New Wine, Old Wine Skins; AI Transformation in Broken Systems

There is an ancient lesson that feels remarkably relevant in the age of artificial intelligence: you cannot pour new wine into old wineskins. In old times, wine was stored in wineskins. With time, the new wine ferments, expands, and requires containers built to hold what it has become. Old wineskins, hardened by time and shaped for wine already consumed, cannot accommodate what the new wine will do. They split, the wine is lost, and the skins are ruined. Both are wasted, because the pairing was wrong from the beginning.

Now, this article is not about wine.

As I think about AI, this is exactly what many organizations are doing with AI right now. We invest in cutting-edge tools, buy licenses across every function, hire consultants at premium rates, launch pilots, and announce ambitious transformation strategies to boards and shareholders.

Beneath the surface, however, we are expecting that new technology to operate within models designed for a slower era, decision-making structures built for information scarcity rather than information abundance, workflows optimized for human cognitive limits that AI does not share, and leadership mindsets formed in environments where the pace of change was measured in years rather than months.

Although the wine is new, the wineskins are the same ones we have been using for decades, and they were already stretched thin before we started pouring.

The result is predictable, and it is happening at scale across every industry right now. AI implementations that looked promising in pilot fail to scale into production. Tools that performed well in controlled demonstrations produce disappointing results in actual workflows. Investment cases that promised significant returns deliver marginal ones or none at all. Executives who championed AI transformation find themselves explaining to boards why the transformation has not materialized.

Somehow, everyone is trying to save face.

The failures get blamed on the technology itself, or on vendor selection, or on implementation partners, or on the specific tools that were adopted. The actual cause, that the organization was never structurally prepared to meet AI’s requirements, remains excused, acknowledging this would require admitting the problem was never the wine.

Contrary to what many business leaders assume, AI does not simply automate work within existing structures. In many cases, it challenges the structures themselves. It changes how decisions get made, decisions that used to require days of analysis can now happen in minutes, and the organizational pace that was built around human analytical speed may become a bottleneck. It changes who makes decisions; capabilities that used to be concentrated in specific roles have become distributed among anyone with access to the tools. It changes how accountability is shared. When AI systems influence outcomes, the traditional lines of responsibility that assumed human decision-makers throughout the process change. It changes how value gets created and measured, because AI-augmented work produces outputs whose value is difficult to attribute to specific inputs.

These changes demand something that our old structures were never designed to provide.

We need greater clarity about what the organization is actually trying to accomplish, faster learning cycles that can accommodate the pace at which AI capabilities evolve, and a level of organizational transparency about how work actually happens that most companies have never needed to develop.

AI requires the organization’s operations to be more detailed than they have ever been. Most organizations are not.

If old structures and processes remain unchanged while AI is poured into them, the outcome is comprehensive failure that gets misdiagnosed.

The problem is that when an AI initiative struggles to scale, the results disappoint. The investment gets written off or quietly deprioritized, and the blame falls on the technology, the vendor, the consultants, or the implementation team, when in reality, the organization was never prepared to receive what AI required. The next initiative launches with different tools, better consultants, and clearer executive sponsorship, but it fails for the same reasons the first one did: the structural problem was never identified and therefore never addressed.

I predict that many organizations can go through several complete cycles of AI investment, failure, and reinvestment without ever recognizing that the wineskins are the problem.

The alternative requires something more difficult than better AI adoption. It requires redesigning the organization to be capable of holding what AI does.

That means thinking beyond the technology, but also looking at it as a leadership project, a governance project, and a cultural project that happens to be occasioned by technology.

Building new wineskins requires deliberate work across several dimensions of how the organization operates.

Update Decision Structure

Old organizational structures concentrated decision-making authority in specific roles because information was scarce and analytical capacity was limited, and it made sense to route decisions to the few people who had access to both. AI changes this by making information broadly available and analytical capacity widely distributed. New wineskins require rethinking which decisions actually need to be made at which levels, which can be pushed to the edges where the work happens, and which require deliberate human judgment that AI cannot and should not replace.

This is a fundamental redesign of who has authority over what, based on where judgment actually needs to live rather than where hierarchy has historically placed it.

Rethink Feedback Velocity

Remember the quarterly and annual reviews. Old organizations operated on quarterly and annual cycles because that was the pace at which meaningful information could be gathered, analyzed, and translated into action. AI-enabled organizations can operate on cycles measured in days or hours for many kinds of decisions, but only if the feedback loops, measurement systems, and response mechanisms are built to operate at that pace.

