• Core Findings: Large-scale randomized controlled trials ($N = 1,222$) reveal that while AI assistance boosts immediate problem-solving performance, it significantly damages a user's independent performance and persistence once the AI is removed.
• Rapid Onset: These negative cognitive effects manifest after only brief periods of interaction with an AI assistant (approximately 10–15 minutes).
• The "Persistence Muscle": Standard AI assistants operate as short-sighted collaborators, providing instant and complete answers. This deprives users of the "productive struggle" necessary for learning, conditioning them to expect immediate results and causing them to give up much quicker when forced to work independently.
• Domain-Generality: The reduction in persistence and the decline in independent success rates were robustly replicated across fundamentally different cognitive domains, specifically mathematical reasoning (fraction-solving) and reading comprehension (SAT-style tests).
• Direct Solutions vs. Hints: The decline in capability is highly concentrated among users who request direct answers from the AI. Conversely, users who leverage AI exclusively for hints, clarifications, or interactive scaffolding show no significant impairment compared to control groups.
• Implications for AI Design: Current AI optimization strategies favor short-term helpfulness, which risks eroding human cognitive capabilities over time. The study highlights an urgent need to pivot AI development toward reinforcing long-term competence.

大多数人认为AI辅助编程只能带来适度的生产力提升，但作者认为Symphony系统实现了500%的代码合并增长率，这是一个惊人的数字。这个数据点挑战了人们对AI辅助编程效果的传统预期，表明正确的AI编排可能带来指数级的生产力提升。

这一观点揭示了‘tokenmaxxing’作为衡量员工AI使用能力的新趋势，暗示了数据消耗成为衡量生产力的一种方式。

大多数人认为AI工具提高了生产力，但作者将其与内容泛滥和平台限制直接关联，暗示AI不仅提高了数量还可能损害了社区质量。这种观点挑战了'AI总是进步'的乐观叙事，提出了技术应用的负面后果。

大多数人认为高AI代码生成比例意味着软件开发效率的大幅提升，但作者指出这只是编码阶段的加速，不包括架构设计、调试、审查等更耗时的环节，因此高AI贡献比例并不等同于整体生产力的提升。

大多数人认为AI生成的代码会被大量采纳，但研究显示只有不到一半的AI生成代码最终被用户保留。这表明AI编程助手的实际贡献远低于表面看起来那么大，用户对AI输出有很高的筛选和修正率。

这一惊人的效率对比揭示了AI时代产品开发的根本性转变：Anthropic仅用极小团队就能构建直接挑战拥有2000名员工的Figma的产品。这挑战了传统软件公司需要大量人力的假设，预示着更小、更专注的团队可能主导未来市场。

这一惊人的数据揭示了AI在工作场所应用中的潜在陷阱。虽然AI被宣传为提高生产力的工具，但近半数员工报告收到过看似精美但不准确或无用的AI生成内容。这表明过度依赖AI可能导致质量下降，挑战了AI总是带来积极效果的假设。

这一功能暗示了一种全新的开发范式，设计师的视觉编辑可以直接转化为生产级代码。这可能会显著减少前端开发中的手动编码工作，但也引发了关于AI生成代码质量和可维护性的重要问题。

令人惊讶的是：AI模型能在不到半分钟内完成完整的应用程序构建计划并编写全部代码，这展示了AI在软件开发领域的惊人效率，远超人类开发者的常规速度。

令人惊讶的是：编程已成为AI在企业中最主要的应用场景，其规模远超其他用例近一个数量级。工程师使用AI工具可以将生产力提高10-20倍，这一惊人的效率提升解释了为什么企业愿意如此迅速地采用AI编程工具，也颠覆了人们对软件开发工作流程的传统认知。

令人惊讶的是：顶尖工程师可能同时管理20-30个AI代理，生产力呈数量级提升。这一事实揭示了AI对软件开发效率的革命性影响，远超大多数人的预期。

令人惊讶的是：文章指出顶级工程师可能同时管理20-30个AI代理，实现数量级的生产力提升。这一数字远超传统认知，暗示AI正在重新定义个人生产力的极限。这种能力意味着未来软件公司的组织结构可能需要彻底重构，从大型团队转向小型高效团队。

令人惊讶的是：高盛经济学家报告显示，正确使用AI的员工每天可节省40-60分钟，与因技术摩擦损失的时间几乎对称。这揭示了一个悖论：AI既可以是效率倍增器，也可以是生产力杀手，关键在于如何实施。

引入了“tokenmaxxing”这一核心概念，将AI生产力提升的本质定义为“最大化token消耗”。这打破了传统节省算力的思维，反直觉地认为用尽全力消耗token才能榨取AI的最大价值，本质上是在探讨如何将电力最高效地转化为智力劳动。

AI agent 完成任务的实际速度比人类快数倍——但这个事实几乎从未出现在主流 AI 能力讨论中。「2 小时时间地平线」被大众理解为「AI 能做人类 2 小时的工作」，但实际上 AI 可能只需 20-30 分钟就完成了这个任务。这意味着 AI 的实际生产力倍数远高于时间地平线数字所暗示的，而低估 AI 效率的讨论普遍存在。

Summary of "AI Fatigue is Real"

