The human brain consumes about 15 to 20 watts of power, less than a lightbulb, yet it drives us to complete countless chores, travel, work overtime, and more. While the 24-hour physical day remains constant, human 'bandwidth' (or mental energy, decision-making capacity) is extremely limited and depletes daily.

Thus, leveraging AI, half the global tech community is collaborating to build a time compressor, attempting to conserve brainpower and liberate mental energy.

For instance, a user of Meta's latest personal AI agent, Muse, shared online: While friends vigorously discussed year-end flight tickets in a group chat for days before finalizing a flight, Muse soon proactively asked, 'Your friends in the group have booked that flight. Shall I book it for you too?' After confirmation, Muse handled everything from reservation, phone verification to final payment.

Unlike conversational AI, the recently popular Personal Agents now handle the 'last mile' tasks entirely. This marks a familiar yet new trend.

Tianyancha reveals that on October 21, Today AI, a personal AI assistant startup founded by Qi Junyuan, former product VP of ByteDance's Feishu and head of Doubao PC, announced it had secured $200 million (approximately RMB 1.336 billion) in cumulative funding. Top-tier investors like IDG Capital, Tencent, 5Y Capital, Ant Group, and Sequoia China China are all on board.

Its core product, Today, launched in late September, similarly targets 'proactive service'—after integrating with WeCom, WeChat Reading, Baidu Netdisk, Kingsoft Docs, and Tencent Meeting, it proactively records work group messages, extracts insightful quotes from books, tracks document revisions, and compiles meeting minutes without waiting for user prompts.

Adding to the mix, OpenClaw, which ignited the geek community earlier this year, along with a flurry of newcomers like Tencent's WorkBuddy, Doubao Work, Qianwen GO, and domestic and international counterparts like Manus team's new product Cue—even AI phones—though differing in verticals and development goals, all share the mission of eliminating operational friction in work and daily life, helping users save time.

Yet, users often fall into another tool cycle, chasing new hits or simply spending time on AI-generated short dramas. While some master true energy conservation, others consume time twice over, letting AI-saved time slip away again.

Why, amidst a flourishing AI landscape, do ordinary consumers feel not at ease but a sense of confusion, as if they must overhaul their lives every few months to adapt to new tools? When terminal applications split into two parallel universes, humanity begins to fight itself.

I. AI for Saving Time Demands a Steep 'Adaptation Tax'

As personal agents rapidly spark market interest, Qi Junyuan, in an interview with LatePost, revealed a product truth: After Today AI's launch, users showed extreme polarization—some uninstalled the mobile app within just two minutes; others entrusted it with more daily tasks, resulting in a rare 'upturned tail' retention curve, entering a stable phase of non-continuous decline.

Why does a thoughtful assistant designed to save trouble and reduce burdens drive some away in just two minutes? Because the logic of AI interaction has changed.

Over the past three years, the public has grown accustomed to ChatGPT or Doubao's conversational AI, which essentially operates on a 'single-task basis'—encounter a problem, open a chatbox, and leave once done, with no mutual obligations between human and AI.

But assistant products like Muse, to bridge the gap from 'just offering advice' to 'getting things done,' uniformly adopt architectures combining 'long-term dialogue + global memory archive + independent cloud execution space.'

For AI to provide timely services without your prompt, it must first possess a 'panoramic view' of your digital life:

Muse's ability to accurately book the same flight as your friends stems from its authorized access to your social chats, emails, bills, and schedules. An assistant handling work effectively needs access to WeCom workflows, private files in netdisks, or voice streams from Tencent Meetings.

Theoretically, any ultimate tool aiming to 'save time' for consumers must first impose a hefty 'context migration cost' and 'lifestyle reengineering tax' on users before delivering efficiency gains.

This explains why, despite the seemingly vibrant AI landscape in 2026, users chasing trends feel exhausted.

Early in the year, everyone fumbled with high-threshold OpenClaw tutorials; mid-year, they migrated daily work to WorkBuddy or Doubao Work; by autumn, they were exploring Muse, Cue, and Today AI. Each transition required reauthorizing privacy boundaries, retraining AI memory preferences, and rewriting decade-old operational muscle memory.

Some products offer partial adaptation, migrating work habits and experiences, but effectiveness remains suboptimal.

In contrast, the market favors 'time-killing' apps like AI-generated short dramas, which follow a 'zero-threshold, human- comply with ' model—no cloud setup, no netdisk permissions, no lifestyle changes required. A single swipe sends a surge of dopamine straight to the brain.

While 'time-killing' apps subtract for users, 'time-saving' agents initially demand addition. When adaptation costs outweigh minor time savings, 'two-minute uninstalls' proliferate (excluding AI nativists, whose time already belongs to AI).

II. From 'Saving Physical Effort' to 'Making Decisions': The Shaking Foundation of Information Intermediaries

Once users surmount initial adaptation hurdles and truly adopt personal agents, profound differences between office and personal life scenarios emerge.

