On Legacy Companies and AI

On Legacy Companies and AI#

October 17, 2025

There is a world of difference between inking a deal with the frontier lab du jour to satisfy a board of directors that the company is “doing AI”, and actually embracing technological phase change.

Legacy companies are not seeing much value from many of their AI pilots so far [1].

Why is that? Cynics and skeptics would have you believe the tech is to blame. I think there is a much more mundane explanation. Most companies are simply not prepared to use the tech effectively. What is required for an organization to use AI effectively?

The recipe is not complicated:

  • precise, explicit understanding of the work to be done

  • informatics infrastructure to route the necessary context wherever it is needed

  • a culture of truth-seeking, empirical improvement, and personal responsibility.

Legacy companies will struggle with AI because the understanding of work to be done is buried in decades of experience across thousands of people. Distilling that experience is hard to do at scale. It requires a level of personalization and individual respect that is inconsistent with corporate perception of workers as fungible automata.

They will struggle because legacy infrastructure and political self-preservation obstruct the flow of vital information at every turn. The VP who built his fiefdom on exclusive access to customer data does not suddenly become an evangelist for transparency. Corporate hierarchies built on information asymmetry will resist accordingly.

They will struggle because AI demands what corporate culture systematically erodes: accountability. AI is easily led astray by unreliable information which must be conscientiously flagged and removed. Pragmatic, measurable standards of accuracy and reliability have to replace naive expectations of omniscient genies. AI is not an abdication of responsibility, despite what salesmen would have you believe. The capabilities of AI create a burden of responsibility higher than ever before. If you can’t be bothered to read and correct a document you just generated, should you really ask a coworker to do so?

In the end, legacy companies will blame AI for failing to fix institutional problems that are all too human. And they will be surpassed.

[1] Estrada, S. (2025, August 18). MIT report: 95% of generative AI pilots at companies are failing. Fortune. https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/