AI Adoption tells Two separate Stories

Author: Michael Spencer from AI Supremacy

Length: • 16 mins

Annotated by Mohamed

Real world lag in AI adoption is worse than BigTech wants you to know. Are Vertical AI apps the solution?

Art Teacher Joyce Hatzidakis poses for a portrait in her classroom on January 22nd, 2026 in Riverside, California. AP: Damian Dovarganes.

Good Morning,

Recently there’s been a slew of new products and models released in the AI world, some of those are on my radar include in no particular order:

Product Radar 📡 in LLMs & Agents

What the AI Adoption!

I wanted to really geek out about AI adoption, or the lack therefore in the Generative AI hype craze.

It’s been three years since ChatGPT launched, and a full seven a half years since the Google paper that started it all. If AI was this transformative technology, wouldn’t you expect more people and workers to be using it more actively?

While workers and even knowledge workers are under a lot of pressure to use AI at work, many of them are not doing so with real conviction or have concerns of the quality, ethics and efficiency of doing so.

I wanted to take a look at several of the recent surveys and reports on AI adoption to try to get a read on the reality in early 2026.

  • The total percentage of employees using AI remains flat, but use varies meaningfully by industry and role type (Gallup Poll).
  • The cited Gallup Workforce survey was conducted this fall of more than 22,000 U.S. workers.
  • I will do my best to cite my sources and display a variety of infographics along with my own notes.

As of November 2025, 12% of U.S. employees use AI daily in their role

Many professionals don’t actually feel comfortable using AI in their professional role however.

Professionals Report Low Comfort in using AI at Work

As of August 2025, just 9% of U.S. employees say they are “very comfortable” using AI in their role.

A Tale of Two Countries (Cities)

AI is now driving a semi-perma K-shaped U.S. economy. A fraction of consumers at the top are driving spending in the economy and the main driver of GDP, the datacenter rollout of capex at all costs benefits mostly a few companies at the top. What does that mean really?

POLL

Are levels of AI adoption in the real world meeting your expectations as compared to the techno-optimist hype?

YesNo

About Half of U.S. Workers say they Never Use AI at Work

Meanwhile, the percentage of total users, those who use AI at work at least a few times a year, was flat in Q4 after sharp increases earlier in the trend. Nearly half of U.S. workers (49%) report that they “never” use AI in their role. - Source.
  • Let me repeat that, nearly half of U.S. workers (49%) report that they “never” use AI in their role.
  • Unless you work as a Product Manager, software engineer, lawyer, translator or a few other professions, or are a recent College graduate, the AI adoption story isn’t impacting you in a huge way.
  • In Q4 of 2025, 38% of employees said their organization has integrated AI technology to improve productivity, efficiency and quality.

The Lag in AI Adoption and Increased Compute Demand is Creating a Mismatch

Global AI computing capacity is technically doubling every 7 months according to Epoch AI.

“(The) Total available computing capacity from AI chips across all major designers has grown by approximately 3.3x per year since 2022 - insights & analysis.

New Bottlenecks to AI Infrastructure and Datacenters Emerging

  • In 2026 Energy (Power, Electricity), a High Bandwidth Memory (HBM) 3D-stacked DRAM technology chip shortage and even Copper are now seen as bottlenecks in AI Infra.
  • Meanwhile the total computing power of the stock of NVIDIA chips is growing by 2.3x.
  • Training compute for frontier language models has been growing by 5x since 2020.
  • The cost in USD of training frontier models has grown by 3.5x per year per year since 2020.

Now according to PwC’s 29th Global CEO Survey, based on responses from 4,454 chief executives across 95 countries and territories. There are some AI adoption stats of note.

The ROI of Generative AI for Companies is not yet There

Most CEOs say their companies aren’t yet seeing a financial return from investments in AI. Although close to a third (30%) report increased revenue from AI in the last 12 months and a quarter (26%) are seeing lower costs, more than half (56%) say they’ve realised neither revenue nor cost benefits.
  • Over half said they didn’t realize more revenue or lower costs.

According to PwC’s survey of CEOs a shocking thing emerges:

Despite $30–40 billion in enterprise investment into GenAI, this report uncovers a surprising result in that 95% of organizations are getting zero return.

Now this might be more of a verdict on ChatGPT Enterprise than anything else, but does show AI diffusion and adoption into companies and workers is much slower than Silicon Valley had us believe.

