Critical Thinking Is the Best Governor of AI

A little over a month after the Department of Defense announced that nearly half of its 3.5 million employees were using AI at work, members of the Army's Combat Capabilities Development Command received an email with an uncomfortable message: they had burned through the entire annual token allocation in months. The Army had promised unlimited access in May 2026. By mid-June, the pool was gone. One Army employee put it plainly: the whole organization had consumed a full year's worth of tokens for a single service in a matter of weeks.

The Army isn't alone. Meta encouraged engineers to "tokenmaxx" — a corporate rallying cry for maximizing AI usage — then took down the leaderboard tracking consumption and began floating caps on individual use. Uber watched its engineers exhaust a full year of generative AI tokens in four months. Three organizations. Three different industries. One identical outcome.

The instinct is to call this a supply problem. Buy more tokens. Raise the limits. Expand the budget. But that instinct is wrong.

This is a value problem. Organizations pushed AI into the hands of their workforce without answering the more important question: what should it actually be used for? When employees are rewarded for high usage, they use it for things that don't need it, for tasks faster done manually, for content no one reads and summaries no one acts on.

Token exhaustion is proof that adoption happened, but not necessarily that AI is working.

How do organizations ensure AI is being used where it actually creates value? That's the question this piece is built to answer.

AI Is Everywhere. The Question Is How.

The enterprise adoption debate is over. AI won. Security teams use it to draft incident summaries. Marketing uses it to generate content. HR uses it for job descriptions. Legal uses it to accelerate document review. Procurement uses it to flag supplier risk. Customer success uses it to write account summaries. Operations uses it to build executive updates. Across every function, in organizations of every size, AI is already embedded in the daily workflow.

The question is no longer whether to adopt AI. It's how.

Unstructured adoption — where employees reach for AI tools without clear objectives, without verified inputs, and without accountable review — creates a different kind of risk. The risk of using it badly: where outputs go unvalidated, conclusions get accepted at face value, speed becomes the metric, and accuracy becomes an afterthought. A sound enterprise AI adoption strategy defines when AI actually improves outcomes. 

More AI Doesn't Mean Better Work

Most enterprise AI dashboards track the wrong things, like users enrolled, prompts submitted, tokens consumed, and hours theoretically saved. These are vanity metrics that just tell you people are using AI, but they say nothing about whether the work is getting better.

The Army is a case in point. By mid-2026, the service had burned through an entire year's worth of AI tokens in a matter of months. Usage was high. Value was unclear. One Army employee told Wired the tools had been unreliable — one model even claimed to have completed a task it hadn't. That's not a usage success story. That's noise dressed up as productivity.

Without structure, AI doesn't produce better outcomes. It produces faster ones — and fast wrong answers are worse than slow right ones. Because when unverified information gets published, overconfident conclusions get escalated and decisions get made on data no one checked. 

Leaders should not measure adoption curves and call it progress. Responsible AI adoption demands a harder question: Are we getting better results, or just generating more content?

More AI isn't the goal. Better outcomes are. And reaching them requires human oversight at every step — not as a policy checkbox, but as standard operating procedure.

Critical Thinking Is the Governor

Technology doesn't determine quality. People do. Critical thinking has always been the variable that separates useful output from expensive noise.

This is the argument most enterprise AI strategies skip entirely. It's easier to buy a platform, set a policy, and measure token consumption than to confront the harder truth: AI quality is a human problem. The model doesn't decide whether the output is worth using. The professional does.

Responsible AI adoption requires a specific set of human behaviors that no software can replicate. Professionals must ask the right questions before they ever open a prompt window, and validate what AI returns instead of trusting it because it sounds confident. They need to recognize bias — in sources, in outputs, in their own assumptions — and understand context the model doesn't have. Judgment has to happen before anything reaches a decision-maker, and the findings need to hold up when someone pushes back.

Critical thinking governs when AI is used, how it's used, and when human expertise takes over entirely. Without it, AI governance workflow becomes theater — policies that exist on paper while bad outputs circulate freely.

The Army's token problem is the clearest illustration of this. Employees were burning through it because no one had built the habit of asking: Does this task actually benefit from AI? That's a critical thinking failure, and no usage cap fixes that. No governance policy fixes a workforce that hasn't learned to use AI intentionally. Limits on consumption are a symptom response, but human oversight of AI requires a deeper intervention — one that starts with what professionals believe AI is actually for.

The governor was never going to be a token counter. It was always going to be judgment.

Good Workflows Naturally Optimize AI

The most effective organizations don't push employees to use AI more. They build workflows that help them use it better.

Pushing for more usage drives the Army problem. Building for better usage creates the opposite dynamic: AI gets applied where it delivers genuine value, and nowhere else.

What does a good AI workflow actually look like? It follows a clear sequence. Start with a defined objective — know exactly what question needs answering before a single prompt is written. Work from reliable, verified source material. Garbage in, garbage out still applies. Use AI to eliminate the repetitive, low-value work: formatting, summarizing, structuring, drafting first passes. Then require human review before any output gets published, sent, or acted on.

When these conditions are met, AI usage becomes intentional, not reflexive. Token budgets stop being a constraint because people stop reaching for AI on tasks that don't justify it.

This is the enterprise AI adoption strategy most organizations skip. They never define what "good" looks like. Without a structured AI governance workflow, every employee is improvising — and at scale, that adds up to real, measurable loss.

AI ROI for enterprise doesn't come from maximizing usage. It comes from maximizing what you get back for the effort you put in. The organizations that figure that out first won't need a token policy. They'll have something better: a workflow that makes waste impossible.

Where Indago Fits

Indago is built on a simple premise: AI is part of the workflow, not a replacement for it.

Analysts and operations teams use Indago to produce executive briefings, supply chain risk assessments, travel risk reports, market intelligence summaries, compliance updates, customer-facing reports, and operational overviews. Not by prompting a chatbot and hoping for the best — but by working within structured, reusable templates that define purpose, audience, and analytical standards before a single word is generated.

The AI handles the repetitive drafting work. The analyst stays accountable for validating sources, applying context, and signing off on conclusions. That accountability isn't optional — it's built into the workflow.

That accountability produces intelligence that can be defended, not just delivered. For organizations serious about responsible AI adoption and sustainable AI ROI, that distinction matters more than any usage metric. 

The Only Governor That Matters

Don't strive to maximize usage. Strive to maximize value. The Army’s problem was never tokens; it was discipline and critical thinking. 

Ready to build AI workflows that produce results your team can stand behind? Book a Demo to see how Indago helps enterprise teams put critical thinking back at the center of their AI strategy.