New research reveals a 20-point gap between companies that buy AI and those that rebuild around it, a lesson we first learned with the electric dynamo a century ago.
We are in the midst of a massive, quiet waste, spending billions on artificial intelligence only to use it as a marginally better photocopier.Last week BCG published its fourth annual AI at Work survey with 11,700 workers across 14 markets, and one comparison buried in the findings does most of the talking. Give a team better AI tools, and measurable business impact moves by roughly 5%. Set a clear strategy and redesign the work around the tools, and impact increases 5x at 25%. (BCG, June 3, 2026)
You can buy the motor for a 5% gain. But the real engine—the redesigned factory—is what unlocks the other 20%.The redesigned workflow is the transformation unlock.
At BasedAI, we're focused on closing that gap. We've been wrestling with how best to frame the challenge, and the most revealing analogy we've found wasn't written about AI. It was written in 1990, and its subject isn't silicon—it's the evolution from steam to the electric motor.
A paradox with a history
This year's disappointment with AI's productivity returns isn't an isolated event—it's a pattern. In February, a study of nearly 6,000 executives found over 80% reported no measurable impact on productivity or employment from AI after three years (NBER, via Fortune, February 2026). For economists, the reaction was less one of panic than of recognition. They saw a familiar ghost in the data and reached for the perfect line to describe it: Robert Solow's 1987 observation, made in the thick of the last technology wave that promised to change everything, that "you can see the computer age everywhere but in the productivity statistics.
When the same paradox shows up twice, forty years apart, the honest conclusion is that the problem is not the technology. It is how companies implement and utilize it after the invoice is paid.
The economic historian Paul David tackled this same paradox head-on in his 1990 paper, The Dynamo and the Computer (David, 1990). To explain the lag, he looked back at the last general-purpose technology to underwhelm for decades before it finally delivered: the electric dynamo.
The factory that changed nothing
The numbers are humbling. Edison opened his first central power station in 1881. By 1899, electric motors accounted for less than 5% of factory mechanical drive in the United States. It took until the early 1920s, four full decades, for even half of factory drive to be electrified, and for the productivity gains everyone had been promised to finally appear in the data.
The reason the productivity transformation lagged is the root cause that should make every operations and finance leader uncomfortable.
The first electric factories kept the architecture of the steam age. A steam engine had driven one central shaft, and every machine in the building was arranged around that shaft, connected by a tangle of belts and pulleys. When electricity arrived, the obvious move was to pull out the steam engine and drop in one large electric motor turning the very same shaft. Cleaner, quieter, fewer boilers. And barely more productive. The factory still had to be built around the drive. The work still flowed the way the belts allowed, not the way the work itself wanted to flow.
The transformational leap came later, and was the result of a modest change. Each machine got its own motor. Once that happened, the central shaft was no longer necessary, and once the shaft was gone, the entire factory could be rearranged.
That is when the gains arrived. The factory layout - the workflow - could follow the flow of materials instead of the geometry of the wiring. Buildings got lighter, lower, naturally lit, because there was no overhead transmission to brace for. A single department could be moved or rewired without shutting down power to the whole plant. Roughly half of the acceleration in manufacturing productivity in the 1920s traces to the spread of those individual motors. The dynamo paid off only when companies stopped treating it as a quieter steam engine and started redesigning the work around it.
The Same Mistake, Ninety Years Later
We should be precise, because David was. He cautioned that computers are not dynamos, and that information flows through a company in subtler ways than current through a wire. The analogy isn’t that AI is electricity. It’s something narrower, and far more uncomfortable: any powerful general-purpose technology, when bolted onto an organization left structurally unchanged, will disappoint—and will continue to disappoint for precisely as long as the organization remains unchanged.
The 2026 data maps perfectly onto that insight. BCG’s survey found that 42% of regular AI users already reclaim at least one full workday each week. Yet two-thirds of those users receive little to no guidance on how to reinvest that time, and more than half never redirect it toward anything strategic. The researchers distilled the leakage into a single, damning observation: without deliberate redesign, saved time simply evaporates.
That is the group-drive model in a sentence. The motor has been installed, but the central shaft remains. The hours returned by the machine drain straight through a floorplan nobody reconfigured.