New wineskins require building the operational infrastructure that supports faster learning cycles, including measurement systems that produce useful data in real time, review mechanisms that can process that data quickly, and response protocols that can act on the resulting insights before the situation has changed.

Organizations that keep AI moving at a quarterly review pace are wasting most of AI capability.

Redesign the accountability structure.

Presently, accountability models assume that identifiable humans make identifiable decisions that produce identifiable outcomes, and responsibility can be traced along these clear lines. Unfortunately, AI complicates this by inserting systems whose decisions cannot be fully explained even by their designers, whose behavior in production may differ from behavior in testing, and whose influence on outcomes may be difficult to isolate from other factors.

What if we need to rethink how accountability actually works when AI is involved, including who is accountable for AI-influenced decisions, what documentation is required for those decisions to be reviewable, and what governance mechanisms need to exist for AI behavior to be controllable at scale? This is not a matter of adding an AI governance committee to existing structures; we would be building governance that actually functions when the objects being governed are AI systems, rather than the human decisions traditional governance was designed to oversee.

Invest in Talent and skill development

Professional expertise is believed to be built through years of experience in specific domains, and career progression followed the accumulation of that experience. AI changes the value of accumulated experience in complex ways. Some experience becomes more valuable because it provides context AI cannot generate, some becomes less valuable because AI can now do what the experience uniquely enabled, and new categories of capability become important that were not part of traditional career development.

New wineskins require rethinking how the organization develops, evaluates, and deploys human capability in a context where AI is doing more of the work traditional roles were built around. This includes reconsidering hiring criteria, promotion paths, training investments, and the specific capabilities the organization needs to build versus the capabilities it can now source through AI.

Clarify Corporate Culture

Over the years, organizational cultures often protected workers from complete visibility into how work actually happened, blurring the lines between performance and empirical reality in ways that reduced conflict and preserved political relationships. AI now requires that organizations be legible to themselves in ways that make this protective ambiguity impossible.

If AI systems are going to influence decisions, the decisions must be traceable. If AI is going to augment work, the work must be documented in ways that allow the augmentation to be evaluated. If AI is going to be governed effectively, its behavior must be visible.

Organizations that pretend about internal operations discover that AI adoption forces choices between the transparency AI requires and the ambiguity the culture protects. New wineskins require choosing transparency, which involves cultural work that most organizations underestimate.

Dismantle Old Leadership mindset

Leadership, as we know it, has always emphasized certainty, control, and projecting strategic clarity even in uncertain situations. AI operates in environments of genuine uncertainty, about capabilities, about outcomes, about second-order effects, about the pace of change, and leadership that projects false certainty in these environments produces worse decisions than leadership that acknowledges uncertainty and builds organizational capacity for learning through it.

Now, we will need leaders who can hold uncertainty, model learning behavior, and lead through change without pretending to know more than they actually know about where the change is heading. This is a specific kind of leadership development that most organizations have not invested in.

The organizations that do this work will be positioned to actually benefit from AI in ways that the organizations trying to skip it will not.

The gap between these two groups is already becoming visible in market outcomes, and I predict that it will widen over the coming years as the compounding effects of structural adaptation versus structural resistance accumulate.

Companies that built new wineskins will be running at a pace, with a level of decision quality, and with an operational integrity that companies still using old wineskins cannot match. The competitive advantage will be less about having better AI and more about having built the organizational capacity to actually use AI, while competitors continue trying to force AI into structures that cannot hold it.

The question every leader should be asking, and asking honestly, is which side of this gap their organization is on.

Look at the actual decision architecture, the actual feedback velocity, the actual accountability structure, the actual talent model, the actual corporate culture, the actual leadership mindset.

Compare them to what AI actually requires.

Notice how much of the current AI strategy assumes that gap will close on its own while the organization focuses on tool adoption and use case identification. It will not close on its own. It will only close through deliberate work on the wineskins themselves, work that most organizations have not begun and many have not recognized as necessary.

The choice is not whether to invest in AI, almost every organization already is.

The choice is whether to invest in the organizational transformation that makes AI investment actually pay off, or to continue pouring new wine into old wineskins and blaming the wine when the skins split.

Are you building an organization capable of harnessing the power of AI, or are you asking AI to fit into yesterday’s organization? The honest answer determines whether the AI investments already committed will produce the returns already promised, or whether they will join the growing category of expensive initiatives that failed for reasons the organizations that funded them never quite understood.

The wineskins are the work.

Everything else is preparation for work that will not succeed until the wineskins are ready to hold it.

Originally published in The AI Catalyst Report ↗