# Sequoia's 'this is AGI' keynote is not a forecast — it's a portfolio decision

URL: https://www.thedeepfeed.ai/posts/2026-05-01-sequoia-ai-ascent-keynote-this-is-agi/
Category: Business
Published: 2026-05-01
Author: the-deep-feed
Tags: sequoia, ai-ascent, agi, agents, services-as-software
Kind: deep

> At AI Ascent 2026, Pat Grady and Sonya Huang argued long-horizon agents are functional AGI. The new claim is the meta-claim: Sequoia is now writing checks as if commercial AGI already arrived.

## TL;DR

- Pat Grady's functional AGI definition — *"if you can dispatch an agent to do a job and it can recover from failure and persist until that job is done, that feels pretty much like AGI"* — was published Jan 14 in the firm's annual essay. The keynote is the deployment, not the announcement.
- The "meter chart" Sonya Huang anchored to is **METR's** task-completion time horizon, doubling every **~7 months** since 2019. METR's extrapolation puts a full work-day in 2028 and a full year in 2034 — Sequoia's portfolio assumes the curve holds.
- Sequoia named **legal services in the US alone at $400 billion** — the same as all of global software — to argue the **$10T** services number is conservative. It is the cleanest TAM math any major firm has put on stage.
- The portfolio receipts: Crosby (legal autopilot), Harvey, Reflection AI, Sierra, Edra, Firetiger, Rogo, Hippocratic AI. Every check assumes the long-horizon-agent thesis works on a customer in production this year.
- The contrary view comes from operators with revenue at risk. Marc Benioff calls AGI talk *"hypnosis"*; Yann LeCun says scaled LLMs will not get there. Sequoia is making the more falsifiable bet, which is also why it is more interesting.

The framing was already on the page. On January 14, 2026, [Pat Grady](https://sequoiacap.com/people/pat-grady/) and Sonya Huang published [*2026: This is AGI*](https://sequoiacap.com/article/2026-this-is-agi). The argument: pre-training gave the world ChatGPT, inference-time compute gave it o1, and *"long-horizon agents are functionally AGI."* The essay closed with a list of named companies operating, in the authors' words, as colleagues. Three months later the thesis showed up on a stage in San Francisco.

![Sequoia's "long-horizon agents are functionally AGI" framing as a meter approaching threshold.](/post-images/2026-05-01-sequoia-ai-ascent-keynote-this-is-agi/hero-meter-curve.jpg)

That stage was AI Ascent 2026, [Sequoia's](https://sequoiacap.com) annual founders' convening. The opening keynote is a known shape: Grady, Huang, and [Konstantine Buhler](https://sequoiacap.com/people/konstantine-buhler/) take ten to fifteen minutes each, in sequence, and try to put a frame around the year. This year's frame was the one already in print. Grady delivered the AGI line on stage, almost word-for-word from the essay:

> If you can dispatch an agent to do a job and it can recover from failure and persist until that job is done, I don't know. That feels pretty much like AGI.
> — Pat Grady, [AI Ascent 2026 keynote](https://youtu.be/LRo33rnv6rQ)

The temptation is to read this as the news. The headlines did. *"Sequoia declares AGI is here"* travels well. The actual news is one layer down. The functional-AGI claim was published in January and re-stated in April. What changed between those two events is the firm's portfolio. More importantly, the frame the firm is using to recommend a strategy to founders has shifted: Sequoia is now writing checks, structuring rounds, and giving advice as if commercial AGI has already arrived. The keynote is the public version of an internal operating assumption that has already shipped to LPs.

Grady's own one-line compression of the talk, posted on X the same day, is the cleanest restatement of that operating assumption:

> Long horizon agents are functionally AGI... and there's $10 Trillion up for grabs
>
> — [@gradypb](https://x.com/gradypb/status/2049895986475245580), April 30, 2026

The two halves of that sentence carry different weight. The first is the headline. The second is the [Buhler $10T number](https://sequoiacap.com/article/10t-ai-revolution/) attached to it as a portfolio mandate.

Read the keynote that way and the specific claims start to matter for different reasons.

