Author: Market Participant; Translator: Deep Tide TechFlow
As the new wave of AI agent fever sparked by OpenClaw and Claude Code swept across social media, the author keenly sensed a frenzy reminiscent of the NFT era of 2021.
This article analyzes how social media amplifies the technology narrative, how Wall Street indiscriminately sells off due to the "AI kills software" bias, and why giants like Salesforce and ServiceNow are still unfairly punished by the market despite delivering impressive results.
The author believes we are in the "mid-game" of a great revolution, and all the extreme optimism and extreme panic are attempts to prematurely exhaust the yet-to-arrive endgame.
The author argues that we are in the "mid-game" of a great revolution, and all the extreme optimism and extreme panic are attempts to prematurely exhaust the yet-to-arrive endgame.
The full text is as follows: This wave of enthusiasm for OpenClaw and Claude Code reminds me of the hysteria of the NFT era. The emergence of new technologies is accompanied by practicality, while simultaneously generating cultural and narrative resonance within the zeitgeist. Like every technology that captures the collective imagination at the right time, it is being processed through the same "twisting machine"—the same machine that once transformed JPEG images of monkeys into a $40 billion asset class. The pattern is identical: genuine innovation arrives, and early adopters discover real value. Then, the social layer takes over—suddenly, the conversation detaches from the technology itself and becomes a performance about "taking sides." Declaring "this is the future" has become a hallmark of insiders. Writing guidelines, deep thinking (think pieces), and exaggerating the value of the status quo garners social approval. The compounding effect of opinions even outpaces the technology itself. (I promise, there will be an opinion on financial markets later.) The Cognitive Distortion Machine X makes things worse. Social media is increasingly seen as a legitimate lens through which reality is viewed, and it distorts the image of facts. The loudest voices are not representative—they are performing "unwavering conviction" for an audience that rewards this behavior. Every major platform operates on engagement, and engagement rewards extremes. "This is fun and useful" won't spread widely; "This changes everything, you're going to lose your job" will. A hundred retweets saying "This changes everything" are not signals, but echoes. Echoes are mistaken for consensus, consensus is mistaken for truth, and truth is mistaken for investable theories. Girard would be thrilled to see this. When enough people put on a “belief show” for a result, the show itself is confused with evidence supporting that result. The NFT era proved this perfectly: people didn’t want JPEGs, they wanted “what everyone else wants” [1]. What is real? The latest modeling capabilities are amazing—far more amazing than NFTs, which have little practical use beyond speculation and cultural signals. I use these tools every day. They improve my efficiency in concrete, measurable ways. The underlying models are truly impressive, and the trajectory of improvement is very steep. The increment is enormous when I compare what I could do with these tools six months ago to what I can do today. Moreover, the broader potential is limitless. AI-assisted programming, research, analysis, writing—these aren't hypothetical use cases; they're happening, creating real value for those who use them effectively. I don't want to be the person who scorned the internet in 1998. That's not the point; I'm very bullish on AI in the long term. The point is the timeline and the gap between potential and reality. What isn't real yet? No—Claude won't immediately catalyze social upheaval. This doesn't mean humans no longer need interfaces to manage work. Nor does it mean Anthropic has won the AI war. Consider what those most breathtaking arguments actually require you to believe: that enterprise software—decades of accumulated workflows, integrations, compliance frameworks, and institutional knowledge—will be replaced in quarters, not years? That the per-seat billing model will die overnight? Would a company with annual revenue exceeding $10 billion and a gross margin of 80% simply vanish because a chatbot can write a function? [2] Dan Ives of Wedbush bluntly stated: "Companies will not completely overturn their past investments of hundreds of billions of dollars in software infrastructure in order to migrate to companies like Anthropic and OpenAI." [3] Jensen Huang, who has more reason than anyone to advocate the disruptive power of AI, called the concept of "AI replacing software" "the most illogical thing in the world." [4] Those who most actively proclaim "Endgame" (thanks to @WillManidis for popularizing the term) are often the ones who benefit the most from your "unwavering belief": followers, inquiries, subscription fees, conference invitations. Incentive structures reward bold predictions that bear no responsibility for timing. What's interesting to me is that the market is making the same mistakes on the other side of the table. On January 30, Anthropic released its Claude Cowork plugin, and in less than a week, $285 billion of software, financial services, and asset management stocks evaporated. [5] The software ETF—$IGV—is down 22% this year, while the S&P 500 is up. 100 of the 110 constituent stocks are losing money. The RSI index hit 16, the