The Charlatans at the Inflection Point

This is an online version of thr newsletter, written by Constellation's Chief Distiller and Board Advisor Esteban Kolsky, shares curious and interesting insights and data points distilled from enterprise technology to identify what’s notable. If you want to receive this in your inbox, subscribe here..

Artificial intelligence has crossed enough adoption, investment, and executive attention thresholds to become mainstream, but it has not yet crossed the threshold into predictable enterprise value. This is the uncomfortable period that accompanies most consequential technology transitions: spending advances faster than understanding, evidence remains incomplete, and the number of people claiming to know the path forward grows much faster than the number who have traveled it.

Executives and board members are surrounded by vendors, consultants, analysts, influencers, internal advocates, and newly minted AI experts offering strategies with remarkable confidence. Much of their advice sounds reasonable because it is assembled from familiar principles and established market language, increasingly with help from the same large language models being discussed. The resulting challenge is determining which information matters, which voices deserve trust, and which recommendations can survive implementation.

First, my take,

Three years ago, I wrote a trilogy addressing three questions that were already becoming central to executive decision-making: Who Can You Trust?, What Do You Need to Know?, and How Do You Get Your Information?. Those posts emerged from conversations with executives who felt increasingly isolated despite being surrounded by advisers, research, and opinion. Vendors brought recommendations shaped by what they sold, consultants brought methodologies shaped by what they could deliver, analysts brought market abstractions, and internal teams brought functional priorities. I described the growing presence of “charlatans of platitude”: people capable of repeating accepted principles with authority but unable to explain how those principles should alter a specific decision, operating model, or investment. The proposed resolution combined trusted relationships, curated information, and informed conversation. Information could prepare the decision, but experience, debate, and the confrontation of ideas were required to test it.

The problem has expanded considerably since 2023. Generative AI has reduced the cost of producing credible-looking expertise to nearly zero. Anyone can create a polished article, framework, presentation, or market position by assembling and repackaging the work of people who spent years developing it. The output may be accurate and professionally presented, yet contain no original experience, accountability, or understanding of what happens after the recommendation is accepted. Visibility, publication frequency, confidence, and audience size increasingly substitute for evidence that the speaker has personally made difficult decisions, managed implementation, lived with the consequences, or corrected a strategy when its assumptions failed. This has created a growing class of vociferous charlatans with a platform, and many are selected because they are loud, accessible, and apparently current rather than because they understand how to navigate the transition or establish a strategy capable of surviving it.

The timing makes this especially dangerous because we are at an inflection point in AI adoption. Generative AI moved from novelty to widespread experimentation in less than two years, and agentic AI moved from an emerging concept to a board-level priority in an even shorter period. Surveys now routinely report significant levels of organizational usage, yet the number of companies that have deeply integrated AI into core business processes remains much lower. Giving employees access to a model qualifies as adoption only in the broadest sense. Sustainable enterprise adoption requires process redesign, integration, governance, infrastructure, ownership, measurement, and the ability to operate the resulting capability after the initial excitement has passed.

Investment has nevertheless advanced at extraordinary speed. BCG reported that companies expected AI spending to rise from approximately 0.8 percent of revenue in 2025 to 1.7 percent in 2026, while Menlo Ventures estimated that enterprise generative AI spending increased from $11.5 billion in 2024 to $37 billion in 2025. PwC found that a majority of CEOs had yet to see either additional revenue or lower costs from their AI investments, with only a much smaller group reporting both. We passed the checkpoints for interest, experimentation, funding, executive attention, and organizational urgency in record time, but with limited agreement about what successful adoption should look like, how value should be measured, or which investments will remain useful as models, vendors, architectures, and pricing structures continue to change.

AI is emerging as a genuine paradigm shift. It will change how organizations structure work, value skills, design processes, make decisions, purchase technology, govern information, and distribute authority. It will eliminate some activities, reshape many roles, and alter the relationship between people and systems. Agentic AI adds another level of complexity because technology may increasingly initiate actions, consume resources, approve or deny requests, and participate in workflows rather than merely generate information. Yet the enterprise work required to navigate these changes remains recognizable. Organizations still need to establish objectives, allocate capital, select architectures, integrate systems, redesign processes, govern access, train employees, manage exceptions, measure results, and determine who owns the outcome. The terminology changes and the tools improve, but the practical work of turning technology into a sustainable operating capability evolves rather than disappears.

This cyclical quality of enterprise technology has fascinated me since my early days in the industry. Each generation believes its challenges are unprecedented, yet many of the same architectural, organizational, financial, and governance problems recur in new forms. The significant difference is that the cycles have compressed. Technology transitions that once unfolded over twenty-five or thirty years eventually narrowed to fifteen or twenty, and the current combination of cloud platforms, open-source development, AI-assisted engineering, and concentrated investment may reduce the window to seven or ten years. Companies have less time to observe before acting and much less time to recover after choosing the wrong architecture, partner, operating model, or source of advice. That compression increases the value of experience because the consequences of a poor decision appear sooner and the opportunity to correct it is smaller.

