AGI has arrived: The 13 most hardcore AI conversations from Sequoia's Annual Meeting

Introduction

At the end of April 2026, Sequoia Capital hosted its fourth AI Ascent conference in San Francisco. The event invited core AI industry players including OpenAI, DeepMind, Anthropic, NVIDIA, and Waymo, as well as startups betting on emerging directions like ElevenLabs, XBOW, Recursive Intelligence, and Starcloud. The 13 dialogues spanned foundational models, programming paradigms, robotics, autonomous driving, chip design, space computing, and novel computing architectures, essentially covering the most cutting-edge threads in the current AI industry.

Compared to previous years, the tone of this year's AI Ascent was more direct: AI is no longer just a tool for improving efficiency but has begun entering real workflows, taking over some complex tasks previously only performable by humans. In its opening speech, Sequoia called this the arrival of "functional AGI"—not that machines are already equivalent to humans in all dimensions, but from a commercial and productivity perspective, long-horizon agents have crossed the threshold from demonstration to usability.

This was also the core backdrop of the conference: as intelligence becomes cheap, callable, and scalable, the focus of AI competition is shifting from "whether the model can do it" to "how to connect it to the real world." Software, services, organizations, hardware, energy, security, and physical space may all be redesigned as a result.

The story Sequoia attempted to tell was clear: intelligence is no longer a luxury but is becoming a new kind of industrial raw material. In the next phase, what truly matters may not be who has the smarter model, but who can understand customers faster, reorganize processes, dispatch agents, and convert this cheap intelligence into a sustainable commercial system.

Therefore, the discussions at this conference were not just about the next step in AI technology, but a larger question: when machines can take on an increasing amount of mental labor, how should humans, companies, and society redefine their own value?

Several Key Themes Running Through the Conference

First, intelligence is becoming a commodity.

Sequoia compared this transformation to "aluminum" in the late 19th century: it was once more expensive than gold, but due to the popularization of the electrolysis process, it became a readily available and ubiquitous industrial material within decades. Today, PhD-level expertise and the cognitive barriers that once defined middle-class competitiveness may be undergoing a similar fate. Advanced intelligence is no longer naturally scarce but is beginning to be mass-produced, called upon, and distributed by models.

Second, the bottleneck is shifting from machines to humans.

Greg Brockman uttered a phrase repeatedly quoted at the conference: when agents can work autonomously, human attention will become the scarcest resource in the entire economy. Karpathy expressed the same judgment more bluntly: when machines can handle almost all execution details, the only ability humans cannot afford to lose is figuring out what they actually want. The problem is no longer whether machines can do it, but whether humans can set the right goals, judge if the results are reliable, and decide what is worth completing.

Third, programming is being solved, but organization is not.

Internally at Anthropic, a large amount of code is already generated by models, with different agents even collaborating autonomously on Slack. Boris Cherny's judgment went further: the real moat is no longer a specific model version, but the degree to which an organization's structure has become "AI-native." For existing companies, this is an unfriendly conclusion—because the gap doesn't just come from tool proficiency, but from whether the company is willing to redesign processes, permissions, collaboration methods, and management structures around agents.

Fourth, AI is returning from the digital world to the physical world.

Jim Fan's robots, Waymo's 20 million autonomous rides, and ElevenLabs' emotional voice synthesis all illustrate from different angles that AI is no longer just a screen tool processing text, code, and images, but is beginning to understand and intervene in light, sound, force, motion, and space. In the past decade, "software eating the world" was the main theme; next, AI may directly enter the physical world, transforming cars, factories, robots, voice interaction, and physical manufacturing itself.

Fifth, the limit of computing power lies in the physical fundamentals.

As land, electricity, and cooling for ground-based data centers hit their limits, a group of more radical companies offer different solutions: Starcloud wants to send chips into space, Recursive lets AI design chips autonomously, Unconventional AI tries to bypass the von Neumann architecture to mimic the brain, and Flapping Airplanes directly questions "brute-force scaling" itself—if humans can learn the same skills with much less data, then today's AI algorithms might be fundamentally too inefficient. The endpoint of the computing power competition is shifting from buying more GPUs to a fundamental reconstruction of energy, chips, architecture, and data efficiency.

