“AGI is here… now.” With that phrase, Sequoia Capital announced this week — one of Silicon Valley’s most storied venture capital firms and a major investor in OpenAI — that we have crossed the threshold into artificial general intelligence (AGI).

In its post, the firm stated, plainly and explicitly, that it was “not bogged down by details at all.” When Sequoia speaks, the tech world listens. The claim dominated discussions across the AI developer community for days.

As someone who is simultaneously a developer, a venture capitalist, and an AI researcher, I see this declaration as deeply useful in one sense — and deeply dangerous in another.

What Is Useful About Sequoia’s Argument?

Sequoia offers a practical definition of AGI: “the ability to discover solutions. Nothing more.” Under this framing, AI systems today can search vast bodies of information, determine a course of action, and then execute it. The core shift, according to Sequoia, is that AI has moved from “talking” to “doing.”

The firm points to concrete examples. It says platforms like Harvey and Legora “act as legal associates,” Juicebox “acts as a recruiter,” and OpenEvidence’s Deep Consult “acts as a specialist.” These are literal descriptions. While I am skeptical of this conceptual framing — more on that shortly — the provocation itself matters.

What Sequoia is doing here is directly challenging developers, and that is important. AI systems can already review contracts clause by clause and engage meaningfully with prospective customers in real time. This is a reminder that we need to think bigger about what is now possible, and that the frontier has expanded dramatically in just a single year.

I sent Sequoia’s post to my co-founders not to debate philosophy, but to push us to rethink the “execution versus conversation” framework it proposed. We need to rise to that challenge.

But Why Is Calling These Systems AGI Dangerous?

Labeling these systems as “artificial general intelligence” causes real harm — both to the credibility of the AI revolution and to the safe deployment of these technologies. It obscures what so-called AI agents can actually do today — and they are certainly not general superintelligence — while offering no guidance on how humans should interact with them. The short answer: do not trust them blindly.

Three examples illustrate these limitations clearly.

First: AI Systems Fail Outside Their Training Distribution

I addressed this in a previous article, but the Greenland crisis provides a live, evolving example. I tested whether generative AI tools — including ChatGPT 5.2 with maximum “reasoning and research” enabled — could analyze this rapidly developing geopolitical event. If these systems are truly AGI, could they help me understand what was happening?

The answer was no. They could not even conceive that the events were possible.

I presented screenshots from Wikipedia documenting the crisis. Every model told me the story was fabricated, “nonsense,” and impossible. When I continued pressing, citing real news sources, ChatGPT repeatedly urged me to “calm down,” insisting that “this is not a real crisis.”

These models are so tightly anchored to traditional Western alliance frameworks that they cannot generate context that contradicts their training data — even when confronted with primary sources. When reality moves outside their training distribution, AI “reasoning” collapses. Instead of expressing uncertainty, the system confidently misleads the user and continues reasoning while wrong. If policymakers or politicians are relying on these tools to understand Greenland right now, that is a genuine risk.

Second: AI Systems Reflect the Beliefs of Their Builders

A study published in Nature two weeks ago made this explicit. Researchers found that large language models reflect the political ideologies of their developers. Chinese models were strongly positive toward China, while Western models were clearly negative.

Even within Western models, bias is evident. Grok, developed by Elon Musk’s xAI, showed negative bias toward the European Union and multiculturalism, reflecting a right-leaning agenda. Google’s Gemini, widely seen as more liberal, was more positive toward both.

This is now widely accepted within the AI community: language models reflect the ideology of the labs that build them. So how can we trust that an “agent” with a supposed blank slate can neutrally “discover solutions,” especially when analyzing complex, large-scale data?

Declaring the existence of AGI implicitly assumes neutrality — or at least gestures toward it — while the evidence points in the opposite direction.

Third: Deterministic Systems Versus Non-Deterministic Systems

Generative AI is inherently non-deterministic. The same input can produce slightly different outputs, or radically different ones.

Humans intuitively understand what should be deterministic and what can be creative. Your shirt size when ordering online is deterministic; choosing a pattern or color is subjective. Even the most advanced models still confuse these categories constantly. We have all seen generative AI treat hard facts as if they were creative suggestions.

This exposes a critical gap in metacognition — awareness of the thinking process itself. Without the ability to distinguish between what must be fixed and what can be generative, AI cannot reliably “discover solutions.”

