I started writing this article at 3:47 AM on a Tuesday, not because I'm a masochist, but because I've been staring at a screen for six hours trying to remember how I used to think before the machines started thinking for me.
I'm 45. I remember when "googling" something meant scanning ten blue links, synthesizing conflicting information, and arriving at your own conclusion.
I remember when writing code meant understanding every line, not describing intent to an autocomplete oracle and praying the output compiled. I remember when "research" was a verb that required effort.
That version of me is dying. And I'm not sure anyone's mourning him.
What follows isn't another breathless list of AI trends. You've seen those. "2026: The Year of Agentic AI!" "Quantum Breakthroughs Ahead!" "Super Agents Will Change Everything!"
The tech press has become a parody of itself, churning out identical predictions with the desperate energy of a casino croupier convincing you the next spin will be different.
No. What I want to explore is something darker, more intimate, and—if I'm honest—more terrifying than any robot uprising narrative. I want to talk about the Great Forgetting: the systematic erosion of human cognitive capability that accompanies every wave of AI "augmentation."
And I want to propose something heretical: that the most important technology story of 2026 isn't what AI can do, but what it's undoing in us.
Part I: The Numbers Don't Lie—But They Don't Tell the Truth Either
Let's start with the data, because I promised you depth, and depth requires evidence.
The Explosion That Hid a Collapse
In 2025, GitHub recorded 1 billion commits—a 25% year-over-year increase. Developers merged 43 million pull requests monthly, up 23%.
On paper, software development has never been more productive. Mario Rodriguez, GitHub's chief product officer, calls this the dawn of "repository intelligence"—AI that understands not just code but the relationships and history behind it.
But here's what nobody at Microsoft will say in their press releases: those 1 billion commits are increasingly written by machines, reviewed by machines, and merged by machines. The human isn't being amplified. The human is being bypassed.
GitHub Copilot, which now writes an estimated 35-40% of code in files where it's enabled, didn't make developers 35-40% better. It made them 35-40% less necessary for the mechanical act of coding. And coding, for all its mystique, was never just mechanical.
It was a form of structured reasoning, of translating ambiguous human desire into unambiguous machine instruction. Every line written was a small act of clarity. Every bug hunted was a lesson in humility.
Now? We "vibe code." We describe intent and validate outputs. IBM's Ismael Faro calls this evolution from "vibe coding to Objective-Validation Protocol"—a sterile term for what is essentially the reduction of human craft to quality assurance.
The Workforce Displacement Nobody's Measuring
Goldman Sachs predicts AI could replace 300 million full-time jobs. The University of North Dakota's ethics research cites this figure prominently. But here's the metric that keeps me awake: we have no measurement for the jobs that aren't replaced but are rendered cognitively vacant.
Consider the knowledge worker who keeps their job because AI "augments" them. They use ChatGPT to draft emails. They use Perplexity to research competitors.
They use Copilot to analyze spreadsheets. They are employed. They are productive. But they are also experiencing something unprecedented in human history: the atrophy of professional judgment without the corresponding signal of unemployment.
A factory worker replaced by a robot knows they're displaced. A lawyer who uses AI to draft contracts but still "reviews" them? They believe they're working.
But after three years of this, could they draft a complex contract from scratch? Could they spot the subtle error that the AI consistently misses because it wasn't in the training data?
We're creating a cognitive precariat: millions of people who appear employed but whose core skills have been hollowed out by dependency. No economist has modeled this. No government tracks it. But I see it in my peers—in the way their eyes glaze over when you ask them to solve a problem without a chatbot open. In the panic that sets in when the internet is down.
The Trust Paradox
A global study found that 73% of consumers trust content produced by generative AI—despite notably low awareness of risks like misuse. This isn't confidence. This is cognitive surrender.
We've outsourced verification itself. When I ask Perplexity a question (and I do, constantly—I'm not immune), it returns a confident answer with source citations. I rarely click those citations. You rarely click them.
The citation has become a ritual of legitimacy rather than an actual gateway to verification. It's the digital equivalent of a doctor's white coat—an authority signal that bypasses critical evaluation.
Microsoft's Aparna Chennapragada says "the future isn't about replacing humans, it's about amplifying them." But amplification assumes the signal being amplified is human.
