IT teams are losing top engineers to better-paying roles in product and AI startups at an accelerating rate. Leaders who have cracked the retention problem share what actually works.
Key Takeaways
The numbers have been building quietly for two years. A joint survey of 400 IT leaders and 800 engineers, conducted by McKinsey's Technology Institute in the first quarter of 2026, found that voluntary attrition among IT engineers has risen 34% since 2023. The increase is not evenly distributed across the workforce: it is heavily concentrated among senior engineers with five or more years of experience, the professionals who carry institutional knowledge, mentor junior colleagues, and own the most complex systems in the environment. These are not the engineers who leave because they receive an unsolicited recruiter message. They leave because something in their current organisation has failed them, and they have decided that something better is available elsewhere.
The destinations are well documented. AI startups are the single largest category of destination employer in the survey, capturing 38% of departing senior engineers. Product-led technology companies come second at 24%. Legacy enterprise IT, including financial services, healthcare, and government technology, is the primary source of departures in each case. The pattern is not simply about compensation, although that is a factor. It reflects a fundamental difference in what these organisations offer in terms of the work itself: the technical problems, the autonomy to solve them, the visibility of impact, and the clarity of career progression. Enterprise IT organisations that have not addressed these structural issues are running a slow-motion talent drain that compounds quarter by quarter.
The financial cost of that drain is often underestimated. When the survey asked IT leaders to calculate the fully loaded cost of replacing a senior engineer, including external recruitment fees, interviewing time, onboarding investment, and the productivity ramp-up period before the new hire reaches full effectiveness, the median estimate was 1.8 times the departing engineer's annual salary. For a senior engineer earning $180,000, that implies a replacement cost of approximately $324,000. Across an IT organisation losing ten senior engineers a year, the cumulative cost runs to more than $3 million annually, before accounting for the knowledge loss, the team disruption, and the delivery delays that senior attrition reliably produces.
The survey asked departing engineers to rank their reasons for leaving across eight categories. The results challenge the conventional wisdom that retention is primarily a compensation problem. Career progression limitations ranked first, cited by 61% of respondents as a primary or significant factor in their decision to leave. Lack of autonomy over technical decisions came second at 54%. Compensation ranked third at 47%, followed by technology relevance at 43%, meaning engineers who felt their current organisation was behind on AI adoption and modern tooling. Only then do factors like management quality, remote work policy, and company culture appear in the rankings.
The AI factor within that technology relevance category deserves particular attention. Engineers who described their organisation as behind on AI adoption were 2.7 times more likely to be actively seeking other roles compared to engineers who felt their organisation was at or ahead of the curve. The mechanism is straightforward: engineers who are not working with AI-augmented development tools, building AI-integrated systems, or contributing to AI product development are watching their market value erode in real time. They understand the direction of the industry. If their current employer is not investing in the tools and projects that keep their skills current, the implied career cost of staying is very high, and eventually that cost exceeds whatever loyalty or inertia is keeping them in place.
Career progression limitations operate differently depending on how an organisation's career architecture is structured. In organisations where the only recognised path to seniority runs through management, technically excellent engineers who have no interest in people management face a ceiling that arrives at a relatively early point in their careers. Many of the most capable engineers in enterprise IT are in exactly this position: highly compensated relative to their tenure, deeply knowledgeable about the systems they own, and with nowhere obvious to go inside the organisation that does not require them to stop doing the technical work they are good at.
The survey identified a cohort of IT organisations with voluntary senior engineer attrition rates below 8%, well below the survey median of 19%. When the research examined what these organisations have in common, three practices emerged with consistency.
The first is a structured technical career ladder that runs parallel to the management track and extends to senior levels with genuine recognition, compensation, and influence attached to each rung. In the leading organisations, a Distinguished Engineer or Principal Engineer designation is not a honorary title. It carries budget authority over technology decisions, a seat in architecture governance, and compensation that competes with the director and VP management equivalents. Engineers in these roles do not need to become managers to progress; they need to deepen their craft and expand their technical influence. Organisations that have built this structure credibly report that their best engineers actively choose to stay on the technical track rather than accepting management roles they do not want.
The second practice is genuine autonomy over tooling decisions. The word genuine matters here: many IT organisations describe themselves as giving engineers autonomy over tooling while actually operating a tightly constrained approved vendor list and a procurement process that makes it impractical to adopt anything outside of it. In the leading organisations, tooling autonomy means that engineering teams can propose, evaluate, and adopt new tools through a lightweight governance process that takes weeks rather than months, and that the default answer to a well-reasoned tooling proposal is yes rather than no. This autonomy is particularly consequential for AI development tools. Engineers who can choose their own AI-assisted coding environments, experiment with new model integrations, and deploy those tools without navigating a multi-quarter procurement cycle feel meaningfully in control of their own productivity.
The third practice is access to meaningful AI projects. This does not simply mean having an AI team that other engineers can observe from a distance. It means that AI project work is distributed across the IT organisation rather than siloed in a specialist unit, that engineers are given protected time to work on AI capability development as part of their core role, and that contributions to AI projects are visible in performance reviews and promotion decisions. Organisations that have structured AI work this way report that their engineers describe it as one of the primary reasons they chose to stay when they were considering leaving. The signal to engineers is unambiguous: this organisation is invested in building the skills you need to remain competitive, and it is making space for you to build them.
"We stopped treating retention as an HR problem and started treating it as a leadership problem. The engineers who left were not leaving for more money. They were leaving for more meaningful work and a clearer path forward. Once we fixed that, the attrition numbers moved."
Claire Nakamura, CIO, Meridian Group
The temptation to respond to rising attrition with compensation adjustments is understandable, because compensation is the most visible and immediate lever available. But the survey data is unambiguous: organisations that have raised compensation without addressing career progression and tooling autonomy have not solved the retention problem. They have delayed it. Engineers who accept a counter-offer without any structural change to the conditions that prompted them to look elsewhere typically resume their search within six to twelve months. The organisations with genuinely low attrition rates are not the ones paying the highest salaries. They are the ones that have built an environment where staying is the rational choice on dimensions that matter more to engineers than compensation alone.
CIOs who want to diagnose their own organisation's retention posture honestly should start by asking four questions. Does the IT organisation have a credible technical career ladder that extends to senior principal or fellow-level roles with real authority and competitive compensation? Do engineering teams have genuine autonomy over tooling decisions, or does the procurement process effectively override that autonomy in practice? Are AI projects distributed across the organisation, or are they concentrated in a team that most engineers cannot access? And finally: when a valued engineer resigned in the past twelve months, what did the exit interview actually reveal, and what changed as a result? The answers to those four questions will tell a CIO more about their retention risk than any engagement survey or benchmarking exercise can provide.
A survey of 600 IT leaders maps the choices separating transformation leaders from laggards in an AI-driven economy.
New research maps the capabilities separating CIOs driving enterprise transformation from those still managing tickets.
