Piece by piece, Alex had built the delivery engine his organization now had. AI agents drafted overnight. Test scaffolding that once consumed a sprint appeared in an afternoon. Features that used to take a quarter shipped in days, at a fraction of the old cost. Every DORA metric on his dashboard was green: deployment frequency up, lead time collapsed, change failure rate at an all-time low. The numbers that had anchored every hard budget conversation were finally, unambiguously good.
So the moment it fell apart was quiet. No outage, no escalation. Just a team lead standing in his doorway asking, for the third time that week, “What do you want us to build next?” The backlog that once stretched eighteen months ran thin by mid-quarter. Work was no longer waiting on engineers. It was waiting on his calendar. Alex had spent his career fighting a slow build engine. Now he stood in front of a fast, inexpensive one, and the organization wrapped around it had not changed at all.
The Constraint Moves
What Alex experienced has a name. Eliyahu Goldratt’s theory of constraints rests on a single insight: every system has one governing bottleneck, and improvement anywhere else is the illusion of progress. For two decades the bottleneck in most technology organizations was delivery, which is why two decades of investment went there: agile, DevOps, cloud, and now AI. For organizations that have done the structural work, the investment is finally paying off. McKinsey finds top performers achieving 16 to 30 percent improvements in time-to-market from AI-enabled development.
But Goldratt’s rule has a second half. A constraint you elevate does not disappear. It moves, and it rarely announces its new address. When the build engine becomes dramatically faster and cheaper, the constraint leaves delivery. For most organizations it surfaces where it surfaced for Alex: in deciding what is worth building.
The bottleneck did not vanish with the faster build engine. It moved to the decision itself: AI can inform what is worth building, but it cannot own the choice, or answer for it.
The Economics Invert
Alex’s first move was to rerun the prioritization model that had settled arguments for years. It came back useless. Every initiative scored “do now.” The math was behaving correctly; the world underneath it had changed. Cost-of-delay rankings, the flow economics Don Reinertsen taught a generation of product organizations, divide value by duration, and AI had collapsed the duration that once separated the options.
Delay still costs money; it just accrues in new places, in front of the decision and after the release, while customers absorb what shipped. So price delay where it now lives: rank work by its value and by how quickly that value can be confirmed, and treat a slow decision the way a slow build used to be treated, as the most expensive queue in the system. And the stakes rose while the math changed: a fast build engine does not forgive weak prioritization; it amplifies it.
For the CFO, the inversion reads differently but lands in the same place. The cost side of the ROI equation, what it takes to build, is falling fast; the return side now depends almost entirely on choosing well and on whether what ships actually gets used. Engineering efficiency will show up in the spend line either way. It reaches EBITDA only when the delivered value changes a revenue or cost curve, and that is a prioritization and absorption question, not a build question.
Put together, the inversion is simple: when building is fast and cheap, the scarce work is deciding what is most valuable and what the customer can absorb into real return. That is a capability, product management and the measurement of customer value, and it is where the operating model must invest next, because investment follows the constraint.
The Wrong Fix: Feeding the Funnel
His second move was the instinctive one: fill the intake. AI generates candidate ideas as cheaply as it generates code, and within a month the funnel was overflowing. His problem got worse. Pushing more input into a decision bottleneck does not raise its throughput. It raises work-in-process and lengthens the very queue it was meant to feed.
The honest objection is worth stating: if building is nearly free, why not build it all and let the market sort it out? Because building is cheap; having built is not. Every feature shipped spends two budgets no tooling can refill. It spends the customer’s, in attention, workflow change, and trust, where unused features clutter the product and erode confidence in it. And it spends the future’s, because everything built must be maintained, secured, and integrated for as long as it lives, and unowned complexity reassembles the old tangle at machine speed.
Ideas are now as cheap as code; the right ones are as hard as they have ever been. What is scarce is decision capacity, and it has two halves. One is cadence: deciding more often than the calendar allows. Gartner finds only 18 percent of CIOs practice dynamic, off-cycle reprioritization, and those who do are 24 percent more likely to be top performers. The other half is judgment: a shared, current definition of customer value, agreed between business and technology up and to the left, before capacity is spent. AI sharpens the inputs to that judgment, telemetry and customer signal at a depth no team had before, but it does not make the call. Cadence can be fixed with governance. Judgment is a capability, and it is the half the industry struggles with most.
The Customer Sets the Pace
Two failed fixes in, Alex called Jonah, an old mentor with a habit of answering questions with questions. Alex described the fast engine, the flooded funnel, the rankings that no longer ranked. Jonah asked only two things: “What limits the system now?” and “Is it even inside your building?” Then, being Jonah, he hung up.
The answer arrived from outside. Account managers began reporting an unfamiliar complaint: customers asking for fewer releases. Adoption lagged each launch a little further. The team was shipping faster than the people it served could change.
