The Forgetting Curve: Ankur Warikoo’s Three Red Flags and the Quiet Crisis of Cognitive Offloading

There’s a moment that arrives silently, without fanfare, when the tool becomes the crutch, and the crutch becomes the limb. Ankur Warikoo, the Indian entrepreneur and author who has built a career on demystifying success and failure for millions, recently published a stark warning that cuts to the bone of our current technological moment. In an editorial for the Economic Times, Warikoo articulated what many of us have felt but few have dared to name: we are losing the muscle memory of thought itself [1]. His thesis is deceptively simple—three dangerous signs of AI dependence—but the implications ripple outward into every corner of the tech industry, from Nvidia’s stratospheric valuations to the $15 billion compute deals that now define the AI arms race [2][3].

This is not another hand-wringing op-ed about the singularity. It is a cold, hard look at what happens when a generation of knowledge workers, executives, and creatives forgets how to operate without a digital co-pilot. As the industry barrels toward autonomous agents that can run for 35 hours without human intervention, Warikoo’s warning feels less like philosophy and more like a diagnostic [4].

The Three Signs: A Diagnostic for the Age of Cognitive Offloading

Warikoo’s framework is elegant in its brutality. The first sign, he argues, is when you have “forgotten how you operated without AI” [1]. This is not hyperbole; it is a measurable phenomenon. When a knowledge worker cannot draft an email, structure a presentation, or debug a line of code without first consulting a large language model, the boundary between augmentation and atrophy has been crossed. The second sign is the inability to start a task without AI’s permission, as if the machine must first bless the endeavor before human cognition can engage. The third is the creeping anxiety that arises when the AI is unavailable—a withdrawal symptom that mirrors the dopamine loops of social media addiction [1].

What makes Warikoo’s analysis so potent is its timing. It arrives as the infrastructure of AI dependence is being laid down at an unprecedented scale. Consider the numbers: Nvidia, whose GPUs power the vast majority of AI workloads, just posted another record quarter [2]. The company also revealed it holds $43 billion in startup investments, a staggering war chest that gives it unparalleled leverage over the entire AI ecosystem [2]. Meanwhile, SpaceX’s IPO filing revealed that Anthropic is paying $15 billion a year to access its data centers [3]. Let that sink in: $15 billion annually for compute, a sum that dwarfs the GDP of small nations. This is not a speculative bubble; this is the cost of admission to the game.

The sources do not specify the exact breakdown of Nvidia’s record revenue, but the trend is unmistakable. The demand for AI compute is insatiable, and every new capability—from autonomous agents to multi-day model execution—only accelerates the cycle. Warikoo’s warning about forgetting how to operate without AI is not just a psychological observation; it is an economic inevitability when the tools are this powerful and this expensive.

The Agent Era: When 35 Hours of Autonomy Becomes the Baseline

To understand why Warikoo’s three signs are so dangerous, one must understand what the AI industry is building right now. VentureBeat recently reported on Alibaba’s Qwen3.7-Max, a model that can run for approximately 35 hours of continuous autonomous execution [4]. This is not a chatbot that answers questions; this is an agent that plans, executes, and course-corrects complex tasks over days rather than seconds [4]. The model supports external harnesses like Anthropic’s Claude Code, meaning it can interact with other AI systems, APIs, and development environments without human oversight [4].

The sources do not specify the exact cost of running Qwen3.7-Max for 35 hours, but the price tag for such capability is likely in the millions. VentureBeat notes a figure of $2.08 million, though the context of that number is not fully detailed in the excerpt [4]. What is clear is that the industry has fully entered what VentureBeat calls the “agent era,” a paradigm where AI models do far more than generate text [4]. They now actively plan, execute, and course-correct.

This environment makes Warikoo’s warning existential. If a knowledge worker has already forgotten how to write a memo without AI, what happens when the AI can run for 35 hours without supervision? The cognitive offloading is no longer a crutch; it is a full delegation of agency. The worker becomes a spectator to their own productivity, a quality assurance manager for a machine that does the actual thinking.

The sources do not provide data on how many organizations already use autonomous agents in production, but the trajectory is clear. Alibaba’s Qwen3.7-Max is a Chinese model, but it supports the same harnesses used by Western AI giants like Anthropic [4]. The technology is global, and the dependence is cross-border. Warikoo’s three signs are not limited to any one geography or industry; they are a universal symptom of a species that has outsourced its cognition.

The $15 Billion Question: Who Owns the Cognitive Infrastructure?

The Wired report on SpaceX’s IPO filing adds a layer of geopolitical and financial complexity to Warikoo’s framework. Anthropic, one of the leading AI safety companies, is paying $15 billion a year to access SpaceX’s data centers [3]. This is not a trivial expense; it is a bet that the compute demands of frontier AI models will only grow. The sources do not specify the exact terms of the deal, but the scale is unprecedented. For context, $15 billion is roughly the annual R&D budget of a major pharmaceutical company. It is the entire market cap of many publicly traded firms.

What does this have to do with Warikoo’s three signs? Everything. When the cost of compute is this high, the pressure to maximize utilization is immense. Organizations that pay $15 billion for data center access will not tolerate human inefficiency. They will optimize for throughput, which means pushing more tasks to AI agents and reducing the friction of human oversight. The very infrastructure of the AI industry is designed to accelerate the forgetting that Warikoo warns about.

The sources do not provide data on how many jobs have been eliminated or restructured due to this compute-driven optimization, but the logic is inescapable. If a human costs $100,000 per year and an AI agent costs $15 billion in compute (amortized across millions of tasks), the math favors the machine. Warikoo’s second sign—the inability to start a task without AI—becomes a feature, not a bug, of the corporate bottom line.