When the AI generates the draft, suggests the edits, and polishes the final output, what exactly is being amplified? The human's capacity to press 'accept'?
Part II: The Architecture of Forgetting—How Systems Design Obliterates Memory
To understand the Great Forgetting, you need to understand how 2026's AI systems are architected. This is where I get technical, because the devil isn't in the headlines. The devil is in the inference stack.
From Models to Systems: The Black Box of Black Boxes
IBM's Gabe Goodhart notes that "in 2026, the competition won't be on the AI models, but on the systems." He's right, and that's the problem. When you interact with ChatGPT, you're not talking to GPT-4.
You're talking to a software system that includes web search tools, code interpreters, memory stores, and agentic loops—each itself a black box, orchestrated in ways even its builders don't fully comprehend.
This matters because opacity compounds. A single model's errors can be studied, benchmarked, eventually understood. A system of models, tools, and routing logic? Its failures are emergent, distributed, and often invisible until they cause harm.
When a "super agent" (IBM's Chris Hay's term) operates across your browser, editor, and inbox, completing tasks you never fully specified, the chain of causality between your intent and the outcome becomes unrecoverable.
You didn't forget how to do the task. The system made remembering irrelevant—and then made the memory itself inaccessible by replacing your workflow with an opaque automation.
The Death of Struggle
Peter Staar, a principal researcher at IBM Zurich, predicts that "robotics and physical AI are definitely going to pick up" as the industry hits "diminishing returns from scaling" language models. But there's another diminishing return nobody discusses: the return on cognitive struggle.
Neuroscience is clear that deep learning—the human kind, not the machine kind—requires productive difficulty.
The struggle to recall information, to navigate ambiguity, to synthesize conflicting sources: this struggle is the mechanism of memory consolidation and skill acquisition. It's why you remember the book you fought through more vividly than the one you skimmed.
AI systems are optimized to eliminate this struggle. Perplexity "felt like a helpful AI sales engineer, asking clarifying questions"—but those clarifying questions narrow your exploration before you've done the messy work of understanding the problem space.
Microsoft Copilot offers "exceptionally readable formatting" and "great use of tables for comparisons"—but readability is the enemy of complexity retention. The more legible the AI makes information, the less your brain is forced to process it.
I'm not being nostalgic for hardship. I'm pointing out a design feature with catastrophic cognitive externalities. Every AI product manager in 2026 is measured on "time to task completion" and "user satisfaction."
Nobody is measured on "user retention of domain knowledge six months later." The metrics that drive AI development are perfectly aligned with human cognitive atrophy.
The Quantum Mirage
Let me address the elephant in the room: quantum computing. Microsoft and IBM both claim 2026 is the year quantum achieves "advantage" over classical computers. Jason Zander at Microsoft says we're entering a "years, not decades" era. IBM's Jamie Garcia says they've "moved past theory."
I don't doubt the physics. But I want you to notice the narrative function of these announcements. Quantum breakthroughs serve as a displacement fantasy—a technological sublime that distracts from the mundane tragedy of human deskilling.
While we marvel at topological qubits and hybrid supercomputing architectures, we ignore that the average office worker can no longer write a coherent memo without AI assistance.
The quantum computer will solve problems classical computers can't. But the classical human, augmented by classical AI, is already failing to solve problems that unaugmented humans managed fifty years ago. We're building gods while forgetting how to pray.
Part III: The Invisible Curriculum—What AI Is Teaching Us Without Our Consent
Every technology has a hidden curriculum. The smartphone taught us fragmented attention. Social media taught us performative identity. AI is teaching us something more fundamental: learned helplessness dressed as empowerment.
The Democratization Trap
Kevin Chung, chief strategy officer at Writer, predicts that 2026 will see "the democratization of AI agent creation" as "the ability to design and deploy intelligent agents is moving beyond developers into the hands of everyday business users." This is framed as liberation. I read it as the final stage of proletarianization.
When "everyday business users" build agents, they're not becoming programmers. They're becoming configuration managers for systems they don't understand. The "lowering of technical barriers" doesn't raise the user's technical capacity. It removes the barrier between them and a dependency they can't escape.
Democratization without comprehension is just dispersal of vulnerability. A developer who understands agent architecture can debug, extend, and eventually replace the system.
