This report by McKinsey, the global consulting firm, sheds some light on what’s happening in the commercial realm with AI. It’s based on an online survey taken in May and June 2026, with 1,719 participants in 97 nations, spread across a full range of regions, industries, company sizes, and functional specialties. 36% of respondents work for companies with more than $1B USD in annual revenue.
The heat map below, from the report, shows the relation of AI use that has passed beyond R&D into scaling phase by industry type and business function. The entries are the percent of respondents who say their firm is into the scaling phase. The business function rows are ranked by total percent.
What stands out?
- How low are supply chain and manufacturing functions!! You would think there would be a lot of applications in these business functions. But, NO! Maybe it’s just because these functions are typically cost-conscious and slow to choose new technology.
- But in advanced manufacturing, industry use in supply chains and manufacturing is close to the highest at 14%.
- IT, knowledge management and software engineering are high. That’s what one expects when a technology such as AI emerges. We saw the same thing with the coming of the internet in the late 1990s.
- Media is high!! What a surprise! The AI engines train on large quantities of media output. If they just operate as a grandiose search engine, they already add value of a kind we can appreciate.
One of the fascinating tidbits from the report is that chatbots are the most frequently employed form of AI. We guess that already, if we use the internet at all today. Every website seems to have one. And if my experience is any guide, they are mostly useless. They give answers to stupid questions, and can’t handle anything even slightly complicated. The most they do is have a real agent call you, and mostly they don’t do that. They direct you to someplace in the company’s sitemap, which you have probably already read without profit before desperately trying the chat.
I question whether this is the best use of all the energy and other resources consumed searching and re-searching all those tokenized training documents.
We haven’t yet seen the real cost of all these searches, over and over again, of all those training documents. It’s measured in tokens, and currently the tokens are underpriced as a loss leader to get us hooked on using AI instead of the far simpler internet search like Google. We see in daily news how many massive data centers are being built, how much resource they consume (electricity, water, grid capacity). Those costs have to be paid for in a viable business. They have no choice but to charge us for the use.
Just like internet and cell phones and cable TV, and phone service before that, we users will have to pay.
Dan Tinkoff, Lieven van der Veken, Michael Chu, and Tara Balakrishnan, August 2026
