In the 1890s, textile mills bolted electric motors onto the same belt-driven machinery they had been using with steam power. Productivity barely budged. It took 30 years and a complete factory redesign before electrification actually delivered on its promise.
George Sivulka, founder and CEO of enterprise AI company Hebbia, thinks we are making the same mistake right now with AI.
His argument, laid out in a new essay, boils down to a single sentence: productive individuals do not automatically equal productive firms. Handing every employee a ChatGPT subscription makes each person faster at writing emails and summarizing documents, but the organization as a whole may not see meaningful returns. The factory has a new motor. The factory floor layout is unchanged.
Seven Ways Individual AI Falls Short
Sivulka breaks the gap between personal AI tools and what he calls "Institutional Intelligence" into seven dimensions.
The most interesting is signal vs. slop. When everyone in a company generates more content faster, the total volume of internal memos, reports, and proposals goes up. But so does the noise. Individual AI tools are optimized to produce output. They are not designed to help the organization figure out which outputs actually matter.
Then there is bias. Consumer AI assistants are tuned to be agreeable. They confirm whatever the user believes. That is a nice user experience. It is a terrible way to make business decisions. Sivulka argues institutional AI systems should challenge assumptions rather than reinforce them.
The remaining five dimensions hit similar notes. Individual tools create coordination chaos when every department runs its own AI workflows with no shared standards. They optimize for time saved rather than revenue generated. They require adoption campaigns instead of genuine process redesign. And they sit passively waiting for queries instead of proactively surfacing risks and opportunities.
The Self-Serving Part (and Why It Still Matters)
Sivulka runs Hebbia, a company that sells enterprise AI for knowledge work, particularly in finance and legal. So yes, he has a direct financial interest in convincing companies that consumer AI tools are not enough and that they need purpose-built institutional solutions. His company is the solution he is prescribing.
That said, the core observation holds up. Most large organizations have spent the past two years distributing AI access broadly without rethinking how decisions get made, how information flows between teams, or how they measure output quality versus output volume. The result is exactly what Sivulka describes: faster individual workers producing more stuff, with no clear evidence that the stuff is better or that the company is actually moving faster as a whole.
The historical parallel to factory electrification is well-worn at this point, but it keeps getting cited because it keeps being accurate. The companies that eventually benefited from electrification were the ones that redesigned their physical layouts, retrained workers, and rethought production flows. The ones that just swapped the motor saw marginal gains at best.
The practical question for anyone running a team: are you just distributing AI access, or are you actually changing how your organization works? The difference between those two things is the entire argument.