A business user who "built" an agent by describing it in natural language has no such capability. When the agent fails, they're helpless. When the vendor changes pricing or terms, they're captive. When the system makes an error with legal or ethical consequences, they're accountable without being knowledgeable.
This is the hidden curriculum of "no-code" AI: you are free to create, but not free to understand.
The Sovereignty Paradox
Capgemini's 2026 trends report identifies "tech sovereignty" as a strategic priority—but notes the "borderless paradox" that full autonomy is unrealistic. IBM's Anthony Marshall reports that 93% of executives consider AI sovereignty a must for 2026. Half worry about over-dependence on compute resources in certain regions.
But sovereignty over infrastructure is meaningless without sovereignty over capability. What good is a sovereign cloud if your workforce can't operate without American AI models? What value is regional data residency if your decision-makers can't think without Chinese reasoning engines?
The sovereignty conversation is technically sophisticated and cognitively naive. Nations are building digital borders while their populations become intellectual dependencies of foreign corporations.
Australia, as I explored in my earlier analysis, exemplifies this: $654.3 million for Digital ID systems, but only $89.3 million for broad cyber resilience. We're fortifying the castle while the villagers forget how to farm.
The Ethics of Learned Helplessness
The University of North Dakota's AI ethics framework emphasizes fairness, transparency, accountability, privacy, and safety. These are worthy principles. But they share a common blind spot: they treat AI as a tool used by humans, rather than a system that reshapes humans through prolonged use.
The "common ethical issues" listed—job displacement, autonomous weapons, intellectual property, surveillance—are all external harms.
What about the internal harm? What about the erosion of human agency that occurs when judgment is consistently delegated? What about the "soft" violence of making entire populations cognitively dependent?
There's no category for this in the EU AI Act. No UNESCO guideline addresses it. The OECD AI Principles don't mention cognitive atrophy.
We've built an ethics framework that protects humans from AI's mistakes while ignoring AI's most successful function: making humans mistake-prone without realizing it.
Part IV: The Resistance—Fragments of a Counter-Movement
I promised you new insights, not just despair. So let me tell you about the resistance. It's smaller than the hype machine, but it's real, and it contains the seeds of something genuinely different.
The "Repository Intelligence" Reversal
Remember GitHub's "repository intelligence"? The AI that understands code context, not just syntax? Here's the reversal: some developers are using AI to surface complexity, not eliminate it.
A small but growing movement—call them "complexity preservationists"—are configuring AI tools to expose the full dependency graph, to highlight edge cases, to deliberately surface the messy history that "repository intelligence" is designed to smooth over. They're using AI as a pedagogical instrument rather than a productivity tool.
This is subtle but crucial. The same technology that deskills can, with intentional design, re-skill. But this requires rejecting the dominant metric of "time saved" in favor of "understanding gained."
No VC-funded startup optimizes for this. No Fortune 500 CIO is measured on it. It exists only at the margins, in open-source communities and academic labs.
The Return of Friction
IBM's Kaoutar El Maghraoui notes that "2026 will be the year of frontier versus efficient model classes"—that "we can't keep scaling compute, so the industry must scale efficiency instead." This hardware constraint is creating unexpected opportunities.
Edge AI—models running on modest accelerators rather than cloud supercomputers—introduces friction by necessity. You can't have infinite context. You can't query the entire internet. You must work with limited, curated knowledge. This constraint, born of energy and cost limits, accidentally preserves human cognitive engagement.
When your AI assistant can't instantly retrieve and synthesize everything, you must choose what to feed it. You must prioritize. You must remember what's important enough to include in the context window. The technical limitation becomes a cognitive feature.
The Quantum of Human Judgment
Let me return to quantum computing, but from a different angle. IBM and Microsoft are building "quantum-centric supercomputing" that combines quantum, classical, and AI processing. The human role in these systems isn't eliminated—it's concentrated at the interfaces between paradigms.
Quantum algorithms require entirely different mental models. They exploit superposition and entanglement in ways that defy classical intuition.
The humans who can operate at this interface—translating between quantum possibility and classical necessity—are not deskilled. They are hyper-skilled in a new dimension.
That actually suggests a pattern: technological revolutions that genuinely expand human capability do so by introducing irreducible complexity, not by hiding it.
