WEBVTT 00:00:00.010 --> 00:00:02.330 For the past couple of years, we've been told 00:00:02.330 --> 00:00:04.730 to treat artificial intelligence like an intern. 00:00:05.230 --> 00:00:07.690 You type a prompt, and the AI helps you brainstorm 00:00:07.690 --> 00:00:10.289 a marketing campaign or draft a polite email 00:00:10.289 --> 00:00:13.089 to your boss. You do the heavy lifting, and the 00:00:13.089 --> 00:00:16.030 machine acts as your digital copilot. This chart 00:00:16.030 --> 00:00:18.429 shows the actual impact of early chatbots on 00:00:18.429 --> 00:00:21.250 a standard 40 -hour workweek. A recent study 00:00:21.250 --> 00:00:23.350 of Danish workers found that using these text 00:00:23.350 --> 00:00:25.649 -based models saved them an average of just 2 00:00:25.649 --> 00:00:28.579 .8 % of their time. their hours didn't drop, 00:00:28.660 --> 00:00:31.420 and their wages stayed exactly the same. But 00:00:31.420 --> 00:00:34.020 the software has evolved past that simple textbox 00:00:34.020 --> 00:00:36.899 interface. The industry is rapidly moving towards 00:00:36.899 --> 00:00:40.640 agentic AI. Agentic AI agents are software entities 00:00:40.640 --> 00:00:42.880 designed to perceive their environment, make 00:00:42.880 --> 00:00:45.640 multi -step decisions, and take actions autonomously 00:00:45.640 --> 00:00:48.119 to achieve specific goals. Instead of waiting 00:00:48.119 --> 00:00:50.679 for you to ask it a question, an agentic system 00:00:50.679 --> 00:00:53.299 like Claude Cowork runs in the background. You 00:00:53.299 --> 00:00:55.600 give it a high -level goal, and the software 00:00:55.600 --> 00:00:58.659 reads your files, writes new code, and executes 00:00:58.659 --> 00:01:02.000 an entire chain of actions on its own. This pie 00:01:02.000 --> 00:01:04.260 chart illustrates how businesses are actually 00:01:04.260 --> 00:01:07.519 deploying these new systems. According to a September 00:01:07.519 --> 00:01:11.500 2025 report from Anthropic, 75 % of companies 00:01:11.500 --> 00:01:14.099 using their software have abandoned the collaboration 00:01:14.099 --> 00:01:17.379 model entirely. They are using it for full task 00:01:17.379 --> 00:01:20.799 delegation. The comforting idea of the AI assistant 00:01:20.799 --> 00:01:23.840 was just a temporary bridge. The transition to 00:01:23.840 --> 00:01:26.239 fully automated white -color knowledge work is 00:01:26.239 --> 00:01:30.180 already underway. In mid -2025, many companies 00:01:30.180 --> 00:01:32.599 started abandoning their early generative AI 00:01:32.599 --> 00:01:35.620 pilot projects because the basic chatbots struggled 00:01:35.620 --> 00:01:38.420 to integrate complex data. The tech industry 00:01:38.420 --> 00:01:40.739 solved this problem by removing the human from 00:01:40.739 --> 00:01:43.219 the middle of the workflow. Under the new model, 00:01:43.459 --> 00:01:46.629 a manager simply defines a final objective. The 00:01:46.629 --> 00:01:49.870 AI agent takes that objective, figures out the 00:01:49.870 --> 00:01:52.810 necessary subtasks, makes autonomous decisions 00:01:52.810 --> 00:01:55.069 about how to sequence them, and does the work 00:01:55.069 --> 00:01:58.769 itself. We saw exactly how capable this architecture 00:01:58.769 --> 00:02:03.209 is in February 2026. A researcher set up an experiment 00:02:03.209 --> 00:02:06.810 where a network of 16 distinct Claude Opus 4 00:02:06.810 --> 00:02:09.969 .6 agents were given a single prompt and left 00:02:09.969 --> 00:02:12.789 to collaborate with each other. Together, these 00:02:12.789 --> 00:02:15.729 agents successfully built a C compiler from scratch 00:02:15.729 --> 00:02:18.930 using the programming language Rust. A compiler 00:02:18.930 --> 00:02:22.009 is a complex program that translates human -readable 00:02:22.009 --> 00:02:24.770 source code into machine code that a computer's 00:02:24.770 --> 00:02:27.490 processor can execute. Writing one from scratch 00:02:27.490 --> 00:02:30.210 is a highly rigorous test of logic and software 00:02:30.210 --> 00:02:33.669 engineering. If a network of agents can autonomously 00:02:33.669 --> 00:02:36.250 map out and build a deeply technical systems 00:02:36.250 --> 00:02:39.310 tool, Routine administrative tasks like drafting 00:02:39.310 --> 00:02:42.069 financial reports or analyzing legal contracts 00:02:42.069 --> 00:02:45.689 are trivial by comparison. This wave of automation 00:02:45.689 --> 00:02:49.210 is pointed directly at the modern office. Researchers 00:02:49.210 --> 00:02:53.169 at OpenAI estimate that 80 % of the U .S. workforce 00:02:53.169 --> 00:02:56.129 will see at least some of their daily work affected 00:02:56.129 --> 00:02:59.169 by large language models. While physical trades 00:02:59.169 --> 00:03:02.110 remain largely insulated for now, Roles rooted 00:03:02.110 --> 00:03:05.550 entirely in processing information, like accountants 00:03:05.550 --> 00:03:08.389 and web designers, and advanced mathematicians, 00:03:08.689 --> 00:03:12.569 are the most highly exposed. Dario Amadei, the 00:03:12.569 --> 00:03:15.449 CEO of Anthropic, was explicit about what this 00:03:15.449 --> 00:03:18.930 means. He warned that AI is on track to eliminate 00:03:18.930 --> 00:03:21.590 white -collar jobs across the board, specifically 00:03:21.590 --> 00:03:24.009 targeting entry -level positions in finance, 00:03:24.150 --> 00:03:26.870 law, and consulting. We've already seen this 00:03:26.870 --> 00:03:29.370 rapid obsolescence happen within the AI industry 00:03:29.370 --> 00:03:33.710 itself. In 2023, prompt engineer, someone skilled 00:03:33.710 --> 00:03:36.129 at typing the exact right instructions into a 00:03:36.129 --> 00:03:38.629 chatbot, was one of the highest -paying, most 00:03:38.629 --> 00:03:42.210 sought -after new careers. By 2025, it was effectively 00:03:42.210 --> 00:03:44.490 a dead profession, because the models learned 00:03:44.490 --> 00:03:47.009 to intuit user intent completely on their own. 00:03:47.469 --> 00:03:50.889 The technology is moving so fast that it is actively 00:03:50.889 --> 00:03:53.530 dismantling the very careers it created just 00:03:53.530 --> 00:03:56.210 two years ago, long before it finishes clearing 00:03:56.210 --> 00:03:59.280 out traditional administrative roles. Look ahead 00:03:59.280 --> 00:04:02.819 to the year 2030. As businesses invest trillions 00:04:02.819 --> 00:04:05.759 into AI data centers and infrastructure, the 00:04:05.759 --> 00:04:08.219 basic mechanics of how a company operates will 00:04:08.219 --> 00:04:10.960 look completely different. These warnings are 00:04:10.960 --> 00:04:13.439 now coming from the heart of American manufacturing. 00:04:14.159 --> 00:04:16.959 Ford CEO Jim Farley predicts that artificial 00:04:16.959 --> 00:04:19.500 intelligence is going to replace literally half 00:04:19.500 --> 00:04:22.100 of all white -collar workers in the United States. 00:04:22.660 --> 00:04:25.060 The workers who remain will not be doing the 00:04:25.060 --> 00:04:28.350 hands -on drafting, coding, or analyzing. Instead, 00:04:28.670 --> 00:04:30.910 their daily routine will look like middle management. 00:04:31.410 --> 00:04:33.930 They will act as supervisors, reviewing outputs 00:04:33.930 --> 00:04:36.610 and managing a fleet of autonomous software agents. 00:04:37.370 --> 00:04:39.769 This graph shows an emerging corporate paradox. 00:04:40.550 --> 00:04:43.069 Historically, industries relied on wide entry 00:04:43.069 --> 00:04:45.550 -level bases as training grounds for future management. 00:04:46.029 --> 00:04:49.430 But if autonomous agents perfectly execute repetitive 00:04:49.430 --> 00:04:52.209 junior tasks, how do graduates learn the business? 00:04:52.649 --> 00:04:55.629 You cannot supervise a process you've never executed. 00:04:56.029 --> 00:04:59.110 The immediate disruption of mass layoffs is only 00:04:59.110 --> 00:05:02.250 the first wave. The lasting impact of agentic 00:05:02.250 --> 00:05:05.170 AI is the permanent removal of the stepping stones 00:05:05.170 --> 00:05:07.949 that build professional expertise, leaving us 00:05:07.949 --> 00:05:10.310 to figure out what an office job actually is 00:05:10.310 --> 00:05:11.930 when the machine does all the work.