AI Can Do Anything. That’s the New Problem

Generative AI has ushered in an era where creative barriers crumble, transforming the act of imagination into immediate possibility. Suddenly, a novel isn't just a novel; it's a comic, an animated series, or a game's lore, all within reach. Each new avenue, however, competes for the same finite human hours and mental bandwidth.

This surge in possibility presents a new kind of problem: the paradox of choice. A wider menu of options, as psychologist Barry Schwartz observed, can lead to more effort spent comparing possibilities and less time actually choosing and initiating a project. With models like Opus 5.5, Astra, and Soul 6.1, the capacity for AI to execute is nearly limitless, but human capacity to direct and consume is not.

David Shapiro, an AI researcher, highlights this shift, noting that while delegation to agents like GrokBot is feasible, the executive function required to review, enjoy, and ultimately use the results remains non-fungible. AI expands what one person can attempt, but it does not expand the number of days in a year or one's inherent capacity to care about every generated outcome. The bottleneck has shifted from AI's capability to human attention.

The constraint has shifted: the bottleneck is no longer the machine, but you. David Shapiro, a prescient AI researcher, observes that as models like Claude Opus 5.5, Astra, and Soul 6.1 grow exponentially more capable, most users hit their personal limits for attention, review time, and decision-making long before they exhaust their token quotas. This marks a critical inflection point in human-AI interaction.

Delegation to AI agents like GrokBot or ChatGPT is now seamless; they can draft, research, or produce assets asynchronously. However, Shapiro stresses that executive function is not fungible. A human still needs to evaluate, steer, and verify outputs. Generating a torrent of files, whether a fanfic or a research dossier, does not make them useful; it simply accumulates digital clutter.

This unchecked accumulation creates more cognitive overhead than value. The real-world utility of an AI-generated novel, for instance, hinges entirely on a human reading and engaging with it. Without this crucial human element of consumption and intentionality, even the most impressive AI output becomes pointless token spend, a digital hoarding of unexamined work.

Easy generation often leads to digital hoarding. Books, research dossiers, and multimedia assets accumulate, unread and unused, a testament to expended tokens, not actual progress. David Shapiro notes that we are "just accumulating files" unless we actively engage with the output.

AI lowers the friction of production, transforming creative ideation into a "slop fire hose," as Shapiro aptly puts it. This deluge, however, does not guarantee originality, usefulness, or an audience. Generating a sequel to Cyberpunk 2077 with Opus 5.5 might be technically feasible, but its value is zero if it never sees human eyes.

This flood of output forces a critical re-evaluation of the quality-versus-quantity challenge. With the bottleneck shifting from token limits to human attention, every project demands a rigorous filter: Will I review this? Will someone use it? Does making it serve a goal or value I genuinely care about?

If the answer is no, treat the experiment as mere entertainment—or skip it entirely. This intentionality isn't about productivity for its own sake, but about reclaiming our finite executive function from the endless possibilities AI presents. For further reading on the cognitive overload of too many options, consider Barry Schwartz: The Paradox of Choice. Otherwise, you’re just spending tokens to create more digital clutter.

Shapiro’s video game metaphor of “min-maxing” provides a potent principle for navigating this new AI landscape. Instead of scattering your finite attention across every appealing generative possibility, strategically concentrate your time and AI tools on a select few high-value projects. Prioritize depth over breadth; a single, fully realized endeavor far outweighs a dozen half-baked concepts.

Adopt a lightweight, outcome-driven workflow. Begin by clearly naming your intended outcome, whether it’s a published article or a finished animation. Before prompting GrokBot or Claude Opus, establish a review checkpoint: will this generation directly advance your stated goal? If the output doesn’t align, stop or redirect your efforts. This disciplined approach prevents the accumulation of unread files and pointless token spend.

This focus isn't a demand for relentless productivity. Curiosity and playful experimentation remain invaluable, but they should be chosen deliberately. The aim is to consciously direct your finite attention, ensuring that your AI-assisted efforts, whether for work or for fun, genuinely serve what matters most to you.

Why can AI’s growing capabilities make it harder to get things done?

More options can create decision fatigue, while reviewing and using AI-generated work still requires limited human time and attention.

What does it mean to say humans are the bottleneck in AI workflows?

AI may generate work quickly, but people must still choose goals, check results, make judgments, and decide what is worth using.

How can people avoid accumulating unused AI-generated content?

Before starting, decide who will use the output, whether you will review it, and how it supports a goal you actually value.

Does focusing AI use mean avoiding experimentation?

No. Experimenting can be worthwhile in its own right; the key is to choose it intentionally rather than treat every capability as an obligation.