taken from the PDF

taken from the PDF

Preamble and a bit of personal story

I recently ended a multi-year consulting engagement. I might publish a reflection on this experience sometime in the future, but for now, let’s just say it was quite mentally draining. I didn’t have any energy left to write public blog posts during my tenure there, as evidenced by the lack of new posts here over the past two years.

This is my first blog post since the consulting engagement ended, and I want to write about something less technical. I became accustomed to writing internal technical materials that often assume some contextual knowledge among readers. The rise of AI in recent years has also been staggering. I am still pondering what I should write in an age when explanatory information is readily available through LLMs. I hope that by writing one piece at a time, I’ll figure it out and will eventually increase my writing output to my preferred level.

This first blog post is about my thoughts after reading the McKinsey & Company report “The State of AI in 2025.” It provides a useful overview of the current adoption of AI in the industry (with some caveats that I’ll mention) and resonates with my experience in recent years.

Key insights of the report

In my view, the results of this survey simply validate (or self-reinforce) some “common-sense” beliefs about AI adoption that have been frequently discussed in the industry:

  1. AI and ML are paradigm-shifting technologies that have been transforming industries since the deep learning revolution around 2015 (Karpathy’s 2017 “Software 2.0” article and his 2025 “Software 3.0” talk provide detailed accounts of this paradigm shift). The most effective way to capitalize on this trend is to redesign applicable workflows to incorporate AI and ML technologies rather than simply augmenting existing workflows.
  2. The company needs to devise a realistic yet ambitious AI roadmap, build technology infrastructure and talent, and rewrite business processes around the technology. Therefore, it is imperative to have sponsorship—or even ownership—from senior leadership for the AI/ML initiatives to succeed (providing the most value to customers and the company).
  3. The new AI/ML technology can be a double-edged sword. Inaccurate results, insecure APIs, intellectual property infringements, the black-box nature of the models, biased predictions, and other issues could all cause trouble for customers and the company. Awareness of these potential problems and proper guardrails against them (e.g., human-in-the-loop processes) are essential for managing the risks associated with adopting AI/ML technology.

The report provided the following data that validate the aforementioned beliefs:

  1. Companies that benefit most from AI use it not only to improve their efficiency (as almost all companies do) but also to achieve growth and innovation.
  2. Among these companies, almost half (48%) show strong senior-leader ownership of, and commitment to, AI initiatives. They are also three times as likely as others (55% vs. 20%) to fundamentally redesign their workflows when deploying AI.
  3. These companies have encountered more negative consequences from using AI, but they also have developed mitigations for a wider range of risks than other companies. A higher proportion of them follow best practices, including human-in-the-loop processes and improved infrastructure, which appear to have contributed to their success in managing these risks.

Key Caveats

The survey nature of this report

This report is based on online survey of 1,993 participents in 105 nations from June 25 to July 29, 2025. This survey is presumably anonmymous, but we should still take the self-reported numbers (e.g., the EBIT impact) with a grain of salt. They can be skewed or inaccurate for various reasons, including the potential bias towards AI usage in participants.

The changing definition of AI

Exhibit 1 from the report

Exhibit 1 from the report

The definition of AI usage in the surveys has become looser and less precise over the years:

  • 2017: “Using AI in a core part of the organization’s business or at scale”
  • 2018-19: “Embedding at least 1 AI capability in business processes or products”
  • 2020-2025: “Adopted AI in at least 1 function” → “Regular use of AI in at least 1 function”

And it gets more confusing with the introduction of the new category “Use of Gen AI (generative AI)” in 2023, where the definition of generative AI is not provided.

It’s not an unreasonable assumption that the increase in AI adoption rates in the reports over the years has been significantly driven by the broader definitions of the term and by the advent of and widespread adoption of large language models (LLMs) such as ChatGPT, Claude, and Gemini.

A marketing department using ChatGPT for copywriting could also be counted as “piloting” AI in the company. Nonetheless, the data still show that more and more companies are willing to claim they use AI in at least one of their business functions.

AI use disclosure

I used AI tools to revise my writing (primarily for grammar and lexical correctness) in this post. However, I wrote most of the content myself; it was not generated with prompts.