AGI đã đến: 13 cuộc đối thoại AI đỉnh cao nhất tại Hội nghị thường niên của Sequoia

Introduction

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

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

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

The story Sequoia is trying to tell is clear: intelligence is no longer a luxury good but is becoming a new type of industrial raw material. In the next phase, what truly matters is perhaps not who has the smarter model, but who can understand customers faster, reorganize workflows, orchestrate agents, and translate this cheap intelligence into a sustainable business system.

Therefore, the conference discussions were not just about the next step for AI technology, but a larger question: as machines can take on more and more cognitive labor, how should individuals, companies, and society redefine their own value?

Key Themes Running Through the Conference

First, intelligence is becoming a commodity.

Sequoia likened this transformation to "aluminum" in the late 19th century: once more expensive than gold, it became a cheap, readily available industrial material within decades due to the popularization of the electrolytic process. Today, Ph.D.-level expertise and the cognitive barriers that once defined middle-class competitiveness might be undergoing a similar fate. Advanced intelligence is no longer naturally scarce; it is being mass-produced, called upon, and distributed by models.

Second, the bottleneck is shifting from machines to humans.

Greg Brockman uttered a phrase that was repeatedly quoted throughout 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 skill humans cannot afford to lose is knowing what they truly want. The problem is no longer whether the machine can do it, but whether humans can set the right goals, judge if the results are reliable, and decide what is worth accomplishing.

Third, programming is being solved, but organizations are not.

Internally at Anthropic, a significant amount of code is generated by models, and different agents can even collaborate autonomously on Slack. Boris Cherny's assessment goes further: the true moat is no longer a specific model version, but the degree to which an organization's structure is "natively built for AI." For existing companies, this is an unsettling conclusion—because the gap stems not just from proficiency with tools, but from whether a company is willing to redesign its processes, permissions, collaboration methods, and management structure around agents.

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

Jim Fan's robotics, Waymo's 20 million autonomous rides, and ElevenLabs' emotional voice synthesis all illustrate from different angles that AI is no longer just a screen-bound tool for processing text, code, and images. It is beginning to understand and intervene in light, sound, force, motion, and space. In the past decade, "software is 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 limits of computing power lie in the physical fundamentals.

As land, electricity, and cooling for data centers on the ground hit their limits, a group of more radical companies are offering different solutions: Starcloud wants to put chips in space, Recursive lets AI design chips autonomously, Unconventional AI attempts 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 far less data, then today's AI algorithms might be fundamentally too inefficient from the start. The endpoint of the computing power race is moving from buying more GPUs to a fundamental restructuring of energy, chips, architectures, and data efficiency.

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

XBOW's agent topped the global white-hat hacker rankings, 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, with the improvement of open-source models, such attack capabilities could rapidly proliferate within the next 6 to 9 months. Cybersecurity is no longer a battle between human hackers; it's an AI arms race with a countdown that has already started.

Putting these clues together reveals that in 2026, the AI industry is in an uncomfortable position: technological capabilities have far outpaced product forms, organizational structures, and societal norms. Models are getting stronger every day, but the "containers" meant to house them—whether corporate workflows, application interfaces, or human attention itself—haven't caught up yet.

Essentially, all the discussions at the conference were trying to answer the same question: in a world where machines can perform an increasing amount of mental labor, what is left for humans?

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

Below is a summary of all 13 panel discussions from the conference.

Forum Summaries

Keynote Speeches

Sequoia Partners' Opening Keynote: This is AGI

The speakers Pat Grady, Sonya Huang, and Konstantine Buhler are three core partners on the AI investment team at Sequoia Capital. Sonya Huang is the author of the globally viral 2022 article "Generative AI: A Creative New World" and is considered one of the first institutional investors to systematically bet on generative AI. The three co-authored the 2026 article "This is AGI," which served as the intellectual framework for this conference. Sequoia Capital itself is a top-tier venture capital firm with a long history in Silicon Valley, 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 shifts only changed how information is disseminated, but AI changes the underlying logic of information generation, causing the floor on which developers build applications to shift daily. The importance of this judgment lies in the fact that during a "rainstorm moment" of unstable foundations, the era of the stable tech stack is over; developers must learn to dance with the ever-evolving model base.