The Nvidia Paradox: Record Profits and the Fragility of Dependence

Nvidia’s record quarter is a double-edged sword [2]. On one hand, it validates the thesis that AI is not a fad but a fundamental shift in computing. On the other hand, it creates a single point of failure that is almost unimaginable in its scale. Nvidia holds $43 billion in startup investments, meaning it has a vested interest in the success of the very companies that are driving AI dependence [2]. The sources do not specify which startups Nvidia has invested in, but the portfolio likely spans the entire AI stack, from model developers to infrastructure providers.

This concentration of power is a systemic risk. If Nvidia stumbles—due to supply chain disruptions, geopolitical tensions, or a technological breakthrough that renders its GPUs obsolete—the entire edifice of AI dependence could wobble. Warikoo’s third sign, the anxiety that arises when AI is unavailable, would become a collective panic. The sources do not provide data on Nvidia’s supply chain vulnerabilities, but the company’s own forecast of slowing revenue growth suggests that even the king of AI is not immune to gravity [2].

The sources diverge in their focus: TechCrunch emphasizes Nvidia’s financial performance and startup holdings [2], while Wired zeroes in on the Anthropic-SpaceX deal [3]. But both point to the same underlying reality: the AI industry is building a cognitive infrastructure that is both incredibly powerful and incredibly fragile. Warikoo’s warning about forgetting how to operate without AI is not just a personal development issue; it is a systemic vulnerability.

The Forgetting Economy: Winners, Losers, and the Unseen Costs

Who benefits from AI dependence? The obvious winners are the infrastructure providers: Nvidia, SpaceX, and the data center operators collecting $15 billion annual checks [2][3]. The less obvious winners are the companies that sell the tools that make forgetting easy. Every AI writing assistant, code generator, and autonomous agent is a product designed to reduce the friction of human cognition. The sources do not provide data on the market size of these tools, but the trend is unmistakable.

The losers are more diffuse but no less real. They include the knowledge workers who lose the ability to think independently, the organizations that become locked into expensive compute contracts, and the society that forgets how to solve problems without a machine’s blessing. Warikoo’s framework suggests that the cost of AI dependence is not measured in dollars but in cognitive atrophy. The sources do not provide data on the long-term effects of AI dependence on human cognition, but the analogy to other forms of technological dependence—GPS eroding spatial memory, spellcheck eroding spelling skills—is well-documented.

The sources also do not specify whether Warikoo’s warning has met with pushback or praise. But the very fact that it appears in a mainstream Indian business publication suggests that the conversation is moving from the fringe to the center. The Economic Times is not a tech blog; it is a newspaper read by executives, investors, and policymakers. Warikoo’s message is reaching the people who are making the decisions about AI adoption.

The Editorial Take: What the Mainstream Media Is Missing

The mainstream coverage of AI tends to oscillate between breathless hype and apocalyptic fear. Warikoo’s contribution is to ground the conversation in the mundane reality of daily work. His three signs are not about superintelligence or job displacement; they are about the quiet erosion of competence. The sources do not provide data on how many people recognize these signs in themselves, but the anecdotal evidence is overwhelming.

What the mainstream media is missing is the feedback loop between infrastructure and dependence. Every record quarter for Nvidia, every $15 billion compute deal, every 35-hour autonomous agent makes Warikoo’s warning more urgent. The industry is not just building tools; it is building a system that rewards forgetting. The cognitive offloading that Warikoo describes is not a bug; it is the product.

The sources also do not address the question of reversibility. Once a knowledge worker has forgotten how to operate without AI, can they relearn? The answer is unclear. The sources do not provide data on cognitive retraining programs or the plasticity of adult brains under conditions of prolonged AI dependence. But the stakes are high enough that the question deserves serious investigation.

Conclusion: The Memory We Are Choosing to Lose

Ankur Warikoo’s three signs of AI dependence are not a prediction; they are a diagnosis of the present. We have already forgotten how we operated without AI, and the infrastructure being built today—from Nvidia’s $43 billion investment portfolio to Anthropic’s $15 billion compute deal to Alibaba’s 35-hour autonomous agents—is designed to ensure we never have to remember [1][2][3][4]. The question is not whether this dependence is dangerous; the question is whether we have the collective will to recognize the danger before the forgetting becomes irreversible.

The sources do not provide a solution, and Warikoo does not claim to have one. But the act of naming the problem is itself a form of resistance. In an industry that measures progress in floating-point operations and token throughput, Warikoo’s warning is a reminder that the most important metric is not compute but cognition. And cognition, unlike a GPU, does not come with a warranty.

---

References

[1] Editorial_board — Original article — https://economictimes.indiatimes.com/magazines/panache/you-have-forgotten-how-you-operated-without-ai-ankur-warikoo-reveals-3-dangerous-signs-of-ai-dependence/articleshow/131255235.cms

[2] TechCrunch — Nvidia posts another record quarter, reveals $43B of holdings in startups — https://techcrunch.com/2026/05/20/nvidia-posts-another-record-quarter-reveals-43-billion-of-holdings-in-startups/

[3] Wired — SpaceX IPO Filing Reveals Anthropic Is Paying $15 Billion a Year to Access Its Data Centers — https://www.wired.com/story/spacex-ipo-anthropic-compute-finances-risks/

[4] VentureBeat — Alibaba's proprietary Qwen3.7-Max can run for 35 hours autonomously and supports external harnesses like Anthropic's Claude Code — https://venturebeat.com/technology/alibabas-proprietary-qwen3-7-max-can-run-for-35-hours-autonomously-and-supports-external-harnesses-like-anthropics-claude-code