《金融时报》专栏与西班牙经济政策智库（Esade Centre for Economic Policy）主任豪尔赫·加林多（Jorge Galindo）进行对谈，探讨代理型人工智慧（Agentic AI）如何彻底改变知识工作者的日常产出。加林多近期运用 AI 将累积多年的西班牙住房危机政策研究转化为一部动画解说影片，并获得超过两百万次观看。他指出，若在没有 AI 的情况下，由于外包制作的机会成本过高，这支影片根本不会问世；这证明 AI 不仅是取代既有工作，更能催生出过去无法实现的全新产出。

这部影片的成功并非仅靠简单的提示词，而是建立在加林多深厚的专业知识与美学品味之上。他向 Claude 模型输入了数百行个人笔记、自己的著作与外部数据，并亲自决定叙事结构、类比方式及情绪板（mood board），而整个制作过程仅消耗了他每月 200 欧元订阅额度中的约 8 欧元成本。对话中提到，像加林多这样同时具备领域专业与视觉美学素养的人相当罕见，因此未来的职场平衡点可能是由领域专家提出构想，再交由能利用 AI 将产能提升十倍的创意专业人员来执行。

然而，当资料分析与启动新专案的成本变得极度低廉时，知识工作者面临的新瓶颈转变为「选择与专注」。加林多表示，AI 让整理分散数据与开展新研究变得轻而易举，导致团队涌现数千个点子，必须严格筛选以免过度频繁发布而稀释内容价值。此外，对于热爱探索新问题的研究人员而言，生成式 AI 非但没有如预期般释放闲暇时间，反而让人更容易在周末不断开启新专案，进而对工作与生活平衡及团队的中长期发展带来隐忧。

The Financial Times sat down with Spanish economist Jorge Galindo, director of the Esade Centre for Economic Policy think-tank, to explore how agentic AI is transforming the output of knowledge workers. Recently, Galindo used AI to turn years of policy research on Spain’s housing crisis into an animated explainer video that garnered over two million views. He noted that in a counterfactual world without AI, this video simply would not have existed because the opportunity cost of hiring external providers was too high, demonstrating that AI enables people to create entirely new outputs rather than merely automating existing tasks.

Rather than relying on a simple two-sentence prompt, the video's success depended heavily on Galindo's substantial domain expertise and aesthetic background. He fed Claude hundreds of lines of notes, his own book, and external data sources, while personally directing the narrative structure, metaphors, and mood board—all at an effective cost of about €8 from his €200 monthly subscription. Because combining deep subject matter expertise with visual aesthetic taste is unusual, a likely future equilibrium in companies is for domain experts to hand off concepts to experienced creative professionals who can use AI to produce ten times as much work.

As the cost of conducting analysis and starting new projects drops dramatically, the new bottleneck for knowledge workers has become choice and focus. Galindo explained that while LLMs make gathering unstructured data effortless and generate thousands of project ideas, teams must carefully select which ones to pursue to avoid devaluing their output. Furthermore, for curious minds driven to constantly ask and answer the next question, generative AI ends up consuming personal time—like Saturday mornings—by tempting them to endlessly open new projects rather than freeing up their schedules, raising concerns about medium- and long-term impacts on teams.