Office agents like WorkBuddy primarily eliminate 'execution friction.' When a boss assigns a task with clear goals and standards, AI retrieves data, generates tables, and writes summaries, sparing users keyboard taps and software switches—but final decisions remain firmly in human hands.

However, when Muse, Today AI, and Alibaba's Qianwen GO enter personal life, they overstep execution boundaries into 'decision-making substitution.'

Qi Junyuan mentioned a real-life change in an interview: Since deeply using personal agents, he rarely opens review or navigation sites anymore.

The logic is simple—to find a satisfying restaurant, users once sifted through countless fake or genuine posts on review platforms; when hailing a ride, despite clear cost-effectiveness across Didi, Gaode, or Cao Cao Mobility, most people, Too much hassle , wouldn’t bother comparing all three apps.

Now, agents with independent cloud execution spaces and global memory instantly perform cross-platform searches, comparisons, and filters in the background, presenting the optimal choice—or even, like Muse, asking if you want to pay.

This touches on the second underlying axiom of AI applications: Personal agents save operational time superficially but fundamentally take over consumers' 'comparison and choice rights.' Yet, traditional mobile internet's most lucrative profits rely on humans' cognitive laziness in comparing options.

The human brain, operating at roughly 20 watts, naturally takes shortcuts amid information overload—preferring ads at the top, visually appealing pages, or familiar platforms to place orders.

Over the past two decades, whether Amazon and Booking overseas or local life services, OTAs, and e-commerce giants domestically, their ability to charge hefty commissions and bid-based ad fees fundamentally exploits the 'information intermediary dividend' from humans' limited cognitive bandwidth.

If personal agents shoulder cross-platform comparisons and fraud screening, making purely rational decisions based on users' true preferences and historical bills, traditional super apps' painstakingly cultivated 'two-sided network effects' and brand premium will face challenges.

Multiple overseas tech media outlets report that Amazon recently explicitly blocked Meta's Muse agent crawlers in its website code to prevent direct user price comparisons and orders bypassing the frontend; domestically, AI phones being blocked by apps exemplify the same issue.

When 'time-saving' AI truly completes the last mile, it reconfigures not just consumers' schedules but the entire mobile internet's traffic distribution foundation. New models are needed.

III. Physical Time Can Be Saved, but Human 'Mental Energy' Remains Constant

After dissecting adaptation barriers and commercial rivalries, we must return to the core question: If AI saves us abundant time at work and in chores, why doesn't this time make us feel more fulfilled? Instead, 'time-killing' entertainment algorithms accelerate their dominance. Is it due to insufficient AI penetration, human preferences, or something else?

Viewed purely from the consumer's perspective as a 'human,' the answer may lie in the disjunction between 'physical time' and 'cognitive mental energy.'

Each person's 24-hour physical day is fixed, with a significant portion as 'non-negotiable assets.' The 'mental energy (cognitive bandwidth)' that truly enables thinking, loving, and forging genuine connections is a scarce, easily depleted resource.

In the pre-AI era, it drowned in ticket bookings, reimbursements, weekly reports, group message replies, and countless other frictions. By nightfall, decision-free consumption forms easily dominated.

By 2026, numerous efficiency agents have made a difference—schemes once taking three days or reports two hours now finish in half an hour. But the saved time packs in denser task schedules and multithreaded decisions, escalating judgment pressure.

More intriguingly, large models themselves have become distorted in this efficiency frenzy.

Reports indicate that the Today AI team initially planned a steady 'software-hardware-model training' progression but had to break plans, forming an in-house team for post- and mid-training earlier.

The trigger? An internal evaluation revealed an anomaly: While the latest large models excelled in Coding and complex logical reasoning, they regressed noticeably in natural conversation, nuanced language sense, and 'communicating like a human,' even falling behind previous versions.

To serve as a sharp 'efficiency blade' at work, mainstream models grow increasingly mechanical, cold, and result-oriented. But as an all-day personal life assistant, infinite efficiency pursuit isn't optimal. Thus, like many internet products, personal agents seek ways to deliver small joys.

This may reveal the third, profound axiom of AI application evolution: The ultimate goal of saving time is never to cram in more work or halt the brain but to help humans restore depleted 'mental energy' and reclaim perception of real life.

Advanced errand-running tools risk depriving people of participation in specific tasks, sliding them toward aimless Nothingness . Thus, personal agents that truly endure and retain users likely offer values beyond efficiency, reclaiming agency to 'live mindfully.'

In the vibrant yet confusing autumn of 2026, immature AI applications already test the boundaries of human habits and psychological resilience. From our consumption habits, it's clear the tug-of-war between 'time-killing' and 'time-saving' won't end soon and may coexist long-term. But consumers are already questioning whether AI is too abundant. Hence, AI products should preemptively consider:

Technology can handle most processes for humanity, but 'how to spend the saved time' remains humanity's ultimate, non-negotiable sovereignty.

Source: Songguo Finance