A Structural Test for Technology Mandates
By Michael Ulrich
Michael is the founder and principal of Thornwright, focused on engineering leadership, technical recovery, and operational execution in complex product environments.
A recent report in The Wall Street Journal (TechFirms Aren’t Just Encouraging Their Workers to Use AI. They’re Enforcing It.) revealed a significant shift within several technology companies. Employees are no longer merely encouraged to utilize artificial intelligence tools; in some organizations, they are mandated to do so. This policy change is not particularly novel. Firms have previously mandated tool adoptions, such as ERP systems, ticketing systems, coding standards, and development frameworks. Ultimately, the leadership of these companies sets the operating environment.
This time, what’s noteworthy is the rapid pace and intensity of this enforcement. AI adoption has transitioned from experimental phases to an institutional expectation within a matter of quarters.
The question worth examining is not whether AI is valuable. It almost certainly is. The question is whether the structure surrounding adoption is ready.
As capital and narrative outpace organizational systems, companies mistake visible adoption for operational improvement, leading to technology adoption failures. This pattern has been observed across multiple waves of enterprise technology adoption, and AI appears to be the latest case.
The Structural Problem With Forced Adoption
Technology enhances productivity by altering the structure of work. However, mandating a tool doesn’t achieve this. It merely modifies behavior without necessarily transforming the system in which it operates. In fact, it can even lead employees to work for the tool rather than the tool working for them, as they are compelled to use it.
This often exhibits a common pattern:
1. Leadership declares adoption a priority.
2. Teams integrate the tool into existing workflows.
3. Early productivity signals are ambiguous.
4. Enforcement increases to ensure compliance.
5. Organizations claim success based on usage rather than outcomes.
From the outside, the transformation appears to be successful. However, inside the organization, the experience is often different. This includes additional verification steps, inconsistent output quality, and a widening gap between official narratives and operational reality. This doesn’t mean the technology is flawed; it means the system hasn’t yet adapted to the technology.
The Adoption Reality Test
Before mandating any new technology, leaders should examine these structural questions. These questions are not philosophical. They determine whether adoption will create leverage or friction.
1. What operational problem does this technology address?
Many technology initiatives begin with a market narrative. Competitors adopt the tool, investors anticipate its success, and the media portrays it as transformative. Leaders feel compelled to act swiftly, which can be understandable. However, strategic timing is paramount. Simply rushing to get the new shiny toy because everyone else is doing it isn’t a sufficient reason.
Without specifically identifying a bottleneck in AI adoption, it demonstrates that the tools are being used without a clear understanding of their objectives. Consequently, people simply use the tools without a consistent vision for their intended purpose. In the absence of a well-defined operational objective, the organization may be operating without a clear direction. The risk is not merely embarrassment; it’s the misallocation of attention and capital.
2. How are systems adapted to accept the technology?
Organizations often assume that productivity gains automatically follow the introduction of new tools. However, in practice, productivity improvements arise from workflow redesign. Introducing a tool without modifying surrounding processes often leads to unexpected friction. Employees must adapt existing systems to accommodate the tool rather than allowing the tool to reshape the workflow. Consequently, the adoption may appear successful, but it actually functions as overhead.
Let’s consider a simple engineering example. If AI-generated analysis and documentation is introduced without modifying review standards, testing protocols, or quality controls, engineers may spend additional time validating the output. The tool accelerates generation but also increases verification, resulting in a new bottleneck.
When this happens, productivity improvements stall. The organization has added a new step without removing an old one. Technology only creates leverage when the structure of work changes alongside the tool.
3. Are employees aligned with company objectives?
Leadership mandates are sometimes necessary. Security policies, compliance systems, and standardized infrastructure often require enforcement. It’s part of business, and employees accept that as part of the job.
But in knowledge work, mandates are often a signal that alignment has not yet formed. If employees believe a tool genuinely improves their output, adoption spreads organically. When adoption must be enforced, it suggests operators remain unconvinced or otherwise misaligned with the company objectives.
This does not mean employees are correct. Operators can underestimate the long-term potential of new tools. But enforcement produces a predictable side effect: cosmetic compliance. Employees demonstrate usage without necessarily changing how they think or work. Output is shaped to satisfy adoption metrics rather than improve results.