In Particular ChatGPT and Copilot were Failures

Companies aren’t embracing Generative AI due to low impact. “Most organizations fall on the wrong side of the GenAI Divide, adoption is high, but disruption is low. Seven of nine sectors show little structural change. Enterprises are piloting GenAI tools, but very few reach deployment.”

According to a Brookings survey (November, 2025)1, among the full sample of 1,163 respondents, 57% report using generative AI for at least one personal purpose, most of whom use it for internet searches or web browsing (74%).

AI Adoption seems to increase with Higher Education

Small Businesses are Experimenting with AI

  • About half of small businesses are experimenting with Generative AI. Most of these were in the tech and finance sectors.
  • SMBs that focus on certain kinds of knowledge work are using Gen AI more.

A 2025 report by the U.S. Chamber of Commerce found that 58% of surveyed small business said they used generative AI, rising from 40% in the previous year. The small businesses with the highest use rates were in the technology and financial services sectors (77% and 74%, respectively).

Bring your own AI (BYOAI) is Infiltrating some Small Businesses

Another early 2025 survey conducted by the Initiative for a Competitive Inner City (ICIC) found that 89% of small business owners reported that at least one employee used AI tools.

Approximately 46%of American private-sector employees work for small businesses. Small businesses account for around 43.5% of the U.S. GDP.

In April, 2025 The Pew Center released a survey2 that I covered that also relates to AI adoption.

In 2025, a Third of U.S. Adults had never used a Gen AI Chatbot

Figures around AI adoption among ordinary citizens is remarkably low even years after ChatGPT and Google Gemini became available.

Daily use of Gen AI is Still a Minority

Those without a College degree in America are high unlikely to be daily active users of chatbots of Generative AI tools. This indicates the TAM (total addressable market) of Generative AI is far smaller than boosterism and social media hype would indicate.

A Tale of Two Stories in AI

The investment in AI Infrastructure and the real-world AI adoption appears to be at odds, even as some knowledge workers and newer companies use AI to their advantage. Most experts and analysts claim this simply all this means is we are early in the adoption cycle.

Outside of the AI adoption stories and lagging data for people and workers, Generative AI is not seeing evidence of new categories of jobs, boosts in productivity or ROI from Enterprise, companies or SMBs at any scale. This to me is very puzzling, especially if AI were a real transformative technology. Others even go as far as to say that the current iteration isn’t transformative but what it can become will be (e.g. AI agents). Then they go on to say in x, y or z years from now.

Adoption Varies Widely by Industry

Gallup Survey, January 25th, 2026.
  • Daily active users of Gen AI is highest in Technology, where only one out of 3 use it daily.
  • Generative AI is therefore more of a weekly use tool in most cases, that would skew to relative lower ROI. Even OpenAI report their users in an odd metric called weekly active users. They use this to also inflate their numbers to appear more successful.
  • Sectors like Retail and Social Services show minimal use of AI, but are significant employers of Americans.
  • Generative AI seems to be impactful in Tech, Finance, Law, Academia and some knowledge workers. Though how impactful remains uncertain.

Have we reached AI Adoption Saturation in 2026?

The Gallup poll noted:

“In industries such as technology where AI use has been most prevalent, growth in total users shows signs of leveling, with gains found primarily among those already using AI.”

While Gen AI adoption has room to grow in Finance and Law, its use in Education and Academia is full of controversy, complications and potential risks and damages. The low figures of job creation from the technology with changes to immigration could hurt the U.S. labor market as a lot of skilled workers are set to retire in the next decade.

While Enterprise and companies will adopt Anthropic’s Claude Code and related tools in the later 2020s, it’s more a case that it’s the best thing around, not that it’s going to deliver incredible ROI necessarily. This as Trillions of dollars are being spent in AI Infrastructure alone, and Billions in the funding of AI related startups and LLM makers. The two stories of the investment vs. the adoption are not adding up.

You would have to be a fairly stubborn Technological optimistic to insist otherwise, with the data, surveys and AI reports I’m seeing. A lot of organizations and lobbying is taking place to try to convince us otherwise, but I’m growing more skeptical. Venture Capital having more power in media is not giving us factual data but trying to boost their own investments.

For People the Usefulness of Chatbots is Not Overwhelming

The Pew Research Center, 2025.

Remote Workers Appear to Adopt AI More

  • Forced RTO may increase real human collaboration, but be correlated to less A.I. usage. This relationship is not entirely clear.
  • Yet another example of “Two divergent stories” around AI adoption.