So What Should We Do?

We have clear tools available.

First, choose narrow, well-defined use cases where bias and out-of-distribution failure are less likely.

Second, provide AI systems with full, customized, real-world context rather than letting agents operate in a vacuum. As I have written before, context is king for AI agents. It also clarifies what must be deterministic and what can be generative.

Third, deploy rule-based filters and supervisory agents that trigger human review when necessary.

Finally, we must acknowledge a core reality: large language models will always reflect their training data and the ideologies of their creators. These models — and their developers — are political actors, whether they intend to be or not. AI should therefore remain under the control of individual human users, not imposed on people as an opaque system. Traceability and accountability are essential — the ability to trace every decision back to a human, no matter how many intermediate steps exist — to ensure governance and safety.

Ultimately, I do not care much what we call these technologies — as long as we do not call them AGI. What we have today is extraordinarily powerful AI, capable of speaking and executing effectively within narrow, well-defined domains. With strict safeguards, deterministic filters, and human-in-the-loop systems, these tools can add trillions of dollars to the global economy.

Call it narrow AI. That is where the trillion-dollar opportunity actually lies today.

US stocks fell on Friday, putting Wall Street’s main indexes on track for a second consecutive weekly loss, as shares of Intel slumped sharply following weak guidance, while ongoing geopolitical tensions continued to weigh on investor risk appetite.

Stocks had rebounded over the previous two sessions after a sharp sell-off on Tuesday, triggered by threats from US President Donald Trump to impose tariffs on European allies unless Washington was allowed to purchase Greenland.

Trump later softened his rhetoric on tariffs and ruled out using force to take control of Greenland. Even so, the S&P 500, the Nasdaq, and the Dow Jones Industrial Average remained on course to end the week lower. At the same time, flows into safe-haven assets persisted, pushing gold prices to a new record high.

The biggest drag on markets on Friday came from chipmaker Intel, whose shares plunged 14.9% after the company forecast quarterly revenue and earnings below market expectations, citing difficulties in meeting demand for server chips used in artificial intelligence data centers. Despite the sharp drop, Intel shares were still up about 50% since the start of the year.

The Philadelphia Semiconductor Index fell 1.6%, pulling back from the record high reached in the previous session, while Wall Street’s volatility index, the VIX, known as the market’s fear gauge, rose after declining over the prior two sessions.

Peter Cardillo, chief economist at Spartan Capital Securities, said: “Earnings season has been good, but one or two stocks have issued less optimistic guidance and sold off accordingly as investors reposition. Guidance has now become more important than ever.”

He added: “Investors will remain cautious because we’re not just watching earnings, we’re also focused on the Federal Reserve. We don’t expect a policy change, but the question is what the Fed will say in its statement.”

By 9:48 a.m. Eastern Time, the Dow Jones Industrial Average was down 320.71 points, or 0.65%, at 49,063.30. The S&P 500 fell 14.68 points, or 0.21%, to 6,898.78, while the Nasdaq Composite slipped 36.50 points, or 0.16%, to 23,399.52.

Anticipation of the Federal Reserve decision

The Federal Reserve is widely expected to keep interest rates unchanged in the 3.5% to 3.75% range at its meeting next week. Investors will scrutinize the policy statement and comments from Chair Jerome Powell for clues about the next move. According to the CME FedWatch Tool, markets are pricing in the first rate cut in June.

Preliminary data from S&P Global showed US business activity remained steady in January, as an improvement in new orders offset weakness in the labor market.

Several members of the “Magnificent Seven,” including Apple, Tesla, and Microsoft, are set to report earnings next week. Their outlooks will be closely watched to assess whether the growth narratives supporting their elevated valuations remain intact.

Supported by the strength of the US economy and expectations for interest rate cuts later this year, market gains had broadened beyond mega-cap stocks into other sectors. Both the Russell 2000 small-cap index and the Dow Jones Transportation Average hit record highs on Thursday.

In other moves, shares of Nvidia rose 1.4% after Bloomberg reported that Chinese officials told companies including Alibaba, Tencent, and ByteDance to prepare for potential purchases of Nvidia’s H200 AI chips.

US-listed mining stocks such as Hecla Mining and Coeur Mining also edged higher by 0.6% and 0.3%, respectively, as silver prices climbed to record levels and approached the $100-per-ounce mark for the first time.