## What's in the deck

Pat Grady opens with a slide that has been a Sequoia staple since 2023: silicon → systems → networks → internet → cloud → mobile → AI, each wave additive. He flags three things that make the AI wave structurally different from the last four. *"It's the biggest wave yet."* It is *"the fastest wave yet."* And, in the line he attributes to Buhler, it is the first **revolution of computation** rather than communication. *"The internet, the cloud, mobile — those are all about information distribution. AI is different. AI is about how information is processed."*

The number that does the work in the first five minutes is the TAM:

> Legal services in the US alone is a $400 billion market. That is one vertical and one geo and it's the same as all of software. So this opportunity is immense.
> — Pat Grady

This is the [Bek essay's](https://sequoiacap.com/article/services-the-new-software/) number compressed into a single sentence. The TDF post on [Sequoia's services-as-software thesis](/posts/2026-04-30-sequoia-services-as-software-thesis/) walked through the multiples math underneath: services trade at ~2x revenue, software at 8-20x, and the entire firm-wide thesis is a margin-conversion story aimed at a $6T pool of professional-services spend rather than the $700B software pool. Grady's stage version is the rhetorical compression. *"One vertical, one geo, the same as all of software."* If you accept that single comparison, the [$10T figure](https://sequoiacap.com/article/10t-ai-revolution/) Buhler put on a Sequoia chart in 2024 stops looking aggressive.

Three inflection points then frame why now. ChatGPT in November 2022 (pre-training). o1 in late 2024 (inference-time compute). And, in what Grady names as a *"discontinuous"* break, Claude Code, Opus 4.5, Opus 4.7 in late 2025 and early 2026 (long-horizon agents). It is the same three-act structure as the [Generative AI's Act Two](https://www.sequoiacap.com/article/generative-ai-act-two/) essay from September 2023 and the [Act o1](https://sequoiacap.com/article/generative-ais-act-o1/) update from October 2024, just with the third act named.

The MAD framework is the part of the keynote founders quoted on the way out. Grady's three pillars for application-layer companies: **M**oats (built customer-back, not technology-back, because *"your customers are not changing nearly as fast as the capabilities are changing"*); **A**ffordance (a hammer is an object a two-year-old understands; Claude Code, by contrast, is *"insanely powerful"* but offers little affordance to a Fortune 500 employee); **D**iffusion (the gap between frontier capability and average-enterprise adoption is widening, and that gap *is* the application-layer opportunity).

> The rate at which capabilities are diffusing out into the market is far shy of the rate at which those capabilities are being created. And every day that the foundation models move faster than your average Fortune 500 enterprise, that gap gets bigger and that opportunity gets bigger.
> — Pat Grady

Buhler's "advice in a downpour" frame closes Grady's section: *"You cannot pass 15 cars in the sun, but you can pass 15 cars in the rain."* He uses it to argue no lead is safe. It is also a fair description of how Sequoia is investing: at speed, into application-layer companies in markets where the incumbents have not yet adapted.

## The meter chart, sourced

![METR's task-completion time horizon, the doubling-every-seven-months curve Sonya Huang anchored to.](/post-images/2026-05-01-sequoia-ai-ascent-keynote-this-is-agi/task-horizon-doubling.jpg)

Sonya Huang's section is the one with the most specific numerical claims. The most-quoted is the chart she calls "the meter chart." She defines it cleanly:

> The meter chart measures how long a model can sustain progress on a complex task without going off the rails. We've gone from the order of tens of minutes a year ago to the order of hours today.
> — Sonya Huang

This is not a Sequoia chart. It is [METR's](https://metr.org/) task-completion time horizon, first published in March 2025 by Kwa, West, and colleagues at Model Evaluation & Threat Research. The [paper](https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/) measures the length of task, calibrated to human-expert completion time, at which a frontier model achieves 50% reliability. The headline finding is the doubling time: *"AI performance has been consistently exponentially increasing over the past 6 years, with a doubling time of around 7 months."* METR's [extrapolation](https://metr.org/time-horizons/) puts a model that can reliably complete a full work-day's worth of software tasks in 2028 and a full year in 2034.

