Over the course of my career, I’ve built teams and transformed organizations through four major technological disruptions: web, social, mobile, and streaming.
Each wave arrived with the same promise and the same trap: Move fast, deploy broadly, figure out the ROI later. I navigated each of those disruptions inside HBO, one of the most beloved and high-profile cultural brands in the world. Bold in its DNA. Deeply risk-averse when it came to exposing that brand. Rooted in decades of success and a long-standing way of doing things.
That tension taught me something that is more relevant than ever: In a period of radical transformation, simplicity is a competitive advantage.
That is the discipline artificial intelligence, the latest wave of disruption challenging business leaders, now demands. When there’s no guarantee that the hype of a new platform or product will be able to deliver when it’s tested in practical team workflows, the ability of organizational leaders to focus their efforts becomes more important than ever.
That means stripping their operations down to the most essential, efficient architecture, then amplifying only where it genuinely warrants it.
I call this approach AI Minimalism.
Near-term realities
With AI, deployment is moving faster than any prior wave of disruption because AI accelerates acceleration itself. Many organizations are already looking toward fully autonomous agentic systems and recursive self-improvement—the ability of an AI system to improve itself with minimal human guidance—not as distant possibilities but as near-term competitive realities. The goal is to compound advantage over time and enable more intelligent, sophisticated automation as systems learn from clean, well-governed inputs.
But covering more ground faster creates better outcomes only if you know which direction to run. And the infrastructure, human and organizational, is buckling under the weight of what we’re building on top of it.
Data from Microsoft released in May shows that using AI was more expensive than paying human employees in some contexts. A week later, Uber reported that the company had exhausted its annual AI budget just four months into 2026, noting that it could not tie AI spending to the development of any new “useful consumer features.”
Just last month, an internal memo from Meta informed employees that it expected to “move toward managing AI tokens [a single unit of data processed by AI models like ChatGPT or Claude] in a more structured way” starting in 2027, noting that it had spent billions for internal use so far in 2026.
As leaders grapple with how to safely and effectively embed AI within their teams and workflows without sacrificing the judgement and creativity that sets them apart, the answer is less, not more.
That is the heart of AI Minimalism.
Three steps
In practice, this means doing three things in tandem.
1. Overhaul your knowledge base.
Your AI is only as good as the information it has access to.
When an organization’s “corpus,” the collection of data used to train a large language model, is massive and disorganized, a team is naturally going to use more tokens because the underlying LLM has to work harder every time it’s prompted in order to find the right information.
Building smaller data sets with intentionally selected files that are hyperspecific to the automation you’re trying to build helps ensure accuracy on top of a more efficient workflow.
This can be as simple as a list of 50 facts or data points about your company that have been verified, giving the AI system proper context and ensuring that the human employee can measure the clarity and accuracy of its outputs.
From there, you have to develop a set of governing principles that dictate who has access to those files and how regularly they’re updated. Even if dozens or hundreds of employees can access a particular AI project or automation, limiting the permission to edit those files to a handful of people ensures that the body of information referenced in the automation is always accurate. And updating the information at a structured cadence means it will always be recent.
Because the automation is built efficiently and accurately based on a very specific set of assets, the token load will naturally be lower.
2. Evaluate your tech stack honestly.
Every company has a tech stack, usually a mix of enterprise tools like Microsoft 365 or Google Workspace, alongside specialized tools specific to their industry.
When adding AI into the mix, many companies are finding that they don’t need nearly as many software platforms as they used to.
Start by creating a list of all the software platforms your team uses, how you use them, who uses them, and how they tie to business workflows. Most importantly, you want to note the specific value that these tools create for your organization.
From there, identify a list of new tech platforms that you think could address any pain points or open up the opportunity for long-term efficiencies.
- Trusted platforms like Microsoft (Copilot), Google (Gemini), and Adobe (Firefly) have added generative AI capabilities into select payment tiers, making for easy experimentation for teams that are already using these enterprise tools.
- Popular new platforms like OpenAI’s ChatGPT and Anthropic’s Claude have emerged as market leaders, and both offer enterprise plans that provide enhanced data security protocols.
- Many leaders are experimenting with open source models like Deep Seek and Kimi, both from China. Open models can be more cost effective because you pay for overall computing power vs. individual token usage, and you generally avoid hefty subscription and licensing fees that the closed models charge.
This is where the value of your team really comes into play. If you are relying on fewer platforms to do the same amount of work, it’s essential to have experts on your team who know what the ideal final product should look like, and can replicate that process to achieve the product inside fewer tools.
3. Identify core workflows to optimize.
Once you have identified which existing platforms are essential to your team, and which new platforms you’d like to optimize, the next step is to focus on a few specific workflows to test. These workflows should offer high potential upside, low risk, and the capability for clear measurement.
In practice, this means layering experimentation on top of real work in rapid iteration cycles. Rather than traditional A/B testing, one team should run a selected workflow twice in tandem—once conventionally, once with AI—and document the delta to note the prompt and tool workflows used, where AI helped, what new problems it created, and the performance against key metrics. The only variable between the two scenarios is the AI itself.
The resulting documentation will help leaders measure the efficacy of any new potential workflow and report their findings to other organizational leaders. It will also allow the wider team to replicate effective workflows for future initiatives. (This is where cost savings comes in—one team burns through tokens during experimentation so that the wider team can use AI in intentional, cost-effective ways.)
The goal is to incrementally scale with tools that are available to you, while strategically testing how new tools might layer on top of them.
Too valuable to break
The corporate culture at HBO was bold by instinct but deeply protective of its carefully curated brand reputation. Running thousands of digital and social campaigns over my time there taught me that introducing something different inside a complex organization required being extremely clear about what it was, why it would make a difference, and how it would leverage everything that already existed. Because the organization was too valuable to break carelessly.
In a world of AI Minimalism, alignment should come before experimentation. That means more collaboration at the strategic level, with departmental leaders sharing priorities and areas of focus before teams start vibe coding on new projects. This requires a culture of safety, where leaders feel comfortable taking risks.
Companies that are serious about AI implementation should also assign an AI enablement lead to be a neutral voice who can moderate their efforts and make sure they continue moving forward.
This is the essential work of this moment that nobody is talking about: turning implicit organizational knowledge into explicit, reusable, consistently updated logic that AI systems can actually work from. It’s not glamorous and it doesn’t get trumpeted in headlines or keynote presentations.
The organizations that come out of this moment stronger will be the ones that chose the right problems before reaching for the tools and treated the infinite nature of AI output as a reason for more discipline.
In a world of AI Minimalism, amplification is powerful only when the signal underneath it is clear.
Less is where the real work begins.
