\u201cAGI is here, now.\u201d<\/p>\n
That\u2019s Sequoia Capital this week, one of Silicon Valley\u2019s most legendary venture firms and a major OpenAI investor, declaring we\u2019ve crossed the threshold into artificial general intelligence.<\/p>\n
Their post also proclaims<\/a>, in big bold letters, that they are \u201cBlissfully Unencumbered by the Details\u201d. When Sequoia speaks, the tech world listens. This post has dominated conversations across the AI builder community for days.<\/p>\n As a builder, venture investor, and AI scientist myself, I find their proclamation both deeply useful and profoundly dangerous.<\/p>\n Here\u2019s what\u2019s useful<\/p>\n Sequoia offers a functional definition of AGI: \u201cthe ability to figure things out. That\u2019s it\u201d. AI can now crawl through information, determine a path forward, and execute. The key shift, as they frame it, is AI moving from \u201ctalking\u201d to \u201cdoing\u201d. Harvey and Legora \u201cfunction as associates,\u201d Juicebox \u201cfunctions as a recruiter,\u201d OpenEvidence\u2019s Deep Consult \u201cfunctions as a specialist.\u201d Those are their exact phrases, and I\u2019m skeptical of the framing, but we\u2019ll get to that.<\/p>\n They\u2019re throwing down the gauntlet for builders, and this matters. AI systems can actually redline contracts now, genuinely reach out to prospects today. It\u2019s a reminder to think bigger about what\u2019s possible and that the horizon has expanded dramatically in just the last year.<\/p>\n I sent this post to my own co-founders, not to debate the philosophy, but I wanted to push us on Sequoia\u2019s \u201cdoing\u201d versus \u201ctalking\u201d framework. We need to rise to this challenge ourselves.<\/p>\n However, calling these systems \u201cAGI\u201d is profoundly damaging, both to the credibility of the AI revolution and to its safe deployment. It obscures what AI agents can actually do today (hint: not general superintelligence) and offers zero guidance on how humans should interact with them (hint: don\u2019t trust them blindly).<\/p>\n Let me give three examples that expose these systems\u2019 real limitations.<\/p>\n AI systems cannot handle out-of-distribution situations<\/p>\n I wrote about this in my last column<\/a>, but the Greenland crisis offers a live, unfolding example. I tested how GenAI tools, such as ChatGPT 5.2 with maximum \u201cthinking and research\u201d mode, would analyze the situation. Could these supposedly AGI systems help me understand a fast-moving geopolitical event?<\/p>\n They couldn\u2019t even conceive of it happening.<\/p>\n I shared a screenshot of the Wikipedia page documenting the crisis. Every model told me it was fabricated, \u201cbullshit\u201d, impossible. When I pushed further, citing real news sources, ChatGPT kept telling me to \u201crelax\u201d \u2013 insisting \u201cthis is not a real crisis\u201d.\u00a0 The models are so anchored in the traditional norms of the Western Alliance, the system simply cannot output context that contradicts its training data, even when confronted with primary sources.<\/p>\n AI \u201cthinking\u201d fails when the situation sits outside its training distribution: gaslighting users instead of escalating uncertainty, refuses humility and keeps \u201creasoning\u201d even when it\u2019s wrong. If policymakers or politicians are using these tools to understand Greenland right now, that\u2019s genuinely dangerous.\u00a0<\/p>\n AI systems reflect the beliefs of their builders<\/p>\n