Sixth, security has entered an asymmetric battlefield of "AI vs. AI."

XBOW's agent topped the global white-hat hacker leaderboard, meaning AI is no longer just an auxiliary tool for security researchers but an autonomous attack system capable of independently discovering, verifying, and exploiting vulnerabilities. More critically, as the capabilities of open-weight models improve, such attack capabilities could rapidly proliferate within the next 6 to 9 months. Cybersecurity is no longer a battle between human hackers but an AI arms race where the countdown has already started.

Putting these clues together reveals that the AI industry in 2026 is in an uncomfortable position: technological capabilities are already far ahead of product forms, organizational structures, and societal rules. Models get stronger every day, but the "containers" meant to receive them—whether corporate processes, application interfaces, or human attention itself—have yet to catch up.

Essentially, the entire conference's discussions were answering the same question: in a world where machines can handle more and more mental labor, what is left for humans?

Sequoia's answer was somewhat counterintuitive: emotion, trust, and those things that cannot be mass-produced. Brockman's answer was "what you want," and Karpathy's was "whether you can judge if the machine did it right." These answers ultimately point to the same thing: when intelligence itself is no longer scarce, intention, judgment, and relationships will become the new hard currency.

Below is a summary of all 13 dialogues from the conference.

Forum Summaries

Keynote Speeches

Sequoia Partners Opening Keynote: This is AGI

The speakers, Pat Grady, Sonya Huang, and Konstantine Buhler, are the three core partners on Sequoia Capital's AI investment team. Sonya Huang is the author of the globally viral 2022 article "Generative AI: A Creative New World," regarded as one of the first institutional investors to systematically bet on generative AI. Together, they co-authored the 2026 paper "This is AGI," which served as the intellectual framework for this conference. Sequoia Capital itself is a top-tier venture capital firm in Silicon Valley with a long history, having made early investments in Apple, Google, Nvidia, Stripe, and OpenAI.

AI is a "computing revolution" that completely overturns the nature of information processing, not merely a "communication revolution" that accelerates distribution. Past internet and mobile changes only altered the distribution path of information, whereas AI changes the underlying logic of information generation. This causes the floor (technological foundation) upon which developers build applications to shift daily. The importance of this judgment lies in the fact that in a "rainstorm moment" of unstable foundations, the traditional stable technology stack is a thing of the past; developers must learn to dance with the constantly evolving model base.

AI will enter a $10 trillion market—ten times larger than traditional software—by directly delivering "professional services." The global software market's TAM is only a few hundred billion dollars, whereas the US legal services vertical alone is worth $400 billion, equivalent to the entire software industry. This advocates for a key shift: AI's commercial value is no longer as a tool sold to humans, but directly taking over and delivering high-value work originally performed by human experts in the form of agents.

From a business practice perspective, long-horizon agents capable of autonomously recovering from failure signify that AGI has already arrived. If a system can be tasked, self-correct when it fails, and persist until completion, it is functionally equivalent to AGI. This counterintuitive judgment reminds us: stop obsessing over academic definitions. An AI capable of independent execution has evolved from a "faster horse" into a "car" that changes the competitive dimension, achieving a 10 to 40-fold leap in efficiency.

In a time of rapidly shifting underlying capabilities, the only logic for building a moat is to be "extremely close to the customer." The MAD strategy—Moats, Affordance (the intuitive and easy-to-use degree of a product), and Diffusion—advocates locking in value through a customer-back approach rather than tech-out. Because human needs change much slower than model capabilities, this deep customer immersion offers more durability than chasing models.