That absorption capacity is measurable, and it is shrinking. Gartner found the average employee experienced ten planned enterprise changes in 2022, up from two in 2016, while willingness to support change fell from 74 percent to 43 percent. Alex’s final constraint was the rate at which his customers could absorb what he delivered. Tooling can soften it, smoothing each change and surfacing adoption signals earlier, but unlike delivery it cannot be scaled on your schedule: the ceiling belongs to the customer. It can be paced to, not owned. Here the value of speed changes meaning: from throughput, doing more, to responsiveness, sensing sooner and delivering the right next thing.
The final constraint is not in your delivery system. It is your customer’s capacity to absorb what you deliver.
What This Does to the Operating Model
An operating model is what turns strategy into delivered value, repeatably. A build engine this fast and this cheap changes every component of it.
- Capabilities and domains. The change here is indirect but relentless. A fast build engine creates new capabilities faster than the business architecture can place them, and every one of them needs a home: a domain it belongs to, an owner who answers for it, and data whose ownership is settled rather than assumed. Capabilities without a home recreate the gray area faster than anyone can own it.
- Structure and boundaries. Boundaries get redrawn in both directions. When a domain needs less engineering capacity, team structures consolidate: accounts payable and invoicing keep their own bounded contexts, but one delivery team can now own both, because the demand side changed. And a new gray area opens between strategy and delivery, a seam that needs a named owner exactly as any unowned boundary does.
- Governance and decision rights. The quiet change is beneath the waterline: every deployment hands AI more decisions, and few organizations can say which ones, or at what threshold of risk and value a person must re-enter the loop. It accrues slowly, then catches the organization off guard. What a portfolio governs, what a product group governs, and what a team governs with AI now look different, and each is managing risk at a speed governance has never had to move at. Where the right measurement does not yet exist, compensating controls hold the line until it does.
- People and roles. The product roles become pivotal. Product managers and owners who connect customer needs, internal and external, to strategy, and who carry the competencies to make those value decisions well, including the decision to stop, now sit exactly where the constraint sits. That is no knock on the AI engineer or the delivery system; it reflects where the scarcity moved. And the division of labor has to be explicit: what AI does, what humans decide, and where they meet, written down rather than assumed.
- Cadence. The deepest change. Annual planning cannot govern a weekly build engine. Strategy, funding, and validation have to move at the rhythm of a continuous flow, with intake ultimately paced to customer absorption rather than team capacity. As Gartner’s Chris Howard puts it, “gone are the days when organizations planned annually and locked in workstreams for the year.”
The diagnostic does not require a framework. Ask one question of your own organization: where does work wait longest today? If the answer is still the build queue, the structural work remains. If the answer is a decision, a funding cycle, or a customer’s capacity to take on change, the constraint has already moved, and the components above are where it is hiding.
An operating model built around a scarce build engine cannot govern an abundant one. Every component, from decision rights to cadence, was calibrated to a constraint that no longer exists.
Where Alex Began
Alex did not fix this in a quarter. He began, and the beginning is the point. He shortened the decision cadence instead of lengthening the roadmap. He pushed the definition of value up and to the left so choices stopped queuing on his calendar. He named an owner for the seam between strategy and delivery, the gap he had been filling himself, one doorway conversation at a time. And he started reading customer adoption, not team velocity, as the system’s true speedometer.
If this piece leaves you with three things, let them be these:
- The constraint moves. When delivery becomes fast and cheap, find its new address before optimizing anything else.
- Upstream, the scarcity is decision capacity, not ideas. Feeding the funnel faster makes the bottleneck worse; deciding better, smaller, and more often makes it work.
- Downstream, the customer sets the pace. Absorption, not capacity, is the final constraint, so build the cadence that senses it.
The quarter’s closing review told Alex what had changed. Nobody asked how much the team had shipped. They asked what customers had absorbed, what the measurements had killed, and what the organization now knew that it had not known ninety days earlier. The build engine was still fast. The difference was that the organization around it had finally started moving at the speed of its decisions, not the speed of its tools.
This post comes from our management consulting practice, which specializes in designing and implementing operating models that align governance, processes, and technology to drive measurable business outcomes.
Sources
DORA, Accelerate State of DevOps Report, Google Cloud. Deployment frequency, lead time, change failure rate, and time to restore as the standard measures of delivery performance.
Goldratt, E. M., and Cox, J. The Goal: A Process of Ongoing Improvement. North River Press, 1984.
McKinsey & Company, “The AI Revolution in Software Development,” 2026.
Reinertsen, D. G. The Principles of Product Development Flow: Second Generation Lean Product Development. Celeritas Publishing, 2009.
Gartner, 2026 CIO and Technology Executive Survey.
Gartner, Workforce Change Survey (2016–2022), and change-fatigue research, 2024.
Howard, C., quoted in “Gartner Unveils Top Predictions for IT Organizations and Users in 2026 and Beyond,” Gartner, October 2025. (Direct quote source.)