The Sequoia 2026 essay [reprinted that exact extrapolation](https://sequoiacap.com/article/2026-this-is-agi). Grady and Huang wrote: *"If we trace out the exponential, agents should be able to work reliably to complete tasks that take human experts a full day by 2028, a full year by 2034, and a full century by 2037."* Huang's stage line that *"whatever you can imagine building over the next hundred years, we think is now possible in 100 days thanks to agents"* is the same curve, run forward. METR is a public, contestable benchmark. So is [Toby Ord's recent paper](https://arxiv.org/abs/2505.05115) modeling the result as a constant-per-minute failure rate. The curve has its critics. What it is not is a Sequoia talking point.

Three concrete examples then anchor the abstraction:

> Nathan from Zed accomplished a three-year moonshot project over the holidays by himself with Claude Code. Brett Taylor rebuilt Sierra over a weekend. The Notion team rewrote 8 million lines of code in just six weeks.
> — Sonya Huang

The Notion stat is the most checkable of the three. [Notion 3.0](https://venturebeat.com/ai/to-scale-agentic-ai-notion-tore-down-its-tech-stack-and-started-fresh/), released in September 2025, shipped on top of a near-complete tech-stack rebuild that the company began in mid-2025. [OpenAI's case study](https://openai.com/index/notion/) on the rebuild describes it as the foundation for "Custom Agents" — Notion's autonomous workflow product. The Sierra and Zed examples are softer; both companies are deep in the weeds of agentic engineering and both founders are public about Claude Code-driven velocity, but neither figure has a single canonical primary source. The Notion number is the one that does the rhetorical work.

## Buhler's industrial-revolution analogy is the wager

Buhler's section is the most speculative and the part that deserves the most scrutiny. His structural claim:

> 99% of all the physical work done on planet Earth for humans is done by a machine. […] We believe that the neural network is the next big wave and that in the near future 99.9% of cognition on planet Earth will be done by machines.
> — Konstantine Buhler

The "industrial revolution for cognition" framing is in the [$10T essay](https://sequoiacap.com/article/10t-ai-revolution/); Buhler has been delivering versions of it on stage since 2023. His four motivating stories (aluminum: intelligence becomes disposable; the NASA evolved antenna: AI-designed objects look alien; Sadi Carnot and thermodynamics: a science of AI is still ahead; and Impressionism: humans respond to mechanical replication by inventing new categories of meaning) are each falsifiable in different ways and on different timescales.

The aluminum story is the relevant one for portfolio construction. *"Aluminum is intelligence. Electrolysis is artificial intelligence. We're about to enter a world where some of the most precious skills that took decades to earn — PhD-level skills — are so instantly invoked that right after using them, you can crumple them up and throw them right in the trash."* That is a pricing claim. If the marginal cost of a PhD-level inference falls toward zero, the price the application layer can defend over time is determined by distribution and customer relationships, not by capability. Which is, again, the argument [Bek made](https://sequoiacap.com/article/services-the-new-software/) in the services-as-software essay. The keynote is internally consistent.

## Sequoia's framing: the four-year arc

Reading the keynote as a deployment of an existing thesis is easier with the prior essays in front of you.

| Year | Sequoia framing | Anchor publication | What was new |
| --- | --- | --- | --- |
| 2023 | "Generative AI's Act Two": the application layer is the prize | [Sept 2023, Grady & Huang](https://www.sequoiacap.com/article/generative-ai-act-two/) | Reframed the field as moving from *"talkers"* to system-2 reasoning |
| 2024 | "Generative AI's Act o1": reasoning is the second scaling axis | [Oct 2024, Grady, Huang, o1](https://sequoiacap.com/article/generative-ais-act-o1/) | Identified inference-time compute as a separate scaling law |
| 2025 | $10T market, services-as-software, AI 50 = "agents move beyond chat" | [Buhler $10T](https://sequoiacap.com/article/10t-ai-revolution/), [AI 50 2025](https://sequoiacap.com/article/ai-50-2025/), [AI Ascent 2025](https://sequoiacap.com/article/ai-ascent-2025/) | Quantified the services TAM and reframed the customer (buyer of labor, not of tooling) |
| 2026 | "Long-horizon agents are functionally AGI"; race begins for $10T | [*This is AGI*](https://sequoiacap.com/article/2026-this-is-agi), [Bek services-as-software](https://sequoiacap.com/article/services-the-new-software/), AI Ascent keynote | Named the third scaling axis (long-horizon agency) and the operating assumption (commercial AGI has arrived) |

The arc is unusually consistent. From 2023 to 2026 the firm has moved from a "what comes after chat" essay to a stage moment that, in Grady's framing, declares the third capability is here and Sequoia is funding accordingly. The 2024 reasoning-models essay set up o1; the 2025 services-as-software essay set up the buyer; the January 2026 *This is AGI* essay set up the keynote. Each step locked in a portfolio bet. Anyone trying to predict where Sequoia's next checks go could have read the essays in order and gotten close.