Agent autonomy is leapfrogging from "minute-level assistants" to "hour-level autonomous employees." The meter chart (task persistence metric), which measures how long a model can stay on track in complex tasks, has jumped from minutes a year ago to hours now, sufficient to support "dark factories" (fully autonomous business processes) without human review. This means the productivity bottleneck has been broken, and extraordinary iterations like "rewriting 8 million lines of code in 6 weeks" are becoming the norm.

Human society is on the eve of a "cognitive industrial revolution," where machines will handle 99.9% of the world's mental labor. Just as the Industrial Revolution replaced 99% of physical labor with engines, the vast majority of analysis, decision-making, and creation in the future will be handled by neural networks. The assertion of this judgment is that intelligence will no longer be a monopolized resource of humans but a low-cost, industrial-grade consumable that can be mass-produced and called upon on demand.

Advanced intellectual skills are about to have their "aluminum moment," transitioning from an expensive luxury to a cheap commodity. Aluminum, once more expensive than gold, became disposable due to the popularization of electrolysis, and AI's instant access to PhD-level knowledge will have the same effect. This foreshadows a brutal future: professional knowledge barriers built over years could collapse in an instant, and intelligence itself will no longer command a scarcity premium.

When intelligence becomes universally commonplace, human relationships and emotional connections will become the only true anchors of value in human society. Photography once pushed art from realism towards Impressionism, which expressed the soul. Similarly, AI's optimal solutions for efficiency often present "alien spaces" that transcend human intuition. The final conclusion is counterintuitive yet profound: in a future where machines handle all work, trust and emotion between people are the ultimate hard currencies that cannot be mass-produced by machines.

If you could only remember one thing from this dialogue, what would it be?

Smartness that used to be valuable will soon become as cheap as plastic bags. In the future, what truly keeps you competitive won't be a brain that can solve tough problems, but an ability to understand others and build trust.

Models and Cognition

Andrej Karpathy: From Vibe Coding to Agent Engineering (OpenAI Founding Team)

The speaker, Andrej Karpathy, is one of the most influential "educator-scientists" in the AI community. He was a founding member of OpenAI, later became Tesla's AI Director responsible for the autonomous driving vision system, and left Tesla in 2024 to found the AI education company Eureka Labs. His YouTube series teaching neural networks step-by-step is the introductory textbook for countless AI engineers. He coined key concepts like "Software 2.0" and "Vibe Coding."

Even top experts feel "outdated" in the AI wave, as technological evolution has leaped from auxiliary tools to autonomous systems. In early 2026, the speaker found he no longer needed to modify code blocks generated by AI; he simply trusted the system to complete complex tasks. The importance of this judgment is that when AI can achieve self-correction and closed-loop delivery, the baseline for developers, once built on accumulated experience, is violently raised, making personal learning speed difficult to keep up with the displacement speed of the technological base.

Modern computing is entering the Software 3.0 era, where LLMs are essentially a new type of computer using context as a lever. Software 1.0 was writing code, 2.0 was training weights, and 3.0 is programming through prompting within a context window (the memory space for the model to process information). This means installing software no longer requires writing complex compatibility scripts; just "feed" a description to the agent. Precise spelling of details is no longer a core competency.

Many existing application architectures are becoming "redundant" because AI can already process raw data directly. The speaker found a menu generation app he painstakingly developed became meaningless, as models can now perform pixel-level rendering directly onto photos. This advocates a profound shift: AI should not just be used to accelerate old business logic; we must realize that the disappearance of middle layers means many traditional product forms have lost their physical foundation for existence.

AI capabilities are "jagged," exhibiting superhuman intelligence only in verifiable domains. A model can refactor a hundred thousand lines of code but might fail at simple common sense, like counting the 'r's in "strawberry." This is because models are primarily reinforced through RL (Reinforcement Learning, a training method using reward signals to guide evolution) in verifiable domains like math and code. This reminds us: we must constantly observe within the loop, remaining wary of weaknesses outside the model's training distribution.

