Featured in reworked: Generative AI Blurs Build vs. Buy Lines

While accepted wisdom is that generative AI will make software building easier, there's nothing to suggest that this might be the case. Here's why.

Giving Senior Developers Room to Grow

The use of generated AI has already affected a growing number of decisions related to how to implement software, but as of the end of 2024, it has still landed inconsistently, author and Confirmed CEO David Radin told Reworked.

Professional developers, he said, can use GenAI to help with their testing, bug identification and fixes as well as shorten the time it takes to develop software projects. The result is reduced costs for building software in-house versus purchasing off-the-shelf software or having outsiders develop custom solutions.



It also allows these professional developers to delve into areas beyond their normal expertise, which means companies can develop software without the broader set of skills previously necessary.

While in theory this means companies can get more done with less experienced programmers, what's lost is the often necessary context that only experienced developers who have encountered a project type or situation multiple times before can provide, Radin added. Ultimately, it allows companies to have more novice to intermediate programmers for each senior developer, which lowers the total cost of in-house development while still maintaining quality.

“This type of reconfiguration of staff, though, will require them to change their processes and the roles of the senior members will take on a higher percentage of mentoring instead of actual coding,” Radin said. “In these situations, the novices can strategize with their senior colleagues, then co-code with GenAI, and then do code reviews with the senior colleagues."

Speeding Traditional Development Cycles

Generative AI democratizes software development within organizations, said Stax's Peter Rodrigues-Renon. Employees across departments can now directly shape and improve their digital workspaces, and respond to operational needs more quickly than traditional development cycles allow, he continued.

The promise hearkens back to the early days of low-code programming.

Rodrigues-Renon argues the shift transforms workplace dynamics in several ways:

1. Distributed Innovation

Innovation becomes more distributed, as employees closest to business challenges can prototype solutions without extensive technical expertise. Teams can iterate faster, leading to more responsive and efficient operations.

2. Organizational Frameworks

However, this democratization requires new organizational frameworks. Leaders must balance employee development with maintaining security and standards. They need to establish clear guidelines for internal tool development while ensuring data protection and system integrity.

3. IT Enablers

IT departments' responsibilities will evolve as a result, shifting from gatekeepers to enablers. They must build platforms that allow safe experimentation while maintaining enterprise architecture. Meanwhile, software vendors need to adapt their products for greater customization and integration, moving from closed systems to more flexible, component-based solutions.

The result is a more fluid digital workplace where the line between users and developers blurs. “Fundamentally, giving the power to 'create' into the hands of more people democratizes the ability to push the boundaries of what a business is capable of and reshapes the frontier of a business's value creation capability,” Rodrigues-Renon said.

Build vs. Buy for GenAI Solutions Themselves

However, not everyone is convinced that generative AI will tip the scales in one direction or the other. David Menninger, executive director of software research at ISG, believes any enterprise predisposed to buying software will continue to do so. "Those that are inclined to build custom software will keep doing that as well," he said. 



“GenAI may give individual enterprises a reason to switch from build to buy or vice versa,” Menninger said. “Realistically, every organization buys software — has anyone built their own email system or word processing system recently? The real question is when they should consider building their own applications.”

He pointed out that when an organization has proprietary processes or intellectual property that can’t easily be captured in pre-built applications, they have no choice but to build their own.

The same applies to building or buying GenAI solutions. ISG's recent research found that 21% of enterprise GenAI solutions are predominantly off-the-shelf, 34% are mainly custom-built and 45% are a mix of both.

While many GenAI tools allow for fine-tuning and retrieval-augmented generation (RAG), and some now offer small language models trained on proprietary data, if these techniques fall short, enterprises may need to consider developing their own applications.

The challenge, Menninger added, is quantifying the trade-offs. GenAI can reduce the software development lifecycle by 20%-50% across various phases. For instance, manual efforts in coding, documentation and test design can be cut by as much as 50%.

However, these savings need to be weighed against the benefits of building a custom application is difficult to quantify. Moreover, he points to the difficulties in estimating the cost of maintaining a custom application in a fast-changing market.

The More Things Change, The More Things ...

Hiswai founder Vaclav Vincalek argues the generative AI boom will not change much in the context of buy vs. build, since everyone will have access to the same tools.

At best it will provide an opportunity (or the illusion of opportunity) for less technical people, that they too can start building software, he said. In theory, you might shorten the cycle when you are writing the code, but the rest remains the same: debugging, QA, deployment, version control, documentation.

“The one thing which will happen for sure is that there will be the slow disappearance of knowledge about these systems and when something goes wrong (and it definitely will) there will be nobody who will be able to troubleshoot the problem,” Vincalek said.

When building a technology solution and using third-party technology, organizations should always strive to buy 'functionality' rather than 'application,' he continued

When you integrate functionality into your own solution, he added, it should be relatively easy to replace it with the same functionality from another vendor. This prevents vendor lock-in and could allow organizations to build that functionality in-house in the future.

Software in the Future

When considering the future trajectory of AI development in software engineering, Vincalek predicts the initial wave of automation will target easily replicable tasks, with tools like Microsoft Copilot and AWS Q Developer serving as early examples. He urged businesses to adopt a measured approach, stating that the real insights will emerge from two or three years of practical feedback, which will illuminate both areas for improvement and new opportunities in the field.



For his part, Radin envisions a strategic evolution of GenAI applications. He anticipates that beyond immediate coding assistance, AI will increasingly contribute to higher-level activities such as strategic planning and competitive research. He believes this dual approach — combining strategic insight with practical coding support — will help ensure that development efforts are both well directed and efficiently executed.

Together, these perspectives suggest a gradual but significant transformation in software development, where AI tools evolve from handling basic automation to supporting more complex decision-making processes. The change, however, will be slow.