Estimated reading time: 2 minutes
Key takeaways:
- AI spend jumped 28x in a year. Feature work stayed stuck at 58%.
- Only 6% of execs can identify AI ROI: faster output just piles up elsewhere.
- The fix? Track the innovation ratio, and deliberately reinvest freed-up capacity, or the org absorbs it.
AI investment is accelerating far faster than engineering innovation. DX’s State of AI Impact in Engineering Report found that median quarterly AI spending in the technology sector jumped from around $1,500 to $44,000 in just 12 months – almost a 28-fold increase.
Yet that spending surge has barely changed how engineers allocate their time. The share of engineering work devoted to new features remained stuck at around 58%, suggesting that the additional capacity created by AI is not translating into more innovation.
Your inbox, upgraded.
Receive weekly engineering insights to level up your leadership approach.
LeadDev’s AI Impact Report 2026 found that 90% of organizations using AI tools now pay for AI models, moving them beyond experimentation and into funded, sanctioned use.
Around two-thirds also pay for both autonomous coding agents, such as Claude Code and Devin, and AI-powered code completion built into traditional development environments.
“The noticeable disconnect between investment and innovation exposes a fundamental misunderstanding about where value is actually created in software engineering,” says Jeff Watkins, chief AI officer at Consultancy NorthStar Intelligence.
The AI efficiency paradox
AI is accelerating code production, but if meetings, CI delays, and unclear requirements still outweigh those gains, are developers using the extra capacity for innovation – or is it being absorbed by organizational overhead?
The Atlassian Teamwork Lab calls this the “AI efficiency paradox.” Just 6% of executives are confident they can identify organization-wide AI ROI, according to Atlassian’s 2026 State of Teams Report, based on 12,000+ knowledge workers and 170+ Fortune 1000 executives.
As AI boosts individual output, work can pile up at reviews, approvals, and other human-judgment gates – eroding those gains.
Watkins says AI is speeding up engineering work, but the wider delivery system is absorbing the gains rather than converting them into business value. Developer throughput is rising, yet developers don’t feel any faster.
“If an engineer can produce a pull request twice as quickly, but it then sits waiting for human review, waits for an overloaded CI pipeline, gets delayed by a security approval, or has to be reworked because the original requirements were ambiguous, this is a problem. It shows that the organization has shifted the bottleneck elsewhere, rather than becoming twice as productive,” Watkins adds.
More like this
What’s the solution?
For engineering leaders, the answer is to track the innovation ratio over time. If it stays flat despite higher throughput, the problem may be prioritization and organizational bottlenecks – not the tools themselves.
For AI to genuinely drive innovation, leaders need to look beyond developer productivity and start measuring flow and outcomes across the entire system, Watkins explains.
“Engineering leaders now need to identify where their new bottlenecks are, redesign the processes around them, and make an explicit decision about where the capacity created by AI should go.”
If they want AI investment to result in more innovation, that released capacity has to be deliberately reinvested in discovery, experimentation, reducing technical debt, and building new products, Watkins says.
“Otherwise, the organization will simply absorb it into the existing machinery.”
