language model to rewrite a core system. The work could have taken between 18 and 24 months while saving the equivalent of only 0.5% of the company’s annual operating budget, he said.
Security and governance present another significant test. A system that works when it is launched must continue to withstand changing cyber threats, remain auditable and comply with evolving regulations.
The executive said roughly half of organisations had seen AI agents exceed their permissions. He cited PocketOS, a car rental management platform, where a coding agent wiped a production database in nine seconds.
Technology leaders were urged to assess whether they could monitor permissions, maintain security and provide adequate governance throughout the life of a custom platform, rather than only during its initial development.
Continuity after a senior technology leader’s departure is another consideration. The executive said the average tenure of a CIO was about four and a half years, while a bespoke platform could take two years or more to build and longer to mature.
That creates a risk that the person responsible for key architectural decisions and integrations may leave before the system is operating effectively. One CIO told him of a previous employer where a custom platform became indispensable, but was left without documentation, a support team or an external supplier after its creator departed.
When building custom AI software makes sense
Building in-house can still be justified when the project creates a genuine competitive advantage by using proprietary data or institutional knowledge.
The executive pointed to a digital and information technology leader at a major shipping and logistics company who was using AI selectively to optimise routes, drawing on decades of internal data and operational expertise.
By contrast, established AI-powered platforms may be more suitable for routine, clearly defined processes such as employee onboarding and holiday requests. These applications can offer clearer safeguards and a more direct route to financial returns, he said.
AI has made custom development easier, but it has not removed the need for businesses to decide carefully where to build and where to buy, the executive argued. For CIOs, the choice can affect both the organisation’s prospects and the way their own technology teams use their limited resources.