AI will penetrate a $10 trillion market—ten times larger than traditional software—by directly delivering "professional services." The global TAM for software is only a few hundred billion dollars, whereas the U.S. legal services market alone is $400 billion, equivalent to the entire software industry. This advocates for a key transition: AI's commercial value is no longer about selling tools to humans but directly using agents to take over and deliver the high-value work originally done by human experts.

From a business practice perspective, long-duration agents capable of autonomously handling failure signify that AGI has arrived. If a system can be dispatched to perform a task, self-repair upon failure, and persist to the end, it is functionally equivalent to AGI. This judgment counterintuitively reminds us: stop obsessing over academic definitions; AI with independent execution capabilities has evolved from a "faster horse" to a "car" that changes the competitive landscape, achieving a 10 to 40-fold leap in efficiency.

In a time of rapidly changing underlying capabilities, the only logic for building a moat is to be "extremely close to the customer." The MAD strategy—Moats, Affordance, and Diffusion—advocates locking in value customer-back rather than tech-out. Since human needs change much slower than model capabilities, this deep wrapping around customers is more durable than chasing models.

Agent autonomy is leaping from "minute-level assistants" to "hour-level autonomous employees." The meter chart measuring how long a model stays on track in complex tasks has leaped from minutes a year ago to several hours now, enough to support dark factories that require no 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 cusp 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 will soon be handled by neural networks. The assertion here is that intelligence will no longer be a monopoly 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 experience their "aluminum moment," transitioning from expensive luxury goods to cheap commodities. Aluminum, once more precious than gold, became disposable due to the popularization of electrolysis; AI's instantaneous access to Ph.D.-level knowledge will have the same effect. This foreshadows a brutal future: professional knowledge barriers built over years could collapse instantly, 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 society. Photography once pushed art from realism to the soulful expression of Impressionism; similarly, AI's optimization for efficiency often presents "alien spaces" that defy human intuition. The final conclusion is counterintuitive yet profound: in a future where machines do all the work, only trust and emotion between humans will be the ultimate hard currency that machines cannot mass-produce.

What is the one thing to remember from this discussion?

Intelligence, which used to be valuable, will soon become as cheap as a plastic bag. What will truly keep you competitive in the future isn't your ability to solve problems, but your ability to understand others and build trust.

Models & Cognition

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

The speaker, Andrej Karpathy, is the most influential "educator-scientist" in the AI community. A founding member of OpenAI, he later served as Tesla's AI Director, responsible for the autonomous driving vision system. In 2024, he left Tesla to found the AI education company Eureka Labs. His YouTube series explaining neural networks step-by-step is considered a primer for countless AI engineers. He coined key concepts like "Software 2.0" and "Vibe Coding."

Even top experts can feel "left behind" by the AI wave, as technology's evolution has leapfrogged from assistive tools to autonomous systems. In early 2026, the speaker found he no longer needed to modify AI-generated code blocks, only needing to trust the system to complete complex tasks. The importance of this realization lies in the fact that when AI can self-correct and deliver results end-to-end, the developer's "baseline" built on experience is violently raised, making individual learning speed difficult to keep up with the shifting technological foundation.

Modern computing is entering the Software 3.0 era, where an LLM is essentially a new type of computer using context as its lever. Software 1.0 was writing code, 2.0 was training weights, and 3.0 is programming within the context window via prompting. This means installing software no longer requires writing complex compatibility scripts; you just "feed" a description to the agent. Precision in spelling out details is no longer a core competency.

Many existing application architectures are becoming "redundant" because AI can now directly process data at the raw data layer. The speaker realized that a menu generation app he had painstakingly developed became meaningless because the model could now perform pixel-level rendering overlays directly on a photo. This advocates for a profound shift: AI shouldn't just be used to accelerate old business logic. We must realize that the disappearance of the middle layer means many traditional product forms have lost their physical basis for existence.