Bolting a chatbot onto an existing workflow is the exact same move those early electrified factories made: swapping the engine while leaving the drive train intact. The assistant merely sits alongside the same handoffs, the same approval layers, the same queues designed for human-only throughput—and it yields the same modest return. Five points of impact, by BCG’s measure, against the twenty-five available to those who rebuilt the flow around the tool.
What Actually Gets Redesigned
Let’s be specific, because “redesign the work” can sound like a euphemism for cutting the team. It isn’t. The data is unambiguous: the substitution play is the weaker bet. The gains belong to the companies that kept their people and freed them.
Picture the most overwhelmed person on a small team. Visualize everything on their plate that doesn’t actually require them: inbox triage, first‑draft research, CRM upkeep, lead qualification, ticket sorting. In the steam‑age design, all of it flows through that single overloaded person, because that’s where the central shaft happens to run.
The redesign is straightforward: give that routine layer its own motor. Introduce an AI coworker that owns that stratum of work and runs it continuously. Then re‑sequence everything downstream, because the bottleneck is gone. The person remains—but they are now pointed at the judgment work, the strategic decisions, the human tasks the redesign was meant to unveil.
This is the augmentation pattern, and it’s the half of the research that compounds. BCG’s respondents in redesigned workflows weren’t just 24 points more likely to report measurable business improvement; they were 20 points more likely to say their jobs had improved. The objective of the unit drive was never to empty the factory. It was to let the factory finally be arranged around the work.
The Forty-Year Lag Is Optional Now
Here, the dynamo story offers a crucial piece of hope it doesn’t get credit for. The forty-year lag was real, but it was anchored in slow capital, immature hardware, and factories made of brick and mortar that couldn’t be redesigned overnight. Today, none of those constraints govern. The redesign facing most companies is organizational, not physical. The motor is already on the shelf.
This isn’t to say the teams who bolted an assistant onto the side were misguided. Most were never offered a real choice between a quieter engine and a new factory. They were handed the engine—on a deadline, against a budget line that demanded a visible return by quarter’s end—while the redesign was left as nobody’s explicit responsibility. We founded BasedAI in part to make it someone’s responsibility: ours. Our aim is to shrink that transformation from a forty-year capital project into a decision a small team can make this quarter.
What we are building for the redesign
BasedAI is the acceleration and commercialization layer for open source AI. We build products that turn the best of open source AI into reliable, useful work for real people and real businesses. The dynamo lesson is, in a real sense, the company thesis: the capability has arrived, and the value lives in the unglamorous work of rebuilding the flow around it.
Hirebase, our first product, is the individual motor for the routine layer of a team's work. You hire AI coworkers that own a defined slice of the work, email, chat, inbox, CRM, calendar, lead signal, and produce output your people review before it goes out. It is built for augmentation rather than headcount replacement, which is the half of the data that pays. The closed beta is open at hirebase.co for solopreneurs and small teams, and design-partner conversations are underway with larger teams that want the same redesign at scale.
BasedAPIs, our second product, is the inference layer underneath: open weight models served reliably for production workloads, with economics that make it viable to put a small, specialized motor on many machines instead of routing everything through one expensive general-purpose call. It is coming soon; if you want to be early, reach us at api@basedai.co.
The rest of what we do is consolidation, the slow work of pulling a fragmented open source AI ecosystem into something a small team can actually act on. Most of it happens out of sight. What surfaces will look like more products.
Start with the layout, not the tool
Paul David summarized his paper by warning his readers off two pitfalls: undue optimism and unrealistic impatience. We will accept the first warning, and push back, gently, on the second. Impatience was unrealistic in 1990 because the lag was anchored to physical capital. This time it is bound to nothing but habit.
So if the question on your desk this quarter is why the AI line item has not moved the numbers, our honest suggestion is to stop auditing the tool and start auditing the layout. Find the shaft. Find the work that only flows the way it does because it has always flowed that way. Give the routine layer its own motor and watch your team rediscover the strategic work they were meant to do, enabling their judgment, creativity, and strategic impact to rise to the surface.
The motor is on the shelf. The factory is yours to rearrange. If the redesign is where your team is headed, the Hirebase closed beta is open at hirebase.co, and we read everything sent to hello@basedai.co.
Hirebase is a product of BasedAI, the acceleration and commercialization layer for open source AI.