Enterprise Companies Do Pilots that Go Nowhere

In 2024 and 2025, most AI pilots at companies didn’t lead anywhere.

MIT NANDA, July, 2025.
“Our research reveals a steep drop-off between investigations of GenAI adoption tools and pilots and actual implementations, with significant variation between generic and custom solutions.” - NANDA MIT

There’s a chasm for companies between in exploring Gen AI capabilities and implementing and investing in serious projects.

  • Another “tale of two stories in AI Adoption”.

Young People aren’t using AI at work as much as you Think

Brookings
  • The age-group of 30-44 appears to feel the most pressure to use AI at work in order to stay relevant.

CEOs and Executives Report Exaggerated AI Adoption and ROI

CEOs Say AI Is Making Work More Efficient. Employees Tell a Different Story. (WSJ)

C-suite members are usually more enthusiastic about AI than their staff, but that is kind of stunning. 40% of workers say AI doesn’t save them any time at all, while nearly 20% of C-Suite say it saves them more than 12 hours a week

  • Leaders seem to be faking it that AI is a transformative productivity-boosting power. Workers know better. (Think Davos elite vs. real world worker).
  • Employees say AI isn’t saving them much in their daily work so far, while many report feeling overwhelmed by how to incorporate it (forced) into their jobs.
  • This increases the chances that Generative AI is not delivering real ROI, when we see discrepancies like this. (Meanwhile tech optimists will often blame the worker or the organizational structure of the org as being the thing at fault).
  • While AI experts suggest that tangible returns come from enterprise-scale deployment consistent with company (high alignment) business strategy.

Just 12% of CEOs have reduced Costs with AI in 2025

PwC Survey

Most companies haven’t yet realised financial returns from their investment in AI. Another way to think about it, is many companies are also failing to reduce costs with AI.

  • A tale of two stories in AI adoption:
  1. Increasing Revenue
  2. Reducing Costs

Many CEOs are trying to do both with AI, it’s not for the lack of trying.

🛠️ 5 Five Bottlenecks to the U.S. AI Infrastructure Rollout 🚧

Another tale of two stories for AI adoption, is the human vs. supply-chain one. Three major bottlenecks have emerged for datacenters project (as well as two additional more minor ones).

  1. Energy (electricity)
  2. HBM Chips
  3. Copper
  4. Community Protests
  5. Access to water

AI adoption at scale requires an huge upgrade of the U.S. electricity power grid, among other issues.

A Majority of Companies don’t have a sophisticated approach to Innovation, R&D & AI Implementation

Most companies don’t have a ton of capital to devote to things like R&D and using AI to boost efficiency. This is visible in their approach to innovation. - (PwC’s survey of CEOs is super interesting).

  • Even if AI agents hit it out of the park (doubtful), the ability of Enterprise companies to adopt the tech readily might be limited and at best, slow.

Companies are Weary to Integrate AI if they don’t Trust the party in their Trust, Safety and Responsible AI element

  • Less rapid adoption of AI at companies that are more Responsible AI orientated (only normal).
  • ChatGPT Enterprise and related OpenAI tools hit a limit here vs. competitors.

Some cases of AI Adoption are forced by Software Providers

Inspite of the lack lustre numbers of most surveys and reports on AI adoption. Consulting firm McKinsey tries to paint a brighter picture3.

  • When you are forced to use disappointing Copilot features by your company, that tends to happen.
  • As you might know, as of October 2025, Microsoft initiated a “mandatory, automatic installation” of the Microsoft 365 Copilot app on Windows devices, particularly those with Microsoft 365 desktop client apps installed.
  • Other studies, likes this one from Microsoft4, tries to make an optimistic take of investment in AI equating with AI adoption literally cherry picking favorable statistics.
Slide 7.

OpenAI’s Anthropic Problem

Analysts are claiming that Anthropic (has already) won the enterprise market thanks to their narrow focus on coding. This as Microsoft is losing marketshare to tools like Cursor and Claude Code.

As OpenAI’s Codex and agent loop lags more nimble competitors. If OpenAI is losing API and developer marketshare to more narrow competitors, it disrupts the entire ChatGPT adoption story that is fading badly in late 2025 and 2026.

The Information.