We are not building "animals" with intrinsic motivation, but "summoning ghosts" from the data distribution. The peak intelligence of a model depends on the training data distribution (e.g., adding a large amount of chess game data rapidly improves chess skill), not on it generating some kind of biological curiosity. This counterintuitive judgment points out that AI doesn't have true "understanding"; it only maximally reinforces specific circuits through statistical simulation. Therefore, users must learn to identify and avoid false capabilities unsupported by data.

Agentic engineering is about maintaining the quality red line of professional software while leveraging stochastic AI. This new engineering method requires developers, while coordinating unstable yet powerful agents, to ensure the system produces no security vulnerabilities. It advocates a new 10x engineer paradigm: the core of competition is no longer the speed of writing code yourself, but the ability to efficiently drive a massive cluster of agents as a director to deliver high-quality results.

When machines take over trivial API details, the true premium for humans will shift towards aesthetics and mastery of the "specification." Developers no longer need to memorize specific interface parameters of PyTorch (a deep learning framework), as these details will be handled by AI "interns" with excellent memory. This predicts a counterintuitive future: foundational principles and design taste will outlast tool details. Humans should transform from "brick-layers" to decision-makers defining "what is good design."

"Thinking" can be outsourced, but "understanding" is the sole bottleneck for humans in the age of cheap intelligence. Although AI can assist us in processing and recompiling vast amounts of information, it cannot decide for us "why to build this" or "whether it is valuable." This advocates an ultimate conclusion: humans remain the sole commander of the system, because only human consciousness can assign goals to the intelligence processing process. This holistic understanding cannot be replaced by algorithms.

If you could only remember one thing from this dialogue, what would it be?

When machines can do all the work for you, even think through all the details, the only skill you can't afford to lose is figuring out what you actually want and whether you can tell if the machine did it right.

Greg Brockman: Human Attention is the New Bottleneck (OpenAI Co-founder)

The speaker, Greg Brockman, is Co-founder and President of OpenAI. Former CTO of Stripe, he co-founded OpenAI with Sam Altman in 2015 and is the core architect of the company's technology and infrastructure. Internally at OpenAI, Altman handles external matters (fundraising, public image, policy), while Brockman handles internal matters (technology, compute, products). His engineer-style of personally writing code and staying up late during releases is well-known in Silicon Valley.

Intelligence has become a standardized, resalable commodity, leading to an insatiable, pathological growth in demand for computing power. OpenAI's business model essentially involves buying or leasing compute, transforming it into intelligence through models, and selling it at a premium. Because the demand for problem-solving is infinite, the projected supply of GPUs in 2026 is nearly zero. The importance of this judgment is that AI is no longer just a software service; it has evolved into a resource-based commodity business where the physical world's compute supply directly determines the upper limit of civilization's intelligence.

The scaling law (the empirical rule that model capabilities improve with increased computation) is a universal empirical truth, showing no signs of hitting a "wall." Although the basic concepts of neural networks originated in the 1940s, as long as massive amounts of compute are continuously invested, a model's various capabilities will accordingly and deterministically improve. This advocates a key point: technological stagnation will not occur in the short term. As long as capital and electricity are continuously invested, we will obtain more powerful intelligence, providing the underlying logic for tech giants' aggressive investments.

From a functional perspective, we have already completed 80% of the journey to AGI, as models now possess the closed-loop ability to execute tasks independently. A system engineer handed a complex optimization problem to a model; the model not only wrote the code but also autonomously ran a Profiler (performance analysis tool) and performed multiple rounds of optimization based on feedback until the task was fully completed. This advocates a counterintuitive point: AGI is not a future moment but an ongoing process. AI has evolved from a "coding assistant" into a "colleague who can solve problems."

Context (the background information a model masters for a specific task) is replacing model algorithms as the core competitive frontier. The new tool Chronicle can record everything a user does on their computer in real-time, giving AI "memory" and eliminating the time humans spend repeatedly explaining background to the machine. The importance of this judgment is that for entrepreneurs, one-time model training is no longer the only moat; building a "data harness"