Silver has a long history of extraordinary price movements, and the latest surge is undoubtedly one of the most notable episodes. Since breaking above the $50 level in late November, prices have followed a sharply rising, near-parabolic path, with little meaningful pause along the way.

Before that, silver had already been climbing steadily, trading at around $23 at the time of Donald Trump’s election to a second presidential term. A combination of industrial demand, constrained mine supply, and monetary demand played a decisive role in this remarkable rally. The most recent phase of the advance, however, has been driven by heavy participation from retail investors, as silver turned into something of an online “trending phenomenon.”

Naturally, some profit-taking can be expected at these levels. Still, it is difficult to bet against precious metals before gold itself reaches the $5,000 mark. Gold’s intraday high earlier today stood at $4,967, and it is currently trading only about $8 below that level.

Silver has always been characterized by sharp price volatility, driven by its dual role as both an industrial commodity and a monetary store of value. The most famous episode in its history remains the attempt by the Hunt brothers to corner the silver market in 1979 and 1980. Motivated by fears of inflation and currency debasement, Nelson and William Hunt accumulated vast quantities of physical silver and futures contracts.

By early 1980, the Hunt brothers controlled roughly one-third of the world’s freely tradable silver supply. Intense buying pressure pushed prices from around $6 to a historic peak near $50 per ounce in January 1980. The bubble burst after exchanges imposed new margin restrictions, triggering what became known as “Silver Thursday,” a market collapse that wiped out much of the Hunt family’s fortune.

Three decades later, silver experienced another major rally in 2011. Following the 2008 global financial crisis, quantitative easing policies and a weaker US dollar drove investors toward hard assets. Silver rose steadily and came close to its 1980 high, reaching around $49 in April 2011, before undergoing a sharp correction after margin requirements were raised again. That rally is widely thought to have been amplified by the emergence of silver-backed exchange-traded funds.

More recently, the “silver squeeze” phenomenon in early 2021 highlighted the growing influence of social media on financial markets. Inspired by the saga surrounding GameStop, retail investors on Reddit attempted to force a squeeze on institutions they believed were artificially suppressing silver prices. While they succeeded in boosting demand for physical silver and ETFs, pushing prices to an eight-year high near $30, the sheer size and liquidity of the global silver market absorbed the shock and prevented a repeat of the Hunt-era scenario.

Today, retail investors are once again trying their luck. The idea has circulated across various corners of the internet for some time, and it is striking — and even enjoyable — to see the upward trend delivering substantial gains and rewarding those who positioned themselves early.

Palladium prices rose on Friday amid positive expectations for continued gains in the industrial metal and stronger investment inflows.

UBS said in a note to clients on Friday that it had raised its palladium price forecast by $300 per ounce to $1,800, citing a sharp increase in investment flows into the metal.

Analyst Giovanni Staunovo said UBS made the revision “driven by the strength of investment demand in recent months,” adding that the relatively small size of the palladium market “often leads to sharp price swings.”

The bank explained that the recent price momentum was not driven by traditional industrial uses, but rather by investor positioning in anticipation of lower US interest rates, a weaker dollar, and rising geopolitical uncertainty.

Staunovo noted that “if investment demand remains strong, prices could rise further,” but warned that “in the absence of investment demand, we see the market as largely balanced,” which explains UBS’s preference for exposure to gold.

Demand for palladium has shifted in recent years after its use in automotive catalytic converters peaked in 2019, the same year prices surged above platinum, triggering substitution away from the metal.

The spread of electric vehicles, which do not use catalytic converters, has also weighed on palladium demand.

However, the bank said palladium has risen alongside platinum and silver since mid-2025, and with palladium now “significantly cheaper than platinum,” UBS expects catalytic converter manufacturers to “switch back to using it… in due course.”

Investment activity in palladium has picked up notably, with UBS pointing to rising exchange-traded fund holdings since mid-2025, alongside a sharp increase in speculative positions in the futures market, after being net short for most of last year.

China may also be supportive of demand. Staunovo said the launch of yuan-denominated platinum futures in Guangzhou “is likely to have supported demand for palladium,” as part of broader trading activity across the platinum group metals.

In trading, March palladium futures jumped 4.1% to $2,007 per ounce by 14:45 GMT.