• The Productivity Paradox: AI significantly speeds up individual tasks (e.g., turning a 3-hour task into 45 minutes), but this doesn't lead to more free time. Instead, the baseline for "normal" output shifts, and the work expands to fill the new capacity, leading to a relentless pace.
• From Creator to Reviewer: Engineering work is shifting from "generative" (energizing, flow-state tasks) to "evaluative" (draining, decision-fatigue tasks). Developers now spend their days as "quality inspectors" on an unending assembly line of AI-generated code.
• The Cost of Nondeterminism: Engineers are trained for determinism (same input = same output). AI’s probabilistic nature creates a constant cognitive load because the output is always "suspect," requiring more rigorous review than code written by a trusted human colleague.
• Context-Switching Exhaustion: Because tasks are "faster," engineers now touch 6–8 different problems a day instead of focusing on one. The mental cost of switching contexts so frequently is "brutally expensive" for the human brain.
• Skill Atrophy: Much like GPS has weakened our innate sense of direction, over-reliance on AI coding tools can cause core technical reasoning and mental mapping of codebases to atrophy.
• Strategies for Sustainability:
  • Time-boxing: Setting strict timers for AI sessions to avoid "prompt spirals."
  • Separating Phases: Dedicating mornings to deep thinking and afternoons to AI-assisted execution.
  • Accepting "Good Enough": Setting the bar at 70% usable output and fixing the rest manually to reduce frustration.
  • Strategic Hype Management: Ignoring every new tool launch and focusing on mastering one primary assistant.

On ai, productivity and shorter work weeks and why that will not happen

• Task Expansion & Role Blurring: AI lowers the barrier to entry for complex tasks, leading employees to take on work outside their core expertise. Product managers and designers are now writing code, while researchers take on engineering tasks.
• Specialist Burden: This expansion creates a "cleanup" tax. For example, senior engineers now spend significant time reviewing, debugging, and mentoring colleagues who produce "vibe-coded" AI outputs, often through informal and unmanaged channels like Slack.
• The "Ambient Work" Phenomenon: Because AI interactions feel conversational and "easy," work has become ambient. Employees find themselves prompting AI during lunch, between meetings, or late at night, eliminating natural mental downtime.
• Intensified Multitasking: Workers are running multiple AI agents in parallel while simultaneously performing manual tasks. This creates a high sense of "momentum" but leads to extreme cognitive load and constant attention-switching.
• The Productivity Trap: AI acts as a "partner" that makes revived or deferred tasks feel doable. This creates a flywheel where people don't work less; they simply take on more volume, leading to "unsustainable intensity" that managers often mistake for genuine productivity.
• Sustainability Risks: The researchers warn that while AI feels like "play" initially, it eventually leads to cognitive fatigue, impaired decision-making, and burnout as the quiet increase in workload becomes overwhelming.

Hacker News Discussion

• Cognitive Fatigue: Users highlighted that "AI fatigue" is distinct from normal work tiredness. It stems from the "constant vigilance" required to audit AI output and the lack of a "flow state" due to unpredictable waiting times for generations.
• Executive Function Strain: Commenters noted that managing autonomous agents is more exhausting than manual work. One user compared it to Level 3 autonomous driving—you aren't driving, but you must remain "fully hands-on" to ensure the AI doesn't touch the wrong files or hallucinate.
• The Jevons Paradox: Several participants pointed out that as the "cost" of work decreases due to AI, the demand for work increases proportionally. Instead of saving time, workers are expected to triple their output, which leaves them more stressed than before.
• Management Expectations: A common theme was that leadership often mandates AI usage and pre-supposes productivity gains, leaving no room for cases where AI makes work slower or lower quality. This forces employees to "perform" productivity while working longer hours.
• Vibe Coding vs. Engineering: There is a heated debate between those who see "vibe coding" (prompt-heavy development) as a massive efficiency gain and veterans who argue it produces "average code" that becomes a maintenance nightmare in large, legacy codebases.

• The author identifies two primary personality traits: "The Builder" (focused on velocity, utility, and shipping) and "The Thinker" (needing deep, prolonged mental struggle).
• "Thinking hard" is defined as sitting with a difficult problem for days or weeks to find a creative solution without external help.
• In university, the author realized this ability to chew on complex physics problems was their "superpower," providing a level of confidence that they could solve anything given enough time.
• Software engineering was initially gratifying because it balanced both traits, but the rise of AI and "vibe coding" has tilted the scale heavily toward the Builder.
• While AI enables the creation of more complex software faster, the author feels they are no longer growing as an engineer because they are "starving the Thinker."
• The lack of struggle leads to a feeling of being stuck, as the dopamine of a successful deploy cannot replace the satisfaction of deep technical pondering.

Hacker News Discussion

• The loss of the "clayship" process: Commenters compared coding to working with clay; skipping the struggle means missing the intimacy with the material that reveals its limits and potential.
• The "Vending Machine" effect: Receiving a "baked and glazed" artifact from AI removes the human element of discovery and learning.
• Risk of mediocrity: There is concern that AI guides developers toward "average" or conventional solutions, making it harder to push for unique or innovative ideas without significant manual effort.
• The tradeoff of efficiency: While some view the current era as the best time for "Builders" who just want to see results, many veteran developers feel a profound sense of loss regarding the cognitive depth of the craft.
• Clear communication as a new skill: Some argue that interacting with AI requires a different kind of "thinking hard"—specifically, the need to express creative boundaries clearly so the model doesn't "correct" away the uniqueness of the project.