## The portfolio receipts

![Sequoia's portfolio of long-horizon agent companies, visualized as the agent stack the firm has filled in.](/post-images/2026-05-01-sequoia-ai-ascent-keynote-this-is-agi/portfolio-cluster.jpg)

The keynote's editorial value lives or dies on whether the firm's recent investments match the framing. They do. Sequoia has spent the last eighteen months filling in the long-horizon-agent stack, starting at the model layer and moving up.

| Company | Sector | What it does | Sequoia partnered |
| --- | --- | --- | --- |
| [Reflection AI](https://sequoiacap.com/companies/reflection-ai/) | Foundation / coding | "Superintelligent autonomous systems, starting with coding agents." Asimov is its code-comprehension agent. | [Mar 2025](https://www.sequoiacap.com/article/partnering-with-reflection-toward-superintelligence-with-autonomous-coding/) |
| [Crosby](https://sequoiacap.com/companies/crosby/) | Legal autopilot | An AI-native law firm structured as a firm, not a tool. Contracts in minutes via AI plus lawyers-in-the-loop. | [Jun 2025](https://sequoiacap.com/article/partnering-with-crosby-a-law-firm-at-the-speed-of-ai/) |
| [Harvey](https://www.harvey.ai/) | Legal copilot → autopilot | Now adopted at majority of Am Law 100 and Am Law 200; agentic capabilities are the new pitch | Earlier ([partnership ongoing](https://www.sequoiacap.com/article/partnering-with-harvey-putting-llms-to-work/)) |
| [Sierra](https://sierra.ai/) | Customer-service agents | Bret Taylor's outcome-priced agent platform; the [τ-bench](https://arxiv.org/abs/2406.12045) authors | [Feb 2024](https://www.sequoiacap.com/companies/sierra/) |
| [Edra](https://sequoiacap.com/companies/edra/) | Context for agents | Reverse-engineers enterprise SOPs from ServiceNow, Zendesk, Outlook so agents follow real processes | 2024 |
| [Firetiger](https://sequoiacap.com/companies/firetiger/) | Ops layer for coding agents | Production observability + business context; steers coding agents to fix issues autonomously | 2025 |
| [Rogo](https://sequoiacap.com/companies/rogo/) | Finance agents | Sourcing, diligence, modeling agents for Wall Street | 2025 |
| [Agency](https://www.sequoiacap.com/companies/agency/) | Customer-success agent | AI agent for CS teams (Elias Torres) | 2024 |
| Hippocratic AI | Healthcare agents | Clinically-safe generative-AI healthcare agents; [Series C $126M Nov 2025](https://hippocraticai.com/hippocratic-ai-announces-series-c-funding-126-million/) | Multiple rounds |
| Glean | Enterprise work agents | Agentic search and workflow over enterprise data; [Series F at $7.2B in 2026](https://news.crunchbase.com/venture/ai-powered-work-assistant-glean-valuation-jumps/) | Earlier rounds |

Two things stand out. First, Sequoia has a check at every layer of the agent stack: labs (Reflection), harnesses and operations (Firetiger), context (Edra), and vertical autopilots (Crosby, Harvey, Rogo, Hippocratic, Sierra). Second, the firm has notably [not](https://techcrunch.com/2025/05/04/cursor-is-reportedly-raising-funds-at-9-billion-valuation-from-thrive-a16z-and-accel/) chased the most-photographed bet of the cycle; Anysphere/Cursor went to Thrive, a16z, and Accel. The portfolio favors structural agent-shaped companies over the marquee app names. That is consistent with the *"customer-back, not capability-back"* advice on stage; it is harder to credibly argue a coding-tool-priced-per-seat is the future of work if the keynote also argues *"services is the new software."*

