Three years ago, with the advent of ChatGPT, Gemini, and Claude, AI for everyone landed with a bang.
Driven by word of mouth, industry, and media, AI literacy exploded as consumers and enterprises learned new tools, models, and prompting techniques. Most professionals made the climb from the plains of the pre – AI era. Now, ChatGPT has crossed 900 million weekly active users, Claude became the #1 US App Store app with one million new users added daily, and AI Overviews handle 1.5 billion searches monthly. At work, enterprise AI deployment has surged to 78% of organisations (Source: Gartner Enterprise AI Adoption Analysis, 2026),and 60% of workers now have access to these tools (Source: Deloitte State of AI in the Enterprise, 2026).
While AI literacy has expanded universally, it has not benefitted everyone uniformly. For most professionals, after the novelty and initial gains, the AI benefit has plateaued. It has not made work easier or lighter.
Limited contextual understanding, session inconsistency, and unreliable outputs mean fixing AI often becomes work itself. The result is intensified effort, validation burden, and correction cycles.
The climb from 2022 is real. But because of the above, there is a sense of a plateau, where the benefit stalls relative to the human input and correction effort required to use it.
And yet, there are a few who use AI exponentially. Kevin Roose observed it from the ground: “People in SF are putting multi-agent Claude swarms in charge of their lives.” As Noah Corduroy puts it plainly, AI polarises the extremes and the rewards are disproportionate for those who move through literacy toward something deeper. Staying put on the plateau seems a missed bus of emergent opportunities,
Continuing the climb from AI literacy to AI fluency requires a new portfolio-based approach to AI augmented work, one designed to reduce the stall and can also genuinely energise us.
In this 69th edition of Mindvista, we explore the structural causes of the plateau, introduce a 2×2 portfolio for AI augmentation and propose actions that one can take to move from literacy to fluency, both in and outside of work.
The AI Literacy Plateau: Cognitive Debt and the Reality of Exhaustion
The plateau looks different depending on whether you are observing it or living in it.
From the outside, there is a constant stream of new models, application wrappers, Agentic capabilities and headlines about new benchmarks, agents and orchestration and productivity transformation.
For most on the inside, there is a more complex reality. I have been publishing Mindvista for over 18 months , using multiple LLMs for tracking, research and advisory work and publishing artefacts. The intensity of work in research and writing each edition has more than doubled and I have a lived experience of intense use and frustration with AI also. A lot of it has to do with the state of technology and LLM vendors and have written about this also two months ago. (https://mindvista.co/ariavista-my-ai-ally-for-exploration-everyones-using-ai-but-few-can-rely-on-it-what-ai-needs-to-do-to-step-up-the-olympian-way/)
In the last few weeks, new research has emerged that deeply supports this lived reality.
Harvard Business Review reports that AI is often intensifying work rather than reducing it, causing a 14% increase in mental fog and up to a 33% spike in decision fatigue among surveyed knowledge workers (Source: HBR, “AI Doesn’t Reduce Work — It Intensifies It,” February 2026.)
The Anthropic Economic Index quantified the reliability dimension. Once you account for the tasks where AI attempts and fails, not just where it succeeds reduces, the annual labour productivity gains falls from 1.8% to 1.0%. (Source: Anthropic Economic Index, January 2026.)
Another striking data point came from a randomised controlled trial using AI in software development . METR studied experienced open-source developers using AI tools on tasks they knew well. Developers using AI tools took 19% longer to complete tasks, while they had expected a 24% speedup. And ironically, they still believed, after completing the tasks, that AI had made them 20% faster. (Source: METR, Early 2025 AI Experienced OS Dev Study, July 2025.)
The gap between perceived and actual performance is the defining feature of the plateau. You feel like you are moving. The data says otherwise especially if you consider the output /input ratio.
Beyond formal productivity metrics, Simon Willison also identified underlying erosion of understanding and judgement from use of AI, calling it cognitive debt. He notes ” Even if AI agents produce code that could be easy to understand, the humans involved may have simply lost the plot and may not understand what the program is supposed to do, how their intentions were implemented, or how to possibly change it.” (Source: simonwillison.net, “Cognitive Debt,” February 2026.)
An Anthropic study shows the cognitive debt that using AI assistance led to a statistically significant decrease in mastery. On a quiz that covered concepts they’d used just a few minutes before, participants in the AI group scored 17% lower than those who coded by hand, or the equivalent of nearly two letter grades. Using AI sped up the task slightly, but this didn’t reach the threshold of statistical significance. This supports Simon Willisons argument on cognitive debt.
The plateau and cognitive debt is clearly real and this is compounded by enterprise mandate traps that makes the climb harder for the experienced.
Expertise Penalty and the Mandate Trap
The altitude sickness hits hardest where expertise is highest, and this is the finding that most enterprise AI programmes have not yet absorbed.
