The Era of AI Adoption
AI usage has been surging in 2025. Better AI is driving this adoption. Enjoy the trend, vibe everything.
Taking AI from Experimentation to Production Use
In 2025, AI has moved from experimentation to unlocking business value. – IBM CEO Arvind Krishna
The theme of the IBM Think conference held earlier this month was described in an X post by Armand Ruiz of IBM:
The era of AI experimentation is over. It’s time to operationalize AI Agents in the enterprise …
He followed that up with describing AI tools that IBM is announcing and delivering to help get AI adopted in the enterprise, including: pre-built AI agents, custom AI agent builder, agent orchestration, agent observability, and enterprise integrations.
IBM is offering domain-specific agents in an AI agent catalog to help solve specific problems in the enterprise, such as IT, HR, and customer support. For example, they offer a Salesforce agent for customer prospecting, HR agent on slack marketplace.
Despite IBM’s history of AI achievements with Deep Blue and Watson, IBM is currently more a laggard than a leader in the AI technology race. They are offering their own AI models but offer the AI models of the foundation AI model developers, through IBM’s 𝗔𝗜 𝗠𝗼𝗱𝗲𝗹 𝗚𝗮𝘁𝗲𝘄𝗮𝘆:
Watsonx now acts as a control plane for models hosted anywhere such as OpenAI, Anthropic, NIMs, Bedrock and beyond.
Of course, IBM is not alone in chasing the AI market this way. Similar to efforts at Salesforce, Microsoft, and others, they are working to get AI into the enterprise by offering AI agents for domain-specific tasks. Microsoft CEO Satya Nadella argues for this approach by claiming that AI models are becoming commodities and the value is in the AI application layer.
AI adoption in the enterprise is getting real. Here is what we see:
AI usage is surging recently.
AI adoption is driven by utility of AI. Improvements to AI drive more adoption.
AI adoption will accelerate as AI models rapidly improve and more capable AI agents roll out.
Join the trend and use AI for your own needs – vibe everything.
AI Crosses the Chasm
AI adoption has accelerated this year. Survey usage numbers confirm that AI use is surging. One source is the Ramp AI index, which has shown an accelerated increase in usage since January.
What’s going on? It looks like AI has “crossed the chasm” of technology adoption. Not just early adopters, but a majority of businesses and people are adopting AI.
Technology companies are the highest adopters, understandable as AI for coding has been a leading application of AI.
A Snowflake report "The Radical ROI of Gen AI," cites compelling reasons for companies to adopt AI:
92% of these early adopters report positive returns. The majority who quantified their ROI see an average 41% return — a figure that's leading them to increase investment across data infrastructure (81% of early adopters), LLMs (78%), supporting software (83%) and talent (76%).
It’s not often when something has a 92% positive response or such a high ROI:
Generative AI is delivering impressive results across the enterprise, with early adopters reporting "game-changing" or "significant" impacts in percentages consistently above 75%.
Early AI adopters got excited in 2023 and tried it out, then did their AI pilots in 2024. Now in 2025 those positive results from early use are driving many enterprises towards widespread adoption. IBM’s theme of moving from pilot to production and experimenting to adoption rings true.
OpenAI’s Adoption
In current market surveys, OpenAI is garnering the lion’s share of user adoption.
ChatGPT has become synonymous with intelligent AI chatbot to many people, like “Google” and “Xerox” became verbs for near-monopoly products. Sam Alman’s goal is to make OpenAI’s AI product the “Core AI subscription for your life,” as he put it in a recent interview. These trends bear out that OpenAI is clearly the leading AI product.
Anthropic market share is a fraction of OpenAI’s, although Anthropic has leading capabilities. Claude 3.7 Sonnet is widely viewed as the best AI model for coding, powering AI coding assistants such as Cursor. However, Claude has lagged on voice interfaces for their app, which may be why there is low interest in their app in the iOS App Store.
Google Gemini and Anthropic Claude are as capable as OpenAI’s AI models in many ways, which might make OpenAI’s lead surprising, since it is so easy to switch and use multiple AI models. However, OpenAI has parlayed the ChatGPT moment into a strong brand and has maintained their lead with models like o1, o3 and o4-mini. The others lack the brand power that OpenAI has gotten from ChatGPT.