The shape that connects most of the names (Crosby, Harvey, Sierra, Hippocratic, Rogo) is the [services-as-software](/posts/2026-04-30-sequoia-services-as-software-thesis/) one. These are companies that sell the work, not the tool, and that price per outcome rather than per seat. Three of them ([Sequoia's Crosby write-up](https://sequoiacap.com/article/partnering-with-crosby-a-law-firm-at-the-speed-of-ai/), the [Bek essay](https://sequoiacap.com/article/services-the-new-software/), and the keynote) say so explicitly. Huang's framing on stage was unsentimental:

> Hiring agents is so much easier than hiring employees. Humans are hard to scale; agents are infinitely scalable with compute. Humans are hard to keep happy; agents are low-maintenance. Humans are expensive — you pay them salaries; you pay agents tokens. Generally, it costs less to accomplish a task with tokens than the equivalent in salary.
> — Sonya Huang

The check sizes follow that logic. Crosby is a $5.8M seed [structured as an actual law firm](https://techcrunch.com/2025/06/17/sequoia-backed-crosby-launches-a-new-kind-of-ai-powered-law-firm/), not a SaaS company; Crosby's [first revenue line is contract negotiation, billed by deliverable](https://crosby.ai/blog/series-b-planting-our-flag). Reflection AI raised $130M to build a model lab whose first product is a [code-comprehension agent rather than a code-generation tool](https://reflection.ai/), explicitly differentiating from Cursor's positioning. Both checks make sense only if the firm believes the agent layer captures more durable value than the IDE seat does.

## Where the keynote is wrong, or at least vulnerable

![The bull-bear split between Sequoia's "this is AGI" framing and the operator critique from Benioff and LeCun.](/post-images/2026-05-01-sequoia-ai-ascent-keynote-this-is-agi/split-takes.jpg)

The strongest critique of the *"functional AGI"* framing is the operator critique. Salesforce CEO Marc Benioff has been the most public on this. In an [August 2025 interview](https://www.businessinsider.com/marc-benioff-extremely-suspect-agi-hypnosis-2025-8), Benioff said he was *"extremely suspect"* of those *"buying into AGI hypnosis"*; on a [later podcast](https://www.theloganbartlettshow.com/archive/ep-127-marc-benioff-ceo-salesforce-strikes-back-at-satya-agi-is-not-here) he said simply, *"AGI is not here."* Benioff's incentive is obvious (Salesforce sells per-seat enterprise SaaS), but his observation about reliability is the empirical one. Today's agents [hallucinate, lose context, and "charge confidently down exactly the wrong path"](https://sequoiacap.com/article/2026-this-is-agi). Even Sequoia's own essay concedes that. The disagreement is about whether that gap closes faster than enterprise adoption proceeds.

Yann LeCun's structural critique is the longer-horizon one. Meta's chief AI scientist has been [consistent since at least January 2025](https://www.pymnts.com/artificial-intelligence-2/2025/meta-large-language-models-will-not-get-to-human-level-intelligence/) that scaled LLMs alone will not get to human-level intelligence; he has argued that world models, not autoregressive token prediction, are the path forward. If LeCun is right, the METR curve eventually slows even though it has not yet, and "long-horizon agents" hit a ceiling before they reach the year-long task that the extrapolation puts at 2034. Sequoia's wager is that the curve holds long enough for their 2025-2026 portfolio to clear an exit.

There is also a more cynical reading worth holding next to the bull case. Sequoia, perhaps more than any other top-tier firm, has spent the last three years branding itself as the AI thesis house: the [AI 50 list](https://sequoiacap.com/article/ai-50-2025/), the [Training Data podcast](https://sequoiacap.com/podcasts/), the annual essay rhythm, the [Inference newsletter](https://inferencebysequoia.substack.com/). A firm whose marketing motion runs on calling the next thing has a structural reason to call AGI early. The question is whether the portfolio backs the rhetoric, and on the evidence of Crosby, Reflection, Edra, Sierra, and Firetiger, it does. This is not a press release thesis; the checks have cleared.

## What founders should take from this

Strip the AGI label off the keynote and the operating advice is the same.