AI capabilities are "jagged," showing superhuman intelligence only in domains that can be verified. A model can refactor a hundred thousand lines of code but stumble on a simple common-sense question like counting the 'r's in "strawberry". This is because models are primarily enhanced via RL in verifiable domains like math and code. This reminds us: we must constantly observe within the loop, staying vigilant against weaknesses outside the model's training distribution.

We are not building "animals" with intrinsic motivation but "conjuring ghosts" within the data distribution. The peak intelligence of a model depends on the training data distribution (e.g., adding vast chess data makes it a chess master rapidly), not on it developing some kind of biological curiosity. This judgment counterintuitively points out: AI doesn't truly "understand"; it just extremely refines specific circuits through statistical simulation. Therefore, users must learn to identify and avoid the false capabilities unsupported by data.

Agentic engineering aims to maintain the quality red lines of professional software while leveraging stochastic AI. This new engineering method requires developers to coordinate unstable but powerful agents while ensuring the system doesn't introduce security vulnerabilities. It advocates for a new 10x engineer paradigm: the core of competition is no longer the speed of writing code oneself but the ability to efficiently direct a large cluster of agents, like a director, to deliver high-quality results.

When machines take over trivial API details, human premium will shift towards aesthetics and control over the "specification document." Developers no longer need to memorize specific PyTorch API parameters, as these details can be handled by AI "interns" with excellent memory. This foreshadows a counterintuitive future: foundational principles and design taste are more durable than tool details. Humans should transition from "bricklayers" to decision-makers who define "what good design looks like."

"Thinking" can be outsourced, but "understanding" is the only bottleneck limiting humans in the age of cheap intelligence. Although AI can help us process and recompile vast amounts of information, it cannot decide for us "why build this" or "is this valuable?". This advocates for the ultimate conclusion: humans remain the sole commander of the system, because only human consciousness can imbue the intelligent processing with purpose. This holistic understanding cannot be replaced by algorithms.

What is the one thing to remember from this discussion?

When machines can do all the work and even think through all the details for you, the only skill you can't afford to lose is knowing 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 the 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, Altman handles outward-facing matters (fundraising, public image, policy), while Brockman focuses on internal aspects (technology, compute, products). He is well known in Silicon Valley for his engineering ethos of personally writing code and overseeing late-night deployments.

Intelligence has become a standardized, resellable commodity, leading to a pathological, relentless growth in demand for compute. OpenAI's business model fundamentally involves buying or leasing compute, transforming it into intelligence via models, and selling it at a premium. Because the demand for problem-solving is infinite, the projected supply of GPUs in 2026 approaches near zero. The importance of this judgment is that AI is no longer just a software service but has evolved into a resource-based commodity business, where the physical world's compute supply directly determines the ceiling of civilization's intelligence.

The scaling law is an empirical truth at the cosmic level, and currently, no "wall" capping it is in sight. Although the basic ideas of neural networks originated in the 1940s, as long as massive amounts of compute are continuously invested, model capabilities will correspondingly and deterministically improve. This advocates for a key viewpoint: technological stagnation is unlikely in the short term. As long as capital and electricity are continually invested, we will obtain more powerful intelligence, providing the underlying logical support for aggressive investment by tech giants.

From a functional perspective, we are already 80% of the way to AGI, as models now possess the closed-loop capability to execute tasks independently. A systems engineer handed a complex optimization problem to a model. The model not only wrote the code but also autonomously ran a profiler and performed multiple rounds of optimization based on feedback until the task was complete. This advocates for a counterintuitive point: AGI is not a future moment but an ongoing process. AI has evolved from a "code-writing assistant" to a "colleague capable of solving problems."

Context is replacing model algorithms as the core frontier of competition. A new tool, Chronicle, can record everything a user does on their computer in real-time, giving the AI "memory" and saving humans the time of repeatedly explaining background to the machine. The importance of this judgment is that for entrepreneurs, one-time model training is not the only moat. Building a "data harness" that allows AI to deeply understand the user's business environment is a truly durable asset.

As the cost of "execution" drops to zero, human attention will become the scarcest resource in the entire economy. When agents can work autonomously, even proactively reporting to managers on Slack when tasks are progressing slowly, human energy