Leadership sees rising usage statistics and interprets them as success. Meanwhile, the operational system remains largely unchanged. The distinction between visible usage and actual integration becomes critical.
4. Has additional oversight been accounted for during adoption?
Organizations reduce workforce capacity based on projected efficiency gains, which poses a more consequential structural risk. Automation historically leads to productivity improvements, but these gains rarely materialize immediately or evenly. When firms compress workforce capacity before productivity gains stabilize, they eliminate redundancy at the same time variability increases.
For instance, in the context of AI-assisted engineering, output may increase, but consistency may decline. More output is produced, but more review is required. Senior engineers are forced to absorb additional verification work instead of focusing on higher-value tasks just to support the tool adoption.
If organizations simultaneously reduce headcount, they may discover that the new bottleneck is human oversight. What was expected to be efficiency becomes operational strain. This risk is not theoretical; it has repeatedly emerged during previous automation waves.
5. Is the technology depleting understanding and comprehension?
The most challenging aspect of structural analysis is measuring its impact. Technologies that generate outputs effortlessly can gradually diminish human capabilities. This effect often manifests gradually, leading to its neglect during the initial adoption phase.
I have personally witnessed this phenomenon firsthand. Junior engineers who once acquired a deep understanding of analysis from the ground up, grasping the underlying physics, now rely on spreadsheets and automated tools without comprehending their context or outcomes. This has resulted in significant gaps in their capabilities when senior engineers retire. What would have once been compounded expertise that transformed a junior engineer into an effective senior engineer has now devolved into button-pushing and visually appealing graphics without a understanding of the results.
If AI extends these automated systems to perform an increasingly significant portion of foundational tasks, including result interpretation, junior employees may acquire proficiency in tool usage without developing the essential skills that these tools rely on. This poses a long-term capability risk. Organizations may maintain productivity in the short term but weaken their expertise pipeline.
In extreme cases, employees transition from creators to validators, reviewing automated outputs instead of constructing systems from scratch. The organization becomes reliant on tools that fewer employees fully comprehend.
The Counterargument
Advocates of aggressive AI adoption raise a valid point: waiting for structural readiness could cause companies to fall behind competitors. Technological advantage often goes to early adopters willing to experiment aggressively.
There’s evidence to support this view. Companies that adopted cloud infrastructure early gained scalability advantages, while those that embraced mobile platforms quickly captured emerging markets. From this perspective, mandates might simply accelerate learning.
This argument has merit. Organizations do need experimentation. Even imperfect adoption generates data.
However, experimentation and enforcement are not the same thing. The danger arises when experiments are prematurely scaled into mandates before the underlying system adapts. At that point, the organization ceases learning and enters a phase of performing adoption.
What Leaders Should Actually Watch
The early signals of structural misalignment are rarely public. They appear inside operational systems.
Leaders should pay attention to:
• Increased verification workload after tool adoption
• Engineers rewriting automated outputs
• Growing ambiguity in performance evaluation criteria
• Rising tool usage without measurable outcome improvement
None of these signals imply the technology should be abandoned. They indicate the system surrounding the technology is still evolving. The correct response is structural adjustment, not stronger enforcement.
A More Useful Standard
Technology adoption should be judged by a simple operational standard:
• Does the tool eliminate steps in the workflow, or merely add new ones?
• If steps disappear, productivity improves.
• If steps accumulate, the organization is absorbing friction rather than gaining leverage.
Mandates cannot solve this problem. Only redesign can.
The Real Constraint
Most organizations underestimate the real constraint in technology adoption. It is not access to tools. Tools are increasingly abundant. The constraint is organizational structure: workflows, incentives, training pipelines, evaluation systems, and decision rights.
When these systems adapt, new technology can produce extraordinary gains. When they do not, adoption becomes a performance rather than a transformation.
AI may ultimately reshape large portions of knowledge work. The evidence so far suggests it will. But the firms that benefit most will not be the ones that enforce adoption fastest. They will be the ones that redesign their systems to make the technology genuinely useful.