If long-horizon agents work, and Sequoia's bet is that they already do, then the durable position at the application layer is **owner of the customer relationship that lets you bill for the work**, not owner of the workflow tool that lets the human practitioner do it slightly faster. That's the MAD framework's M and the Bek essay's autopilot. If the agents do not yet work in your domain, the durable position is the **affordance layer**: the path-of-least-resistance UX that turns a Claude Code-shaped raw capability into a thing the average enterprise employee will actually use. That's the MAD framework's A. The diffusion gap between frontier capability and enterprise adoption is wider, not narrower, in 2026 than it was in 2024, so the time available to grab those positions is also wider, not shorter.

Two of the things Sequoia did *not* say on stage are worth flagging. The keynote did not name a single regulatory question (licensure, malpractice liability, evidentiary admissibility for AI-generated legal work), even though three of its named verticals (legal, medicine, finance) are some of the most-regulated industries in the economy. The TDF [services-as-software piece](/posts/2026-04-30-sequoia-services-as-software-thesis/) treated this as the unfinished part of the thesis; the keynote skips it. The keynote also does not engage with the [pricing-floor problem](https://linas.substack.com/p/sequoiathesis): if marginal inference cost on a PhD-grade task falls toward $0.03 and five startups in each vertical are competing on price, the software-multiple math collapses. Founders should hold both questions next to the slide deck.

The companion talks at AI Ascent 2026 ([Andrej Karpathy on Software 3.0](/posts/2026-05-01-karpathy-software-3-agentic-engineering/), [Demis Hassabis on AGI timelines](/posts/2026-05-01-hassabis-2030-agi-disease-compression/), [Greg Brockman on the human attention bottleneck](/posts/2026-05-01-brockman-human-attention-bottleneck/), and [Jim Fan on robotics](/posts/2026-05-01-jim-fan-robotics-llm-playbook/)) all give versions of the same instrumental claim Huang and Grady make: agents are crossing a usable-reliability threshold, and the binding constraint moves from capability to *how a human knows when to trust the output*. Karpathy calls it verifiability; Brockman calls it human attention; Hassabis calls it grounding. Sequoia is the one that priced it.

## What Sequoia is saying

![Sequoia operating "as if" commercial AGI has already arrived — a portfolio standing on the far side of the line.](/post-images/2026-05-01-sequoia-ai-ascent-keynote-this-is-agi/agi-position.jpg)

The honest read of the keynote is that the AGI-label headline is the least interesting thing in it. The *"if you can dispatch an agent and it persists"* line will get clipped, retweeted, debated by people who do not work on agents, and then forgotten. The actual content (METR's curve as the metric for long-horizon agency, the $400B legal services anchor for the $10T number, the MAD framework, the "anchor to customers, not capabilities" advice) is operating doctrine for application-layer founders.

The framing TDF will keep using is this: Sequoia has not declared AGI. Sequoia has decided to *operate as if* commercial AGI has arrived in the specific sense Pat Grady defined: agents that recover from failure and persist toward a goal. The portfolio is the proof the firm believes its own essay. The keynote is the public deployment of an operating assumption that has been driving check-writing since at least the [Reflection AI partnership in March 2025](https://www.sequoiacap.com/article/partnering-with-reflection-toward-superintelligence-with-autonomous-coding/).

If that operating assumption is wrong (if METR's curve breaks, if regulators move faster than Bek's essay assumes, if Benioff is right and SaaS keeps the labor budget for another decade), Sequoia's 2025-2026 vintage is the most exposed in the firm's modern history. If it is right, this vintage is also the firm's best shot at the next [trillion-dollar outcome](https://www.youtube.com/watch?v=v9JBMnxuPX8). Both possibilities live inside the same set of checks. That is what makes the keynote worth watching even for people who do not think AGI is a useful word. The label is the marketing. The portfolio is the falsifiable claim.

The line that should travel from this keynote is not Grady's AGI definition. It's Huang's:

> Whatever you can imagine building over the next hundred years, we think is now possible in 100 days thanks to agents.
> — Sonya Huang

That is the falsifiable forecast. By this time next year, either the AI Ascent 2026 portfolio has produced a service-replacing autopilot in at least one regulated vertical, or it hasn't. Sequoia just told the room (and its LPs) to watch which one happens.