ChatGPT also keeps getting better for day-to-day tasks. With its own embedded search and reasoning, it is a great “Answer engine” to answer any question, from deeply technical to personal advice. The o4-mini is the best AI model out there. Another factor has been the success of the image generation capability in GPT-4o.
AI Utility Is Driving AI Adoption - METR Capability Index
METR, which stands for Model Evaluation & Threat Research, conducts research and tests AI capabilities, specifically evaluating autonomous capabilities of AI systems, i.e., AI agents, to conduct AI R&D. They do this to serve their mission of assessing AI catastrophic risks.
As shared in their paper, Measuring AI Ability to Complete Long Tasks, METR has tried to find a benchmark that better reflects real-world AI performance. Doing super-human on LSAT or medical license tests doesn’t mean an AI model can practice law or medicine. METR developed a different metric:
To quantify the capabilities of AI systems in terms of human capabilities, we propose a new metric: 50%-task-completion time horizon.
What is startling is that AI systems are improving on METR’s metric at an exponential rate, doubling the length of task they can do every 7 months.
AI Utility Drives AI Adoption – A Personal View
As the METR metric shows, AI has crossed the capability threshold for more tasks. As it grows in capability, it opens up more use cases, driving more AI adoption.
I’ve observed this personally with Deep Research. I found it such a time-saver that it became my go-to advisor and assistant for many types of questions. Utility drove adoption.
I just spent several days doing a research deep-dive on a non-AI topic. I used ChatGPT’s o4-mini as a research assistant, asking o4-mini highly technical questions then getting answers with citations and sources in under a minute. Google’s Gemini-powered AI views in search helped as well.
I called upon o4-mini and other AI tools dozens of times in the last 3 days, far more than I’ve done in the past. . I went through multiple iterations on ideas, answered technical questions in minutes that would have taken hours, accumulated and utilized dozens of relevant scientific papers, and wrote up 37 pages of notes.
Comparing this research process to the past makes it clear this accelerated in my research:
1980s (pre-internet): Go to library, read academic sources to help understand a question. Write findings by hand in notebooks. Learning cycle – 1 day.
2000s (internet-based research): Search on internet, find online papers to answer a question. Write findings in computer documents. Learning cycle – 1 hour.
2020s (AI-supported research): Pose question as a query for AI model, AI-enabled search and/or ‘deep research’ tool. Read the response and incorporate findings into online learning documents. Learning cycle – 5 minutes.
Most of us have found AI useful far beyond ‘deep research’ in personal tasks. One very different use case is sending a photo to ChatGPT or Gemini AI app on your smartphone to identify an animal, help with a home repair, or find a replacement part. Like with internet search or a smartphone, once you use it, you can never go back.
Conclusion - Vibe Everything
Not only is AI usage rising, but there has been an acceleration in AI adoption since early 2025, after Gemini 2.0, o3-mini, and DeepSeek R1 came out. More capable AI agents and improved AI coding assistants have come out that have driven more usage.
AI has gotten good enough to do real things, and so adoption is surging as real people use it more for real-world applications.
Vibe coding became a trend because AI got good enough to complete significant software tasks and projects with zero human code writing. Instead of coding, send the problem to the AI model and see how far it can go. You’ll find that AI can go pretty far.
AI has gotten really good at many things beyond coding. It can research, write, and think better. With those three skills, it can assist with a surprisingly wide set of tasks, making it more personally useful.
One final piece of advice to take advantage of AI’s recently improved capabilities: Vibe everything. For any task you have, let AI take a crack at it.
Are you writing a report or making a business or marketing plan? Let AI make an outline and draft it. Job-hunting? Let AI search and find employer prospects, critique and redo your resume, draft cover letter emails, and more. AI can handle many technical, business, medical, and personal questions and tasks well, up to a certain point of complexity.
Don’t just play with AI, get value out of it. Go as far as you can with AI.
The surge in AI adoption is a sign has crossed the chasm, to the point where now most people are using AI. AI itself is still improving rapidly, which will drive further increased adoption.
5 minutes seems pretty quick for AI-supported research. How are you checking the accuracy of the answers your AI tool gives you? Personally, I've found that accuracy-checking can be a real bottleneck.