Agentic Fads continue to be Manufactured - Motbot Case Study

One of the ways you know real AI adoption is lagging is the rise of somewhat PR (X is no longer a reliable ecosystem) enabled fad-like products. In the context of AI, Moltbot (formerly known as Clawdbot) is a viral, open-source personal AI assistant that has gained massive popularity in late 2025 and early 2026.

Unlike a standard chatbot that just answers questions, Moltbot is an supposedly an autonomous agent designed to actually “do” things on your behalf by interacting with your computer’s files, applications, and the web. Not unlike Minimax has already released with Minimax Desktop Agent, that’s a more pragmatic and working Claude Cowork.

Suffice to say more Manus AI like products are on the way in 2026. Ofen these are based in part on open-weight models like those from Qwen. Agents with persistent memory, proactive Intelligence, full-access system, and a messaging friendly interface are being hailed as breakthroughs by relative unknown actors. In this case Peter Steinberger. When a Billionaire suddenly comes up with a "real Jarvis” that goes “viral” on X, call me skeptical.

Workslop is Undermining the Productivity Boost Narrative

The rise of AI “workslop” has productivity risks and market effects

  • AI “workslop” refers to AI-generated work content that appears useful but lacks substance, is incomplete, or contains inaccuracies. Such content undermines productivity by forcing recipients to interpret, correct, or redo the work (Niederhoffer et al., 2025; Madsen & Puyt, 2025). Workslop may be a key reason why individual productivity gains are not seen at the group or organizational level.
  • In Niederhoffer et al.’s (2025) survey of 1,150 U.S. employees, 40% received workslop in the past month, estimated at 15% of content. Most slop flows between peers (40%), but it also moves upward (18%) and downward (16%) in hierarchies.
  • Workslop is part of the broader generative AI “slop” phenomenon, which is reshaping markets by flooding them with low-cost and low-quality content (Miklian & Hoelscher, 2025; Tullis, 2025; Pendergrass et al., 2025).

Don’t even get me started on AI slop in academic papers.

The 7Vs of AI Slop and Workslop, 2025

Read: Microsoft New Future of Work Report 2025

AI Continues to Infiltrate more Tasks and Departments

This from an already outdated McKinsey survey of Nov, 2025.

  • 2025 was heralded as the year of AI agents, if anything that narrative flopped. At least it built us some basic scaffolding and protocols for the future.
  • If anything, 2025 was the year of Job-redesigns caused by AI
  • We did see some Job impacts on entry level positions, for a variety of reasons most of which have nothing to do with AI.
  • While mass layoffs didn’t occur in 2025, streamlining of job roles for more efficiency did even to pay for the AI pilots (with very mixed results I might add)

Will Generative AI Create Enough New Jobs?

According to the International Monetary Fund (IMF)5, “New Skills and AI Are Reshaping the Future of Work” - Bridging skill gaps for the future.

While new jobs don’t seem to be appearing quickly, new skills are popping up in the post Generative AI era.

“About one in ten job vacancies in advanced economies demands at least one new skill, often appearing first in the United States. The incidence is about half in emerging economies.” - IMF

  • Less immigration in the U.S. might actually increase wages and inflation. Not due to productivity boosts from AI by the way in my opinion.
  • The IMF assumes new jobs are appearing, but does not show evidence of this.
  • Yet another techno-optimistic narrative:
  • Job postings requiring AI-related skills now command a wage premium of roughly 8% to 11% in the U.S. compared to similar roles without those requirements.
  • But it’s not true for everyone. The lack of affordability continues to move faster than wage gains even as the U.S. labor force continues to deteriorate with historically low consumer sentiment levels (lowest levels since 2014, or the last twelve years).
  • If AI were a transformational technology, wouldn’t you expect ordinary people to be a bit more optimistic?

A Tale of Two Stories

According to the IMF:

  1. Countries broadly fall into two categories in the Skill Imbalance Index. Those with high demand for new skills but relatively low supply—like Brazil, Mexico, and Sweden—need to invest in training and ensuring better education in science, technology, engineering, and mathematics. They may also need to outsource or rely on foreign-born workers with skills.
  2. Other countries, such as Australia, Ireland and Poland have abundant talent but more modest demand. Their challenge is to stimulate innovation and help companies absorb available talent. Reforms that foster innovation and the creation of new firms and improve business access to finance would help.
IMF
  • Many of the so-called new AI skills however become redundant quickly, like the “prompt-engineering” of 2024.
  • You can learn about AI agents, and then the next year they are fundamentally something else entirely with new protocols and new capabilities.
  • LLM improvements and software innovations can also lead to unexpected outcomes in skills, job resigns and even entry level career ladders in some specific fields.