The danger isn't quantum computing. The danger is that we'll use quantum systems through AI interfaces so seamless that the quantum nature becomes invisible—another black box among black boxes.
The preservation of human judgment requires deliberate exposure to irreducible complexity. This is the fight: not against AI, but against the seamlessness that makes AI invisible.
Part V: What Comes After—Scenarios for 2030
I've painted a dark picture. Let me offer three scenarios, grounded in the data but extrapolated with intention. These aren't predictions. They're provocations.
Scenario A: The Managed Decline (Probability: 60%)
In this scenario, the Great Forgetting continues, managed but not reversed. Governments introduce "AI literacy" programs that teach citizens to use AI tools effectively—reinforcing dependency under the guise of empowerment.
The cognitive precariat grows to encompass most knowledge workers. A small elite maintains deep technical skills, operating the systems that everyone else depends on.
By 2030, "critical thinking" has become a niche skill, like calligraphy or blacksmithing. Universities offer it as a "wellness" elective. The economy functions—AI handles the complexity, humans handle the exceptions. But the exceptions are increasingly incomprehensible to those who must handle them.
This is the path of least resistance. It requires no conspiracy, only the accumulation of small optimizations that individually make sense and collectively destroy capacity.
Scenario B: The Great Reckoning (Probability: 25%)
A catastrophic AI failure—financial, medical, or infrastructural—kills thousands and exposes the fragility of human-AI systems where humans no longer understand the tools they depend on.
The aftermath resembles the post-2008 financial crisis, but deeper, because the "toxic assets" are human capabilities that have atrophied beyond recovery.
In the reckoning, nations institute "cognitive reserve" requirements—mandatory periods of AI-free operation in critical sectors. Education systems are restructured around "struggle-based learning." The EU AI Act is amended to include "human capability preservation" as a core principle.
This is painful but potentially regenerative. Like a forest fire that clears deadwood for new growth. But the cost is measured in lives lost and trust destroyed.
Scenario C: The Intentional Renaissance (Probability: 15%)
A coalition of educators, technologists, and policymakers recognizes the Great Forgetting early enough to intervene. They don't reject AI—they redesign it around human development rather than human replacement.
Key elements:
- Cognitive impact assessmentsrequired for AI deployments, measuring not just efficiency but skill retention
- "Friction by design"standards that prevent interfaces from being- tooseamless
- "AI sabbaticals"—mandatory periods where professionals operate without AI assistance to maintain baseline capabilities
- Education systemsthat use AI to identify knowledge gaps- in order to target human instruction, not to replace it
By 2030, this coalition has demonstrated that AI-augmented humans can outperform both unaugmented humans and fully automated systems—but only when augmentation is designed to expand, not replace, human capability.
This is the scenario I work toward. It has the lowest probability because it requires fighting every incentive structure in the tech industry. But it has the highest potential payoff.
Part VI: A Personal Coda—The Writer's Dilemma
I told you I started this article at 3:47 AM. Here's the rest of that story.
At 2:00 AM, I had an outline and a blinking cursor. At 2:15, I opened ChatGPT and asked it to "help me think through the structure." At 2:47, I realized it had written three paragraphs that I was about to paste into my draft—paragraphs that were competent, well-structured, and not mine.
I closed the tab. I paced. I made coffee I didn't want. And I returned to the cursor, forcing myself to find my own words for ideas that the AI could have expressed more smoothly.
This article is worse than it would have been with AI assistance. It's messier. The transitions are bumpier. Some sentences are inelegant. But every inelegant sentence is evidence that a human struggled with an idea and emerged with something imperfect but owned.
This is my practice of resistance. Not rejection of AI—I used it to research, to verify facts, to check my memory of GitHub's commit statistics. But rejection of the seamless, the smooth, the frictionless. Rejection of the voice that isn't mine but sounds better than mine.
The Great Forgetting isn't inevitable. It's a design choice, repeated billions of times, by billions of users, in billions of moments of "I'll just use AI for this one thing."
The "one thing" becomes everything. The exception becomes the rule. The tool becomes the mind.
I'm asking you—not as a reader, but as a fellow human navigating the same currents—to notice the next time you reach for AI without thinking. To pause. To ask: What am I preserving by doing this myself? What am I losing by delegating it?
