Solving The AI Productivity Paradox In Software Development
Lance Knight is the Chief Value Stream Architect at Broadcom and former CEO at ConnectAll.
AI has genuinely changed the pace of software development. Code that once took hours now takes minutes, and developer velocity has measurably improved at the task level. Here’s the catch: While the writing phase is accelerating, the flow of value to production is not. Review, security scanning, compliance checks and deployment cannot absorb the volume that AI-assisted developers are generating.
The result is a classic local optimization problem. This is the AI productivity paradox: moving faster internally while business outcomes stay flat. The answer is not more AI at the point of coding. It is intelligent value stream automation, connecting and orchestrating the systems across the entire delivery pipeline so that the speed AI creates at one end actually flows through to the other.
Why Faster Doesn’t Always Mean More
The data is more counterintuitive than most teams expect. METR’s 2025 study of experienced open-source developers found that AI coding tools increased task completion time by 19%. The engineers themselves believed they were approximately 20% faster. Perception and reality were moving in opposite directions.
Faros AI’s analysis of more than 10,000 developers surfaces the same tension at scale. Teams with high AI adoption merge 98% more pull requests. But pull request size has grown by 154%, and code review time is up 91%. One developer captured the dynamic precisely: One person’s 10x output becomes another person’s 10x review burden.
AI acts as an amplifier. In organizations with well-integrated value streams, it accelerates throughput. In organizations where the bottleneck lives downstream from coding in review, governance and deployment, it accelerates the rate at which work accumulates in queues. Leaders see more output. Developers feel productive. Time to market barely moves.
The 2026 Strategic Context
According to ValueOps and Dimensional Research’s 2026 value stream management survey, 93% of companies are continuing to invest in delivery pipeline management. The executive agenda has converged on three priorities: reduce costs, grow revenue and improve customer outcomes. Intelligent value stream automation connects directly to all three.
But most organizations are not making this connection. According to the same survey, business strategy and software development are out of sync for 64% of companies. And visibility is declining: Only 6% of organizations report excellent visibility into their delivery pipelines this year, down from 14% last year. As AI increases the volume of code in motion, siloed data and fragmented value streams are making it harder, not easier, to see where value is stalling.
A Historical Parallel
When factories first adopted electric motors in the late 19th century, the most common approach was to replace the steam engine with a motor while keeping the existing shaft-and-belt architecture intact. The result was marginal efficiency gains. The productivity gains from electrification did not materialize until factories redesigned entirely around unit drive, with individual motors in each machine. The lag from invention to measurable productivity gains was approximately 40 years. Not because electricity failed to work, but because capturing its full benefit required redesigning the factory, not just changing the power source.
The economist Paul David identified this as a general property of transformative technologies: The gains are not in the technology itself but in the organizational redesign the technology makes possible.
Organizations deploying AI coding tools today are doing the factory equivalent of replacing the steam engine while leaving the old architecture in place. Intelligent value stream automation is the unit drive moment for software delivery: not a faster tool in a broken chain, but a connected, orchestrated system built around the volume AI actually produces.
Three Things That Must Change
Most organizations already sense where the paradox lives. The problem is knowing which three moves actually break it.
1. Make the bottleneck visible.
Start with lean flow metrics, principles drawn from lean manufacturing and value stream management. Flow time measures how long it takes a work item to travel from start to done. Flow efficiency reveals the ratio of active work to total elapsed time, exposing how much time work spends waiting rather than moving. Flow load measures work in progress in the system.
In an AI-assisted environment, flow load rises, flow efficiency collapses and flow time increases even as individual coding tasks get faster. This is the paradox made visible in data, and it is invisible on a DevOps Research and Assessment (DORA) dashboard showing green across all four deployment metrics. DORA metrics remain valuable at the deployment level, but they tell you what happened, not why.
2. Automate the handoff at the constraint.
Flow efficiency data shows precisely where active work time ends and wait time begins. That transition point is the constraint. Automation applied anywhere else produces activity without throughput improvement. Identify the constraint from flow data first, then automate the handoff at the constraint.
3. Scale quality gates with volume.
A quality gate that worked at previous volume levels becomes a bottleneck as volume increases. DORA’s change failure rate catches deployments that break production but misses the quiet churn of AI-generated code that deploys successfully and gets rewritten within days. Flow distribution, which tracks the balance between features, defects, risks and technical debt, surfaces this pattern before it becomes a production incident.
Measuring What Actually Matters
When adoption metrics become targets, they stop being reliable evidence of value delivery. Green dashboards coexist with rising total cost of ownership and flat time to market. Lean flow metrics connect engineering activity to business outcomes in a way adoption dashboards cannot. Flow time answers what executives actually care about: How long does it take for an idea to reach the customer?
Start with one value stream. Measure flow time and flow efficiency to establish the baseline. Find where the active-to-wait ratio breaks down. Automate one handoff at that point. Measure the change. Then expand from there.
The organizations that will sustain a durable advantage from AI are not the ones with the highest adoption rates. They are the ones that built a value stream intelligent enough to convert developer speed into customer value.
Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?
