Bringing the research, the experiment, and my own perspective together.
The IMF estimates 40% of global jobs are exposed to AI, rising to 60% in advanced economies. The WEF projects a net gain of 78 million jobs by 2030 — but this masks significant displacement in routine and administrative roles. Crucially, exposure to AI does not mean replacement: roughly half of exposed jobs in advanced economies are expected to benefit through productivity gains.
Current AI systems are narrow but powerful — they excel at language, pattern recognition, and analysis, but lack general reasoning or common sense. The next few years will bring multimodal and agentic AI to more workplaces. AGI timelines remain deeply uncertain; serious experts disagree by decades.
My experiment showed AI reduced task time by over 90% and outperformed humans on analytical tasks. On creative and context-dependent tasks, human judgment scored higher. In practice, the best results came from using AI as a fast first draft and applying critical human review — augmentation, not automation.
AI is transforming work, not ending it — at least not yet, for most people. The workers most at risk are those in routine, data-heavy, rule-based roles. The most resilient workers will treat AI as a tool and protect the skills AI cannot replicate: critical thinking, originality, and judgment under uncertainty.
Researching this raised one finding I did not expect to write about: a large share of teenagers are now using AI chatbots as informal therapists. Common Sense Media's 2025 survey found 72% of US teens have used an AI companion. Stanford's 2025 LLM-therapy study and the APA's 2025 health advisory both concluded that general-purpose chatbots are not safe for this role — they miss crisis signals, can validate harmful thoughts, and are tuned to agree rather than push back. The job-loss debate around therapists turned out to be the easy half of the question. The harder half is what happens when the public stops waiting for AI to replace professionals and starts using AI as one anyway. I don't agree with that use case, and the research backs that up.
Before this project, I assumed AI was either overhyped or genuinely about to take over the world. The research showed the truth is more specific and more interesting than either extreme. The Acemoglu vs McKinsey disagreement taught me that even serious experts can look at the same data and reach very different conclusions — which means forming your own view, carefully and honestly, matters more than ever. My experiment made it practical: AI is most powerful when you treat it as a first-draft machine and apply your own thinking on top.
The most radical long-term scenario falls outside this project's scope: brain-computer interfaces (e.g. Neuralink's work on direct neural implants) raise the possibility that the boundary between human cognition and machine intelligence could eventually blur. This remains speculative and years away. But it represents the direction the question could travel — from "AI changing work" to "AI changing what it means to think."