Engineering leaders are at an AI-driven turning point that challenges some long-held assumptions about how teams create value.
Engineering 2028: Leading Human + AI Teams Responsibly, produced in partnership with Damilah, draws on insights from senior tech leaders to map a reality more nuanced and ambitious than headlines suggest, including an interesting economic paradox sitting at the heart of the AI productivity debate.
The (simplistic) conversation around AI and engineering headcount tends to go one of two ways.
Either the technology is supposed to hollow out your team, or it’s supposed to make everyone ten times more productive. Both of those options are very dramatic, and neither quite captures what’s actually happening at the moment.
In reality, when tasks that used to take weeks are compressed into hours, engineering leaders aren’t looking for the exit. Instead of reducing headcount, organisations are discovering that increased efficiency simply reveals a massive, previously suppressed demand for more features, more innovation, and more scale.
This is the Jevons Paradox applied to software development.
What is the Jevons Paradox?
First observed by economist William Stanley Jevons in 1865, the paradox originally identified that making coal consumption more efficient didn’t reduce the total amount of coal burned. Efficiency lowered the cost of use, which expanded demand, which then led to increased consumption overall.
The same dynamic is now playing out across engineering organisations.
AI tooling reduces the cost of building things, and when building costs less, organisations tend to ask for more. Feature backlogs grow, new markets become viable to explore, and technical debt that was previously uneconomical to address suddenly enters the conversation. The work expands to meet the newly available capacity, often exceeding it. Many engineering leaders who understand this paradox plan for a time of growth, not of contraction.
However, this dynamic may shift as AI pricing continues to mature. The cost of these tools is low right now, but historical SaaS pricing patterns suggest that won’t hold indefinitely, even if the appetite already unlocked is unlikely to contract with it.
The ’10x Developer’ Myth
Much of the Silicon Valley framing around AI productivity focuses on the individual: what can a single developer now accomplish? It’s a compelling idea, but misses the more significant question, which is what happens to the organisation when that individual becomes more capable.
The data from our report reflects this. When asked what productivity leap AI would realistically be expected to unlock by 2028, 68% of our respondents were clustered in the 11-50% range. This would be very meaningful progress, but a long way from the dominant 10x narrative.
Organisations don’t simply maintain their current ambitions and reduce their teams accordingly. They instead reassess what is now possible and adjust their ambition, and leaders planning for a leaner, more efficient team may find themselves surprised by the volume of demand that better tooling unlocks.
“We’re not interested in doing the work of 400 people with 200 and reduce heads. We’re interested in doing the work of 800 people with 400.”
– Lee Provoost, CTO, Flagstone
Efficiency Gains as a Growth Accelerator
The firms best positioned for 2028 aren’t those treating AI as a way to do the same things more cheaply. They’re treating it as a way to take on projects and responsibilities that were previously impossible.
How you measure and frame productivity shapes what conclusions you may draw from it. When measured against a fixed workload, efficiency gains look like a cost reduction, but measured against an expanding set of opportunities, they look like a growth driver. The second framing is closer to what most engineering organisations are actually experiencing.
“Efficiency gains are unlocking exponential demand rather than shrinking teams.”
– Duncan Lawie, VP of Technology, Kline Company
When code generation gets cheaper, human judgment becomes the scarce resource, and the ability to look at what’s been built, assess whether it’s right, and then direct the next decision becomes the premium skill. Without a person there who knows how to properly steer AI tools, the expanded capacity they can offer is meaningless.
What This Means for Planning in 2028
It’s worth being honest about what’s consuming the efficiency gains many teams are generating.
For some organisations, AI tooling has genuinely freed up capacity for higher-value work, while for others the time saved has been absorbed almost immediately by new feature requests, accumulated technical debt, and the overhead of managing the AI tools themselves.
Neither outcome is inevitable, and the difference tends to come down to whether leaders have been deliberate about where reclaimed capacity flows, or whether it’s simply been pulled in by whichever direction the demand is loudest at a given moment.
Senior engineering leaders who have internalised the Jevons dynamic tend to plan with that question front of mind. Rather than asking ‘how do we maintain output with a smaller team?’, they ask ‘what does expanded capacity allow us to take on that we couldn’t before?’. That reframing has practical consequences, as hiring decisions change, conversations with stakeholders change, and the way you structure roadmaps and manage expectations across the business changes too. It also shifts how you evaluate AI tooling, since the relevant question stops being ‘does this cut cost?’ and becomes ‘does this create value?’
The organisations set to be best positioned at the end of this decade are those building the capacity to absorb expanding demand now, not those trimming to match current expectations. The work isn’t shrinking, nor is it likely to start shrinking, and the question is whether you’re building a team that can grow into what the business will eventually ask of it.
Our full Engineering 2028 Leading Human + Al Teams Responsibly report, created in partnership with Damilah, goes deeper into this and much more, including how senior leaders are thinking about headcount, AI adoption maturity, and the governance challenges that come with scaling human and AI teams together. You can download it here