lowest reading since September 2001. [6] Hedge funds are aggressively shorting software stocks and continuing to add to their positions. [7] The narrative is that AI is killing SaaS (Software as a Service). Every software company that charges per seat is a “walking dead”. This sell-off is indiscriminate. Companies with completely different risk profiles due to AI are all treated as the same trading benchmark [8]. When 100 out of 110 names in the index are falling, the market has stopped analyzing and is indulging in the climax of the narrative. Note: The recovery may have begun since I started writing this article. Throwing out the bathwater also means losing the baby. Let’s see what’s really going on inside those companies that are thought to be facing annihilation. Salesforce’s Agentforce revenue grew 330% year-over-year, with annualized revenue exceeding $500 million and generating $12.4 billion in free cash flow. The forward P/E ratio is 15. They just announced a revenue target of $60 billion for fiscal year 2030. [9] This isn't a company being disrupted by AI—it's a company building an enterprise delivery layer for AI. ServiceNow's subscription business grew by 21%, its operating margin expanded to 31%, and it authorized $5 billion in stock buybacks. Their AI suite, Now Assist, has an annual contract value (ACV) of $600 million and aims to break $1 billion by the end of the year. [10] However, its stock price has fallen 50% from its peak. Should these names be moderately downgraded because of the risks? Perhaps. But smart people started pricing this out years ago. As many smarter people than me have pointed out: this sell-off requires you to believe simultaneously that "AI capital expenditure is collapsing" and "AI is powerful enough to destroy the entire software industry." [11] These two things cannot both be true at the same time. Choose one.
Identifying the real risks
Will some companies be truly replaced? Yes.
Point solutions that provide standardized, single workflows are vulnerable. If your entire product is just an interface layer built on non-proprietary data, you're in trouble. LegalZoom fell 20%—for this type of company, the concern is real [12]. When AI plugins can automatically perform contract review and confidentiality agreement (NDA) classification, the value proposition of paying traditional vendors for the same functionality becomes hard to defend.
But companies with deep integration, proprietary data, and platform-level foundations are a completely different story. Salesforce has penetrated the technology stack of every Fortune 500 company. ServiceNow is the system of record for enterprise IT.
Datadog's consumption-based model means that more AI computing directly translates into more monitoring revenue—their non-AI business growth actually accelerated to 20% year-over-year [13]. Selling digital infrastructure because "AI kills software" is as absurd as selling construction equipment stocks because buildings are springing up. We've been through this before. The SaaS crash of 2022 was instructive. The sector fell by more than 50%. The median forward revenue multiple dropped from 25x to 7x—below pre-pandemic levels [14]. Meanwhile, earnings were strong. The subsequent rebound was remarkable—the Nasdaq rose 43% in 2023. Admittedly, the trigger at the time was more interest rate shocks than fundamental deterioration. The DeepSeek panic of January 2025 is closer. Nvidia crashed due to fears that cheap Chinese AI models would render the entire AI infrastructure development meaningless, but subsequently recovered completely.[15] That fear was structurally similar to today's: a single product launch triggered an existential crisis reassessment of the entire industry.[16] Many observers have drawn a direct analogy to the early stages of the dot-com bubble burst—tech stocks fell while consumer staples, utilities, and healthcare stocks rose.[17] But there is one thing about the dot-com bubble burst: Amazon fell 94% and then became one of the most important companies in the world. The market was trying to price in the "endgame" halfway through the game, creating one of the greatest buying opportunities in history. Jim Reid of Deutsche Bank made a very true statement: “Identifying long-term winners and losers at this stage is almost pure guesswork.” [17] I bet he’s right. And this uncertainty—the admission that we don’t know how it will end—is what makes this indiscriminate selling a mistake. The Endgame Fallacy The speculators on X and the panic sellers on Wall Street made the same mistake on both sides of the chessboard. One group said that AI has won, the future is here, and all institutions and job functions should be rewritten from now on. Another group said that AI has killed software, subscription revenue is dead, $10 billion in free cash flow doesn’t matter because the business model is obsolete. Both sides jumped to the "endgame" when the game still had many moves to go. The gap between our present state and the technological vision will be filled by chaotic, incremental, company-specific progress. Some software companies will integrate AI and become more powerful; a few will actually be replaced; most will adapt—a slow, uneven process, not suitable for tweeting. The actual trajectory is more volatile and uncertain than hype or panic suggests. Those who will do well from now on will be those who can tolerate this ambiguity, not those who rush to grasp a premature narrative. Great managers always find a way.