The answer isn't always "do it yourself." Sometimes AI genuinely expands what's possible. But the question must be asked. The unexamined delegation is the mechanism of forgetting.
Appendices: The Data Beneath the Narrative
For those who want to verify, challenge, or extend my analysis, here are the key sources and statistics referenced:
| Claim | Source | Date |
|---|---|---|
| 1 billion GitHub commits in 2025 (+25% YoY); 43M monthly pull requests (+23%) | Microsoft News: "What's next in AI: 7 trends to watch in 2026" | Dec 8, 2025 |
| AI could replace 300 million full-time jobs | Goldman Sachs, cited in University of North Dakota AI ethics research | Nov 10, 2025 |
| 73% of consumers trust generative AI content | Global study, cited in UND AI ethics research | 2025 |
| 93% of executives consider AI sovereignty a must for 2026 | IBM Institute for Business Value, cited in IBM Think | Jan 1, 2026 |
| "Systems, not models, will define AI leadership" | Gabe Goodhart, Chief Architect, AI Open Innovation, IBM | Jan 1, 2026 |
| "2026 will be the year of frontier versus efficient model classes" | Kaoutar El Maghraoui, Principal Research Scientist, IBM | Jan 1, 2026 |
| "The future isn't about replacing humans, it's about amplifying them" | Aparna Chennapragada, Chief Product Officer for AI Experiences, Microsoft | Dec 8, 2025 |
| "Software practice will evolve from vibe coding to Objective-Validation Protocol" | Ismael Faro, VP Quantum and AI, IBM Research | Jan 1, 2026 |
| "We're seeing the rise of what I call the 'super agent'" | Chris Hay, Distinguished Engineer, IBM | Jan 1, 2026 |
| "The democratization of AI agent creation... moving beyond developers into the hands of everyday business users" | Kevin Chung, Chief Strategy Officer, Writer | Jan 1, 2026 |
| "AI is eating software" / "The paradigm moves from 'writing code' to 'expressing intent'" | Capgemini: Top Tech Trends 2026 | Mar 10, 2026 |
| "Tech sovereignty returns to the top of the agenda, but the race is now for resilient interdependence" | Capgemini: Top Tech Trends 2026 | Mar 10, 2026 |
| "People are getting tired of scaling and are looking for new ideas" / "Robotics and physical AI are definitely going to pick up" | Peter Staar, Principal Research Staff Member, IBM Research Zurich | Jan 1, 2026 |
| "Open source AI is a necessity... Otherwise, you end up with fragmented silos, or a winner-take-all platform" | Anthony Annunziata, Director of Open Source AI, IBM and the AI Alliance | Jan 1, 2026 |
| "Every agent should have similar security protections as humans... to ensure agents don't turn into 'double agents'" | Vasu Jakkal, Corporate VP, Microsoft Security | Dec 8, 2025 |
| "AI will generate hypotheses, use tools and apps that control scientific experiments, and collaborate with both human and AI research colleagues" | Peter Lee, President, Microsoft Research | Dec 8, 2025 |
| "The most effective AI infrastructure will pack computing power more densely across distributed networks... measured by the quality of intelligence it produces, not just its sheer size" | Mark Russinovich, CTO, Microsoft Azure | Dec 8, 2025 |
| "Quantum advantage will drive breakthroughs in materials, medicine and more... The future of AI and science won't just be faster, it will be fundamentally redefined" | Jason Zander, EVP, Microsoft Discovery and Quantum | Dec 8, 2025 |
Final Words
I've produced approximately 4,000 words. An AI could have written 40,000 in the same time, better organized, more comprehensively sourced, with perfect grammar.
But it couldn't have written this—this particular arrangement of anxiety and hope, this specific admission of complicity, this exact call to consciousness. The imperfections are the signature. The struggle is the substance.
The Great Forgetting is real. It's happening now, in your phone, in your browser, in your workflow. The question isn't whether AI will change the world. It already has.
The question is: will you remember what you knew before you forgot?
And if the answer is no—if you can't reconstruct your pre-AI capabilities—then the deeper question: what are you, if not the sum of your delegations?
I don't have an answer. I'm still searching, still struggling, still closing tabs at 3:47 AM to find words that are mine.
Join me, or don't. But notice. Always notice.