Few technologies have ever swept through the world's boardrooms with as much breathless urgency as generative AI. Executives have poured tens of billions into pilots and platforms, expecting clarity, efficiency, and competitive advantage. Instead, most have found themselves mired in frustration. Far from simplifying work, AI initiatives often produce what the Harvard Business Review recently dubbed “workslop” — an avalanche of half-useful drafts and repetitive outputs that clog inboxes, and slow decision-making down to the proverbial molasses in January.
The scale of this disillusionment is extraordinary. According to Project NANDA at MIT, which examined more than 300 corporate AI initiatives, 95 percent of enterprise projects show zero measurable return. Pilots abound, enthusiastic "must have" implementations are ordered, yet only 5 percent reach production with any real impact on profit and loss.
Tech and media show some movement of the efficiency needle, but seven of nine major sectors, including finance, healthcare, and energy, remain unchanged despite years of hype. The researchers call this gulf the GenAI Divide: a small handful of organisations capturing real value while the vast majority tread water and spend vast sums of money, without ever getting far off the starting line.
The reasons are not defective model quality, stifling regulation, or – certainly – lack of investment. The problem is that most tools do not adapt, and do not integrate with the living tissue of work. They churn out outputs, but they lack memory and persistence. They cannot evolve alongside shifting contexts or embed themselves deeply in workflows. That is why generic chatbots like ChatGPT feel immediately useful to individuals – they are flexible, conversational, and responsive, yet enterprise systems costing millions fail to cross into daily operations.
The cost of this failure is more than wasted money – it ends up as strategic paralysis. Leaders find themselves managing ever more data and outputs while struggling to extract genuine insight. The temptation is to push harder, to demand replacement of human work with machine-generated efficiency, but the evidence shows this road leads only to brittle systems and mounting disillusionment. The way forward lies not in replacement, but in augmentation: human ingenuity amplified by machine processing power.
Why art thinking belongs at the core
This is where art thinking becomes indispensable. Unlike managerial logic, which is trained to reduce ambiguity and enforce efficiency, art thinking thrives in ambiguity, iteration, and open-ended discovery. Artists do not treat uncertainty as a failure state; it is their raw material. They are practiced in reframing problems, finding meaning in gnarly and incomplete inputs, and treating each “output” as step in an ongoing process; provisional and subject to change.
Here lies the crucial distinction. In business, “innovation” is often treated as an add-on: a project, a lab, or a sprint layered awkwardly on top of operations that otherwise seek stability and control. It is little wonder that such innovation efforts often fail. The road to innovation nirvana is truly littered with countless corpses of corporate VC efforts and co-creation programmes.
For an artist, by contrast, continuous innovation is simply the work itself. A painter will rework a canvas dozens of times, sometimes discarding it or painting another image altogether over the top. A writer rarely considers a book “finished” simply because it is published; there is always another chapter, another revision, another story gestating. A musician never stops looking for the next riff, the next variation on a familiar melody, the next collaboration that changes everything. Each output is not an end but a milestone in a longer journey, a stepping stone to the next work and the next. Getting wet in the river of making something new IS the work. It is not to be avoided, it is to be relished.
Business tends to look at products as endpoints – perfected, scaleable, and able to be delivered in volume. Artists look at their works as waypoints on a river that has been flowing long before and will continue long after. This difference of horizon is not trivial. It goes to the heart of why AI transformation has stalled. Too many organisations are looking for perfect outputs – reliable, scalable, automated deliverables – when the more urgent need is to learn from imperfect inputs, iterating continuously and embedding adaptation into the system.
The GenAI Divide as a failure of iteration
The MIT study makes this painfully clear. Organisations that remain stuck on the wrong side of the GenAI Divide are those treating AI as an endpoint generator of perfect outputs. They buy brittle enterprise systems, expecting polished answers, only to discover that these systems repeat the same mistakes, require constant re-prompting, and fail in edge cases. Ah, yes, edge cases – they are now the norm, in case you've not heard.
A corporate lawyer described paying $50,000 for a contract-analysis tool but reverting to ChatGPT because at least it allowed iterative dialogue. Yet even ChatGPT she would not trust for high-stakes contracts; it remembers context incompletely and only learns from edits after multiple re-prompting.