A study in the Quarterly Journal of Economics examined AI impact across skill levels in professional settings. Less experienced workers gained 30–36% in speed and quality while Highly skilled workers only saw marginal speed gains and small quality declines. (Source: Quarterly Journal of Economics, “Generative AI at Work,” 2025.) AI substitutes for the lower order components of expert work. It struggles with the higher order judgment, the accumulated pattern recognition, the contextual instinct and knowing when not to act.
The HBS and BCG jagged frontier study added an organisational dimension. GPT-4 improved average output quality across a team, and simultaneously reduced the diversity and variability of ideas generated. (Source: HBS, “Navigating the Jagged Technological Frontier,” 2024.)
Corporate mandates can create a counter effect of diminishing productivity and ability. Organisations like Accenture have tied staff promotions to the tracked usage of its AI tools, for senior managers and directors. If this does not take into account the AI burden and cognitive debt then it could become a game of compliance metrics rather than build genuine new capability in employees. (Source: The Guardian, “Accenture Links Staff Promotions to Use of AI Tools,” February 2026.)
Mandated AI adoption without fluency development produces better average work and less original work. For enterprises whose competitive advantage lies in original thinking, this is a structural risk dressed as a productivity gain.
Increasing AI usage for continued gains and progressing from literacy to fluency needs newer ways to structure, assess and adopt a portfolio than a task based approach, that works for individuals both in and outside of work.
A 2x2 Portfolio based approach for AI Fluency
To navigate rapidly evolving AI capabilities, a simple 2×2 portfolio framework helps structure where and how AI adds value.This allows us to capture incremental gains while exploring the art of the possible in our AI adoption along two axes
Axis One: Context Need for AI as a spectrum from light to heavy to act reliably and consistently
Axis Two: Nature of Task as a spectrum from repetitive execution to building new.
The four quadrants can be assessed as below:
Quadrant 1:Repetitive, Context Light.This is where enterprise AI delivers measurable ROI today. Analytics pipelines, data formatting, scheduling, standard reporting, and structured document processing. Agentic systems are beginning to own this quadrant at scale. Ai led Automation works here.
Quadrant 2: Creative, Context Light. The current LLM sweet spot. Ideation, brainstorming, short-form content, and discrete creative tasks where the human reviews and iterates quickly. The collaboration is productive because the stakes of any single error are contained and correction loops are fast. Most individual professionals begin their AI journey here, and many stay here.
Quadrant 1 & 2:(Light Context / Repetitive & Creative) is where AI thrives today. Adobe data shows that just 5% of tasks—primarily standard operational and dynamic problem solving workflows drive a staggering 59% of AI use.
Quadrant 3: Creative, Context Heavy. This is where most senior knowledge work actually lives with long- form reasoning, strategic analysis, editorial judgment, research synthesis, advisory work. This is also where LLMs fail most. An EMNLP study demonstrated that even when models retrieve information correctly, performance degrades by 13.9% to 85% as input context length increases. (Source: EMNLP, “Long-Context Failures,” 2025.) This is a frustration zone.
Quadrant 4:Repetitive, Context Heavy. The unsolved gap. This quadrant represents complex workflows, institutional memory, traceability, consistency and reliability for operational compliance. Success here is not driven by native chat models. It is solved by what we defined as the new enterprise architecture: combining, Systems of Execution (the agents) with the Fabric (the connective mesh of Context, Control, and Observability). Reference Mindvista 66th Edition ( https://mindvista.co/software-ate-the-world-ai-isnt-devouring-it-its-sharing-the-table-the-high-stakes-fight-for-enterprise-power-and-value/)
Building AI Fluency does not mean just forcing your way through the heavy quadrants alone. It means adopting a portfolio mindset for finding ways to use, test, and explore AI across all quadrants as possible, both in and outside of work.
The Climb: Building AI Fluency at Work and in Life circa 2026
While LLM challenges and incremental gains remain, there are also new advancements in automation with Agentic tools like Claude Cowork, Perplexity computer, Chatgpt Codex. Enterprises are also engaged in rearchitecting for agency, fabric and record and reimagining processes for AI led automation.
AI fluency can be built step by step both at work and personal use cases for building new.
1.Exploring what use cases can help in context light and repetitive work. For example I have semi automated the process of scanning for signals from 100+ sources on email and X in bi weekly digests across work, life and society and civilisation . For me this has given me a time efficient, knowledge uplift for Mindvista and gives a cutting edge in my strategic advisory work to AI tech companies.
2. Evaluating new capabilities in context light and creative work to build and experiment for coding, content creation (presentations, audio/video generation) for reach and impact.