Are Scanandevian European Countries Really Well positioned in AI Skills?

According to the IMF, everything is great:

IMF

“Software is Brittle, LLMs and Agents are Fragile in Execution”

In another AI adoption survey in the FAll of 2025 by Wavestone6, their survey of executives claims AI is a “central pillar” in their business strategy inspite of being in the paradox of a technology celebrated in boardrooms yet fragile in execution.

  • Digital sovereignty is now a mainstream concern, with 84% of organizations factoring it into their AI strategies.
  • Only 34% of businesses survey treat AI as a top priority, while others balance it against performance and cost.
  • AI adoption begins with plug-and-play tools embedded in productivity and enterprise software, often accelerated by employee experimentation.
  • AI alignment of workforce is slow. Even in 2025, on average, 30% of target users have (had) meaningfully changed the way they work (with Generative AI tools).

On average, 13% of the IT Budget is allocated to AI of the survey participants.

Skim over the entire PDF.

In 2025, CEOs still seemed over-confident about the capabilities of AI.

Wavestone.
  • Business leaders are financially incentivized to be overconfident about the capabilities of new technologies like AI.
  • Many business leaders and CEOs lack a deeper understanding of what the tech can actually do today.
  • Ironically many of the other CEO surveys I read showed a decrease in confidence in AI systems closer to the end of 2025. (The sample size of this one was small).
  • Wavestone itself is a “digital transformation” consulting firm. It’s their job to hype the technology to make money off of it.

When considering something like the real figures behind AI adoption: BigTech, consulting firms, Venture Capitalist funds and AI creators, are likely among the least trustworthy sources. Even many academic papers are likely funded by special interests groups and many organizations that you’d hope to be objective in reality are little more than lobbying PR firms.

That being said, let’s move on to ICONIQ’s January, 2026 The Execution of AI Era Report7.

The audience here are AI product builders themselves.

Where is AI Adoption Shifting to?

ICONIQ

AI in the Real World?

They call this the Execution era simply because they noticed more C-level executives in AI startups are building more Vertical AI applications. That is, Vertical AI applications are purpose-built to solve problems within a specific industry or niche.

  • The goal here is not hogwash AGI or some scalable AI companion but rather nearly 70% of these companies are building vertical AI applications, reinforcing that durable value is being created through domain specific workflows rather than generalized intelligence.

Changes in Developer Ecosystems

  • OpenAI appears to have lost marketshare to Google Gemini
  • Meta and Mistral appear to have lost marketshare to Qwen, xAI and Databricks

Where the Value at? 💎

ICONIQ
  • Adopting the right model for your startup isn’t just about benchmarks, but real world factors like reliability, cost, security, customization, privacy, latency, competitiveness, explainability and cultural fit.

Enterprise and Startup Considerations for Adoption for Consumer Facing Products

Evaluating Models is more Art than Science

ICONIQ
  • Benchmarks are not useful in real-world adoption.
  • Customer satisfaction and adoption are more important.

The Rise of Synthetic Data Generation

ICONIQ
  • In-house data engineering is now the norm.
  • Cloud-based data processing is the optimal workflow.
  • Synthetic data generation is becoming more popular.
  • Data labeling services are now plentiful and specialized.

Capex and R&D budgets are Skyrocketing

Getting a competitive edge has become more difficult.

ICONIQ

Gross Margins Ramp for AI Products at Scale

ICONIQ
  • Costs shift from talent to inference at scale.

Anyways this concludes our tour in the AI adoption theme. Check out the related Reports for more details.

1Brookings: How are Americans using AI? Evidence from a nationwide survey - Link.

2How the U.S. public and AI experts view AI - Link

3Superagency in the workplace: Empowering people to unlock AI’s full potential (Jan, 2026) - Link

4Microsoft New Future of Work Report 2025 - Link

5IMF: New Skills and AI Are Reshaping the Future of Work - Link

6Wavestone: Global AI survey 2025: The paradox of AI adoption - Link

7State of AI: Bi-Annual Snapshot - Link

What else?

Lately Michael Burry of Cassandra Unchained Newsletter is recommending AI Supremacy, if you come from there my work that is most related is Emerging Technology Investments.

You are reading AI Supremacy, where we explore AI at the intersection of business, society, work, culture and technology.

View in app