AI Made Us Faster. So Why Are We Fried?

AI was supposed to make us faster. So why are we fried? Engineer Cole Medin articulates a paradox: autonomous AI agents and sophisticated second-brain systems have made him more productive than ever, yet he concludes his days feeling utterly depleted, his attention span shattered. This isn't just about working hard; it’s about a new form of cognitive strain.

The core issue is that increased capability begets increased pressure. Medin notes a common sentiment: if you can ship ten times more, you feel compelled to. When an AI agent is processing a task, waiting for its output feels like wasted time. This incentivizes users to spin up another agent, initiating a new task or conversation, leading to multiple parallel workflows.

This constant task-switching creates what Medin calls the “fried brain effect.” He describes jumping between agent conversations, estimating he context-switches 100 times more frequently than just a few years ago. While Medin acknowledges this specific figure might be an exaggeration for effect, he cites Tibo’s similar account of extreme context switching as a widely recognizable pattern, not an isolated complaint. The result is a fragmented attention economy, leaving users exhausted, not invigorated.

AI agents excel at parallel processing, spinning up multiple sessions simultaneously to tackle diverse tasks. Humans, however, cannot parallelize. Our brains are single-threaded, forcing us to constantly switch attention between agent conversations, decision points, code reviews, and new prompts. This creates a relentless cycle of cognitive load.

Consider the real-world workflow: engineers initiate tasks in several agent sessions, monitor each for questions or outputs, respond, and then attempt to return to unfinished work. Each switch incurs an attention residue, leaving mental fragments of the previous task. This persistent switching is the engine of the "fried brain effect" Medin describes.

Frequent task switching decimates deep-work time. The American Psychological Association (APA) highlights the significant switching costs involved, noting that even brief interruptions can derail focus for extended periods. While Medin’s estimate of switching "100 times more" than five years ago might be hyperbolic, the underlying mechanism of cognitive fatigue is undeniably real.

Before AI, engineers often spent hours in uninterrupted flow states, focused on a single problem. Now, orchestrating multiple agents feels like the only path to "ultra productive" output, leading to the abandonment of sustained, deep work. This shift means less time for focused problem-solving and more time managing a swarm of digital assistants.

The allure of spending several uninterrupted hours solving a single, complex problem has faded, replaced by the perceived inefficiency of serial work. Why dedicate a full afternoon to one task when a fleet of agents could be orchestrating a dozen in parallel? This shift fundamentally redefines what "productive" means, yet our biology struggles to keep pace.

Cole Medin, a software engineer and educator, highlights how this new paradigm pushes individuals to seek sustained attention outside work. He cites a yearning for the "flow state" of deep coding, now replaced by the constant monitoring of agentic processes. Medin, an avid chess player and reader, finds himself turning to these activities for the focused engagement his workday no longer provides.

Cal Newport's observation that "the ability to focus is becoming the scarcest commodity of the 21st century" resonates more profoundly than ever, Medin notes. AI agents, while powerful, demand constant supervision, intensifying the challenge. This is not a call to abandon agents; Medin values their gains. Rather, it’s an acknowledgment that the human brain, unlike its silicon counterparts, does not parallelize. For more on the cognitive toll of rapid task switching, see Multitasking: Switching Costs - American Psychological Association. Until AI agents achieve true autonomy, we remain stuck in a high-output, high-fragmentation loop, where the pressure to ship 10 times more means feeling guilty for simply waiting.

Medin proposes a crucial trade-off: instead of yielding to the temptation of adding another agent session with every capability improvement, developers must invest in a more autonomous, less babysat workflow. This means building better harnesses that allow agents to operate for extended periods without human intervention.

Practically, this involves clearly defining intent upfront, providing agents with reusable rules and skills, and implementing a robust plan–implement–validate loop. Generic goal commands are insufficient; autonomy must be earned through reliable, quantifiable validation.

Batch decisions and review completed work together, requiring human approval only at critical junctures. This approach shifts the developer’s role from constant context-switcher to strategic orchestrator, leveraging AI’s parallel processing without fragmenting their own attention. True productivity comes from enabling agents to run independently, freeing up precious human focus for deep work, not from juggling a hundred micro-conversations daily.

Why can AI agents make developers feel more tired?

Running several agents creates frequent monitoring, decision-making, and task switching. The work may move faster, but human attention remains limited.

What is the AI productivity paradox?

AI increases the amount of work one person can produce, but that extra capacity can raise expectations and cognitive load rather than reduce working hours.

How can developers reduce context switching when using AI agents?

Batch agent check-ins, limit parallel sessions, and improve workflows so agents can plan, implement, and validate tasks with less supervision.

Does reducing agent monitoring mean trusting AI blindly?

No. The goal is to define intent and validation clearly, then review completed work in batches while retaining approval over important decisions.