By contrast, the rare organisations that cross the divide are those treating AI as a collaborator in a process of ongoing iteration. They may, for example, deploy “agentic” systems – tools designed to remember, learn, and evolve. (Yes, there is much hype about those systems right now, and, you know, caveat emptor.)
In early pilots, customer service agents now handle entire enquiries end-to-end, and financial processing agents monitor and approve transactions with feedback loops built in. These systems succeed because they are designed less like a finished product and more like an evolving practice, one that develops over time through continuous input, correction, and refinement. In other words, they succeed because they operate in a way familiar to people, and even more so, to artists.
What business can learn from artists
The implications are profound. If innovation in business is often an awkward add-on, then executives need to study those for whom innovation is simply life: artists. From them, leaders can learn three things that directly apply to AI transformation:
First, that no work is ever “finished.” A product launch or pilot is not an endpoint but a point of departure. AI systems must be expected to evolve, accumulate memory, and adapt over time – just as a painter revisits a canvas or a musician reworks a melody.
Second, that outputs are milestones, not monuments. Success lies in seeing business processes as rivers, where each iteration contributes to the flow. AI is not there to deliver a perfect artefact but to enrich the current, providing inputs and provocations that humans can shape, correct, and redirect.
Third, that imperfect inputs are more valuable than perfect outputs, exactly because the process is more important than the mythical end goal. Artists do not crave polish; they crave material that can be worked – polish comes later. For business, this means valuing systems that provide adaptive, “gnarly” data to work with, rather than polished but shallow deliverables. An AI draft that provokes new thinking may be more valuable than a “finished” report that misses the point.
Where augmentation already delivers
Examples of augmentation working in practice are emerging. The MIT study highlights back-office automation, a neglected area where AI has delivered millions in savings. Organisations that reframed their efforts away from polished front-office demos and toward iterative process improvement were able to eliminate outsourcing contracts, cut agency spend by thirty percent, and reduce risk-checking costs by millions annually. These were not cases of replacing humans with machines. They were cases of augmenting human oversight with adaptive systems, freeing staff from tedious work while retaining judgment and accountability.
Similarly, the so-called “shadow AI economy” shows what happens when individuals are given freedom to experiment. Employees using personal AI accounts have already automated large portions of their workloads, far more effectively than official initiatives. They succeed because they work iteratively, adjusting prompts and reworking drafts until they get where they want to be. In other words, they use AI the way an artist uses a sketchbook, not to produce a final artefact but to explore possibilities. Organisations that harness this shadow economy, rather than suppress it, will find themselves closer to the creative flow that produces transformation.
The narrowing window
Time, however, is short. Project NANDA warns that within eighteen months many enterprises will lock themselves into vendor relationships that will be nearly impossible to unwind. As adaptive systems accumulate memory of workflows and data, switching costs grow prohibitive. Companies that cling to the fantasy of perfect outputs risk binding themselves to brittle tools whose costs can only grow. Those that embrace augmentation – pairing art thinking with adaptive AI – will compound their advantage with every cycle of learning.
Conclusion: the river, not the rock
The lesson is clear. Generative AI is not a machine for perfect outputs, as much as the hype spinners would like us to believe it. It is a collaborator in a process of imperfect inputs, continuous iteration, and adaptive discovery. Artists have always known this way of working, as their work is never finished, their outputs are milestones on a river of creation, and they thrive not on polish but on working material that demands revision.
Business, by contrast, too often imagines its products as perfected rocks: polished, scaled, and immovable – used to skip across the river so as not to get wet. That mindset is precisely what has left 95 percent of AI initiatives stranded on the wrong side of the GenAI Divide.
To cross the divide, leaders must relish the idea of getting wet and step into the water without knowing what is at the bottom. They need to begin to think more like artists. They must accept that innovation is the work, not an add-on to the work. They must learn to value imperfect inputs, continuous discovery, and outputs as milestones rather than monuments. And above all, they must embrace augmentation — human ingenuity working with machine power — as a path to transformation that is real, resilient, and lasting.