It also creates ideal conditions for charlatans because executives must act before a mature body of evidence exists. A charlatan does not need to provide false information. Most rely on technically correct statements and familiar principles: focus on outcomes, begin with the business problem, prepare the data, establish governance, and keep humans in the loop. These statements are generally true but insufficient to guide a decision. “Focus on outcomes,” for example, offers little practical direction until someone determines whether the relevant outcome is lower cost, shorter cycle time, increased revenue, improved margin, reduced risk, additional capacity, or the creation of a capability the enterprise will need later. The outcome must then be measured over an appropriate period, assigned to an owner, and separated from the data, infrastructure, process, and organizational investments required to make AI useful. That’s before we determine how an outcome varies across organizations.

The same gap exists in governance. Recommending that an enterprise govern AI is easy. Determining what an agent may access, which actions it may perform, when it must stop, how its decisions will be observed, and who remains accountable requires operating knowledge. Policies must become controls, controls must be incorporated into workflows, and exceptions must be handled without undermining the operating model. The adviser who can articulate the principle and the operator who can make it work provide different forms of value.

This is why experience matters more than expertise alone at an inflection point. Expertise explains the technology, describes what is possible, and provides the vocabulary required to discuss it. Experience recognizes recurring patterns, anticipates second-order effects, and understands which technically correct recommendation is likely to fail inside a particular organization. No one has previously operated the complete agentic enterprise now being imagined, so experience cannot provide a finished playbook. Its value lies in distinguishing the genuinely new elements from familiar enterprise problems presented through new terminology. Experienced operators know that architectures must be maintained, governance must survive exceptions, benefits require owners, integrations create dependencies, and implementation will expose compromises absent from the original strategy.

The information challenge I described in 2023 has also changed. Executives no longer face only a knowledge avalanche. They face an expanding supply of synthetic knowledge that may be correct, derivative, incomplete, decontextualized, or confidently wrong. The objective cannot be to consume more content or follow more sources. It must be to select information that contributes to a decision framework and obtain it from people capable of explaining its implications. Trust can no longer rest primarily on credentials, institutional affiliation, visibility, or publishing volume. It must increasingly be based on evidence of judgment: what the adviser has operated under, which decisions they have owned, what failed, what changed after implementation, and under which conditions the current recommendation would be wrong.

This connects directly to last week’s post, The Return to Simple Times Is Near. AI is beginning to assume its proper place as one element of a broader enterprise strategy rather than the organizing principle for every technology decision. Growth, resilience, and sustainability still depend on data, infrastructure, security, integration, governance, observability, institutional knowledge, and experienced human judgment. Models, vendors, and pricing structures will continue to change, but the enterprise must retain control of its architecture, proprietary context, operating processes, governance, and ability to learn. Returning to simple times means restoring a disciplined basis for navigating complexity: understand the outcome, know the starting point, assign ownership, preserve flexibility, and build capabilities the organization can govern and operate.

At an inflection point, choosing the people who influence the decision is itself a strategic decision. The wrong guide can shape architecture, capital allocation, talent, governance, and organizational design for the remainder of an increasingly compressed technology cycle. The charlatans will remain visible, prolific, and loud because the market rewards certainty and presence. Executives and board members must apply a different standard based on operating experience, demonstrated judgment, intellectual honesty, and the ability to convert a knowledge avalanche into a decision appropriate for the enterprise before them.

Here are some reading resources:

• Stanford Institute for Human-Centered AI, “The 2026 AI Index Report.” Provides a broad view of AI diffusion, investment, technical development, economics, and social impact.
• BCG, “As AI Investments Surge, CEOs Take the Lead.” Examines the rapid increase in corporate AI investment and the shift of accountability toward CEOs.
• PwC, “Leading Through Uncertainty in the Age of AI.” Reviews the continuing gap between executive expectations and realized financial outcomes.
• Deloitte, “The State of AI in the Enterprise 2026.” Explores the organizational and operating capabilities required to move beyond experimentation.
• Menlo Ventures, “The State of Generative AI in the Enterprise 2025.” Tracks the expansion of enterprise generative AI spending across applications, models, and infrastructure.
• McKinsey, “State of AI Trust in 2026.” Examines weaknesses in governance, risk management, and organizational trust as AI becomes more autonomous.
• Yang Yu and colleagues, “AI Adoption in S&P 500 Firms.” Distinguishes broad claims of adoption from deeper integration into business processes.
• Spyridon Alvanakis Apostolou, Jan Bosch, and Helena Holmström Olsson, “Agentic AI in Industry.” Identifies continuing barriers between agent demonstrations and production deployment.

What’s your take? We are fostering a community of executives who want to discuss these issues in depth. This newsletter is but a part of it. We welcome your feedback and look forward to engaging in these conversations.

If you are interested in exploring the full report, discussing the Board’s offering further, or have any additional questions, please contact me at [email protected], and I will be happy to connect with you.

(*) A normal distillation process produces byproducts: primary, simple ones called foreshots, and secondary, more complex and nuanced ones called aftershots. This newsletter highlights remnants from the distillation process, the “cutting room floor” elements, and shares insights to complement the monthly report.