A wave of AI restructuring is sweeping the market, with organisations announcing layoffs in an attempt to pivot towards greater and more strategic AI adoption. Many, including the likes of Meta, are justifying these workforce changes as a necessary cost of building an AI-native future, and in some cases, ensuring the organisation’s future survival. However, this future is missing something vitally important.
The engineers who held a deep understanding of how the underlying systems worked, including knowledge of how they were built, why certain decisions were made, and where the fragile points were hidden, are on their way out. In fact, many of these staff left their organisations before the AI race even began, and they left for a reason that had nothing to do with artificial intelligence.
It’s the stack, not the technology…
There is some important scene-setting to consider here. For years, enterprises of all sizes have been asking engineers to spend the majority of their time maintaining systems those engineers did not want to be maintaining. Engineers by nature are intuitive and technically curious with a need to continuously iterate and push the boundaries, yet this is often not where they find themselves. New industry research found that more than half of security engineers and platform professionals spend at least 30% of their working week on systems maintenance rather than building. In primarily legacy environments, that figure climbs to a staggering 85%.
What this is unveiling is the crux of a retention problem disguised as a productivity problem, and for engineers, the frustration is rarely abstract. Being the custodian of legacy infrastructure, debugging systems built on decisions made a decade ago, and being constrained by tools that no longer reflect where the industry has moved to, is uninspiring. When alternative tools, platforms and systems exist that are demonstrably better suited to the work and motivations of modern engineers, the gap between what engineers are doing and what they know they could be doing becomes difficult to ignore.
So, when they spend their time on keeping outdated systems running rather than developing new capabilities, working on skills, or building something they find meaningful, it should be no surprise they stop seeing a future in their role. Naturally, they tend to then leave.
The research bears this out, and it’s an unfortunate picture. Four in ten technical professionals have either left a role, or seriously considered doing so, because the technology stack they worked with felt outdated, or it was actively holding back their professional growth. It’s not a niche grievance but a dynamic that has been playing out across organisations for a long time. Yet with the big push towards AI, it’s a problem now appearing quite significantly on the risk register.
Restructuring at the worst moment
What many organisations do not realise is that the consequence of this slow departure spans far beyond headcount loss. There is an erosion of something far harder to recover: the deep, accumulated knowledge of how critical tools and systems actually function in practice. Knowledge lives in people, including engineers who understand why particular integrations were built the way they were, and developers who understand which parts of digital infrastructure are genuinely fragile. They know exactly what to consider, what to ignore, and precisely when action is needed.
This knowledge walks out the door quietly, long before senior leadership announces any transformation programme. Most tellingly, 69% of respondents expect the talent gap between legacy and modern architecture expertise to grow over the next two to three years, with only 6% expecting it to shrink. The people closest to the problem already know where this is heading.
Against this backdrop, these same organisations are now racing to re-organise around AI. FAANG companies have been explicit about linking workforce restructuring to AI investment, framing reductions in headcount as a reallocation of resources toward automation and intelligent systems, and it is encouraging other companies to be similarly vocal. The logic of this action seems reasonable, yet there is a fundamental problem with the sequencing. Enterprises are accelerating into an AI-enabled future while simultaneously losing grip on the institutional knowledge they need to maintain the systems the AI is meant to enhance or replace – it’s the definition of running before they can walk.
A quarter of organisations surveyed are not confident they can attract and retain the engineering talent needed to maintain their current systems over the next five years. A further 50% say they are only somewhat confident. These figures sit alongside ambitious AI investment announcements as a deeply uncomfortable contradiction. You cannot transform what you cannot maintain, and you cannot maintain what you no longer have the people to understand.
The compounding dynamic here matters most. Legacy-heavy environments create the very conditions that drive talented engineers away – high maintenance burden, limited opportunity for growth, and tools that feel like a step backward relative to what the market offers. These environments then struggle to attract the talent needed to modernise, because the professionals who could lead that modernisation know what the stack looks like and factor it into their decisions.
Naming the risk clearly
The ability to retain the engineering knowledge required to keep critical infrastructure operational is not currently treated as a strategic risk in most organisations. It tends to surface in hiring conversations, in engineering team retrospectives, occasionally in security review, yet it rarely appears at the boardroom level. And, most concerningly, it almost never sits alongside cybersecurity and infrastructure resilience on enterprise risk registers. It’s a frightening thought, given that 48% of organisations already acknowledge some level of security exposure in their current infrastructure.
The fix is not purely a hiring or compensation question. Organisations need to evolve their systems to give engineers something worth staying for. That means investing in the tools, platforms and stacks that make meaningful work possible – moving away from where the majority of a developer’s week is consumed by maintenance, and towards architectures that enable them to build, iterate and grow. Modern composable tooling is increasingly a baseline expectation of the talent many organisations need most.
This is where the AI ambition and the retention problem converge. The enterprises best positioned to succeed with AI are not simply those investing the most in it, but those who have kept their engineers engaged, equipped them with modern stacks, and maintained the institutional knowledge required to build on top of what already exists. Every enterprise would do well to remember that AI adoption does not happen in a vacuum, but on top of infrastructure maintained by people who understand it. If those people have left, or are on their way out, no amount of AI investment will close that gap quickly.
Keep the engineers, keep the knowledge. Give them the tools they want to work with, and the AI future becomes something they can actually help build.