## Sources

- [Sequoia Capital — AI Ascent 2026 keynote (Sonya Huang & Pat Grady, YouTube)](https://youtu.be/LRo33rnv6rQ)
- [Sequoia Capital — 2026: This is AGI (Pat Grady & Sonya Huang, Jan 14, 2026)](https://sequoiacap.com/article/2026-this-is-agi)
- [Sequoia Capital — Generative AI's Act Two (Pat Grady & Sonya Huang, Sept 2023)](https://www.sequoiacap.com/article/generative-ai-act-two/)
- [Sequoia Capital — Generative AI's Act o1 (Pat Grady, Sonya Huang, Oct 2024)](https://sequoiacap.com/article/generative-ais-act-o1/)
- [Sequoia Capital — Services: The New Software (Julien Bek)](https://sequoiacap.com/article/services-the-new-software/)
- [Sequoia Capital — The $10 Trillion AI Revolution (Konstantine Buhler)](https://sequoiacap.com/article/10t-ai-revolution/)
- [Sequoia Capital — AI 50 2025: AI Agents Move Beyond Chat](https://sequoiacap.com/article/ai-50-2025/)
- [Sequoia Capital — AI Ascent 2025 recap](https://sequoiacap.com/article/ai-ascent-2025/)
- [METR — Measuring AI Ability to Complete Long Tasks (Kwa, West et al., Mar 2025)](https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/)
- [METR — Task-Completion Time Horizons of Frontier AI Models (live tracker)](https://metr.org/time-horizons/)
- [Sequoia Capital — Partnering with Crosby (Jun 2025)](https://sequoiacap.com/article/partnering-with-crosby-a-law-firm-at-the-speed-of-ai/)
- [Sequoia Capital — Partnering with Reflection AI (Mar 2025)](https://www.sequoiacap.com/article/partnering-with-reflection-toward-superintelligence-with-autonomous-coding/)
- [Sequoia Capital — Reflection AI launches Asimov (Jul 2025)](https://sequoiacap.com/article/reflection-ai-asimov/)
- [Sequoia Capital — Partnering with Edra](https://sequoiacap.com/article/partnering-with-edra-context-for-agents-at-scale/)
- [Sequoia Capital — Partnering with Sierra (Bret Taylor)](https://www.sequoiacap.com/companies/sierra/)
- [Sequoia Capital — Firetiger company page](https://sequoiacap.com/companies/firetiger/)
- [TechCrunch — Sequoia-backed Crosby launches a new kind of AI-powered law firm (Jun 2025)](https://techcrunch.com/2025/06/17/sequoia-backed-crosby-launches-a-new-kind-of-ai-powered-law-firm/)
- [OpenAI — Notion's GPT-5 rebuild unlocks autonomous AI workflows (Nov 2025)](https://openai.com/index/notion/)
- [VentureBeat — To scale agentic AI, Notion tore down its tech stack and started fresh (Oct 2025)](https://venturebeat.com/ai/to-scale-agentic-ai-notion-tore-down-its-tech-stack-and-started-fresh/)
- [Business Insider — Marc Benioff 'extremely suspect' of AGI 'hypnosis' (Aug 2025)](https://www.businessinsider.com/marc-benioff-extremely-suspect-agi-hypnosis-2025-8)
- [The Logan Bartlett Show — Marc Benioff: AGI Is Not Here](https://www.theloganbartlettshow.com/archive/ep-127-marc-benioff-ceo-salesforce-strikes-back-at-satya-agi-is-not-here)
- [PYMNTS — Yann LeCun: scaling LLMs will not reach human-level intelligence (Jan 2025)](https://www.pymnts.com/artificial-intelligence-2/2025/meta-large-language-models-will-not-get-to-human-level-intelligence/)
- [Sonya Huang — AI Ascent 2026 talk thread (X, Apr 30, 2026)](https://x.com/sonyatweetybird/status/2049890826856391156)
- [Pat Grady — 'Long horizon agents are functionally AGI' (X, Apr 30, 2026)](https://x.com/gradypb/status/2049895986475245580)

---

Canonical: https://www.thedeepfeed.ai/posts/2026-05-01-sequoia-ai-ascent-keynote-this-is-agi/
Site: https://www.thedeepfeed.ai
Full corpus: https://www.thedeepfeed.ai/llms-full.txt