In a recent LinkedIn post, Betsy Tong discusses why companies’ AI investments may falter when they build on data that systems cannot reliably interpret or access. This article reports on Tong’s argument in that post; it is not written by Tong. She contends that the central obstacle is often a lack of shared meaning across organizations, rather than a problem that can be solved simply by buying more AI tools.
“The more AI changes.
The more the problem stays the same.”
Why more AI may not resolve old problems
Tong describes a recurring pattern in major technology transformations: organizations introduce new tools while relying on data they do not fully trust, systems cannot readily reach, and employees must manually reconcile. She warns that adding AI agents to this environment could raise the stakes because agents can make decisions and act across systems at a pace that leaves little time for human correction.
According to Tong, failed pilots are a sign that organizations may be moving ahead without addressing the foundations those projects depend on. She recalls feeling regret over “perfect pilots that died,” suggesting that time spent making data usable could have changed their prospects.
The challenge is shared meaning
Tong is skeptical of treating the goal as simply “cleaning” data. In her view, organizations are too complex for a universal clean-up effort to be quick or straightforward. The more practical starting point, she argues, is agreeing on what important business terms mean and establishing a structure that AI can use.
“Customer means one thing to sales, another to finance, a different thing to support.”
As Tong notes, definitions can be technically correct within individual systems while still conflicting across departments. Her example of Susie, who uses a spreadsheet created in 2017 to translate between those definitions, illustrates how institutional knowledge can be concentrated in a single employee. Tong says Susie is retiring in two weeks, underscoring the risk of relying on informal workarounds.
A layer between data and AI
Tong proposes creating a layer between complex underlying data and AI applications. In her description, that layer would define key business entities, such as customers, accounts and orders, along with the actions an agent is allowed to take. AI would work through those definitions and rules rather than directly navigating inconsistent systems.
“AI operates through that layer, not at the scary mess underneath.”
That approach, Tong argues, could give companies a way to govern what AI can read and do. The post does not offer a detailed implementation plan or claim that this layer alone guarantees successful AI projects. Instead, it makes a leadership point: decisions about definitions and authority belong to the organization.
Tong closes with a pointed reframing: “You don’t have a data problem. You have a meaning problem.” By casting the issue as a decision problem, she challenges executives to resolve disagreements across teams before blaming a model when an AI deployment falls short.
📝 About This Content
This article is based on insights shared by Betsy Tong on LinkedIn.
📅 Originally posted on November 7, 2026 | View original post on LinkedIn →