3.Reduce fatigue with both repetitive and creative context heavy work by actively leading and engaging in AI projects at work and explore side projects and personal use cases using newer capabilities. For example, I am exploring more context capture and personalisation (using Claude skills) and building automated analytics for LinkedIn, website, X and subscriber and follower conversations, using Claude Cowork.
There are many hands on leaders and experts who are doing even more. Andrej Karpathy documented his own shift in real time and in weeks, going from 80% manual and 20% agentic to the reverse. (Source: @karpathy, January 2026.) Satya Nadella runs what he calls a “Council of AIs,” orchestrating multiple models simultaneously to debate, cross-validate reasoning, and challenge his own thinking as a personal project. (Source: Analytics India Magazine / LinkedIn, 2024.)
The practical experience gained by problem solving, working around technology limitations is an invaluable skill and learning when you observe, do, learn and share.
Enterprise Mandates for AI Adoption to AI Fluent Employees
For organisational leaders, the plateau on AI literacy is not only a metric problem. It is a talent architecture problem, and the two are connected in ways that standard AI deployment frameworks have not yet addressed.
McKinsey’s State of Organisations 2025 found that companies framing AI adoption as workforce capability development consistently outperform those framing it as technology deployment. But capability development is not about training or mandates alone.
It needs to recognise the context of employee work and augment work to create path ways to enable each to move along the four quadrants. It may also mean bringing together deep domain expertise for employees with AI technologies.
HR and Business leaders can develop a AI fluency quotient along the quadrants and let the employees chart what they need to at work and give them the time and space to explore AI personally as well.
For an accelerated path to senior leadership, I explored the Igbo apprenticeship model as a framework for Apprenticeship for AI Allyship by pairing a senior executive or domain expert who holds institutional context, judgment, and the weight of Quadrant 3 and 4 work, with a digitally native professional who holds technical AI fluency and the experimental agility of someone building in Quadrants 1 and 2. The exchange is mutual and necessary. The senior brings the “why.” The junior brings the “how.” Together they cover the full map that neither can navigate alone. (Source: MindVista Edition 16, “Apprenticeship for AI Allyship,” 2024- https://mindvista.co/apprenticeship-for-ai-allyship-a-new-model-integrating-multigenerational-talent-for-a-human-centric-ai-driven-enterprise/)
The Apprenticeship also helps solve a talent scarcity in future. This threat is already visible in the macroeconomic data: Stanford research using ADP payroll data reveals a 16% employment decline for young workers aged 22 to 25 in AI-exposed fields, signalling a creeping structural risk to the enterprise talent pipeline
It may be a good idea for enterprises mandating AI and pursuing build a AI fluent workforce to evaluate and pilot this model.
For AI native talent, the pathway could be different to lead them to be engaged on AI-led initiatives to drive business outcomes.
Personal Motivation: Edmund Hillary the Himalayan Climb and the Stop
At the end of the day being AI literate is table stakes and AI fluent a winning hand that impacts everyone.
It comes down to human personal quest.
Edmund Hillary when asked as to why he climbed Everest. His answer was simple and yet profound : “I did it because it is there.”
Getting to the plateau from the plains was not easy and consumers and enterprises deserve credit for making the climb. But the complexities of context, challenges of the technology, the busyness and stress at work all compound.
Progressing from AI literacy to fluency needs the same persistence and curiosity but also new methods which has been the attempt of the 69th edition acknowledging the truth and the portfolio approach to build and explore the art of possible with AI.
There is one point of difference with Hillary’s Himalayan climb. Unlike its relentlessness, in this journey there are also things that you do without AI.
Jacob Collier coined “Creative Infinity Syndrome” for his own condition — boundless generation that refuses to stop and consequently produces paralysis rather than product. In an age of infinite AI-assisted output, the premium human skill is not generating more. It is the judgment to stop, evaluate, and decide what is actually worth keeping. (Source: @SridharRavi90 / Jacob Collier, February 2026.)
Fluency includes restraint. In 2026, with agents that can run autonomously and generate at scale, restraint may be the most undervalued capability of all.
AI Fluency is not about becoming more like the machine. It is about cultivating the judgment, restraint, and contextual wisdom the machine cannot possess. That is the humanness that powers the next ascent.
- How do we cure the “Brain Fry” and cognitive debt that accumulate from managing imperfect AI?
- What steps are you actively taking to build AI fluency across different workflows for yourself and your teams?
- Can an “AI Allyship” pairing senior experts with tech-fluent juniors rescue the enterprise talent pipeline?
- What are you deliberately doing without AI to apply restraint and avoid “Creative Infinity Syndrome”?
- How are you balancing personal AI experiments with work mandates?
Love to hear your comments, thoughts and ideas. Explore and stay tuned for more!
Best wishes,
External Sources & Key Takeaways
Great Read