Opinion: Will ChatGPT and Claude wipe out every single job?

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯5806 words

a16z partner Joe Schmidt IV points out that large model labs will only dominate horizontal tasks, while the real AI application opportunities are hidden in vertical scenarios and complex workflows. (Context: Altman retracts "AI will destroy human jobs" prediction: I'm glad I was wrong, really? ) (Background: Google leads investment in AI routing platform OpenRouter, valuation $1.3 billion, 240% growth in one year) Operators and potential employees keep asking me the same question: Is there any room left to build in the AI application layer? Or will OpenAI and Anthropic kill everything? Behind this question hides a specific type of "AI anxiety." Some have concluded that to avoid becoming a permanent underclass, the only place to have a lasting foothold is either inside large labs or in frontier fields like robotics and hard tech—theoretically, "anything the labs can't touch." If every piece of software is about to be swallowed, whether by having your job directly replaced by Codex or Claude, or by having everything you build rendered unnecessary by future models, then run! Listen, I am an AI maximalist like almost everyone else, but I think they are only half right. Labs are indeed eroding a large portion of the application landscape. But the "application layer" is not a single, homogeneous opportunity. The correct mental framework is: Are you on the "Yellow Brick Road" that the labs are traveling, or are you somewhere else in the land of Oz? The Yellow Brick Road is our shorthand for the path the labs are on, where they are pouring staggering resources. The reason labs are best suited to solve problems like code generation, writing, or image creation is that these problems improve as "raw model capability" increases: every dollar spent on pre-training and post-training directly improves product quality. Meanwhile, the rest of Oz is full of more complex, often vertical problems. These problems are not as simple as giving enterprise users a "general-purpose tool" with standard tools and computer operation permissions. Their value comes less from the raw capability of the underlying model (though this remains important!) and more from the scaffolding built around it—the very architecture that makes the output trustworthy, compliant, and ready for actual operational deployment in specific industries. We are seeing this play out in real-time. OpenAI and Anthropic are effectively sending a message to the market: they cannot solve every problem with a general-purpose AI coworker. They have already announced large-scale frontline deployment joint ventures, building entire companies around configuring and customizing models for enterprises. If you think the release of the next model will solve everything, you would never pour billions of dollars into these projects. Therefore, if you want to get rich building AI applications—avoid the Yellow Brick Road and go explore the rest of Oz. Here is what we and some entrepreneurs in our portfolio have learned about what actually works. If you are starting a company, the Yellow Brick Road is the most obvious path, but it is also the most dangerous. Take a high-performance model, plug in some off-the-shelf connectors (like Google Drive, Slack, Salesforce, Notion, GitHub), and launch some kind of Agent orchestration layer on top. It’s magic! The problem here is that this is exactly what the labs are doing via Cowork and Codex. Obviously, they own the models themselves, which gives them better margins, control, and the ability to exert pricing power over any downstream vendor. But perhaps most importantly, they also control the "architectural choices" that determine which problems their products can perfectly solve. So far, they have been deliberate about the "model plus tool calls" pattern, which is exactly what is needed for the horizontal, low-step workflows on the Yellow Brick Road. Even if a startup can somewhat outperform Codex or Claude Code, the labs possess massive distribution channels and the strongest brand halo in AI. If you are an AI application company copying this playbook, using the same connectors, with no sub-agents or deep configuration underneath, and no distribution channels, then you are likely on a dead-end path to a bottomless pit. For startups, it’s not all doom and gloom. There are huge opportunities outside the Yellow Brick Road, where startups have a clear path to owning their customers and solving complex problems. These companies are building Agent experiences that weave models into a complex network of tools, automation, and integrations (in other words: software), which makes most of these startups vertical by default. They can focus on multi-step and multi-role collaborative work, and set up sub-agents for specific roles and vertical tasks, which Anthropic and OpenAI cannot reach with a general-purpose platform: collecting context across systems and then routing it to multiple people who must approve at different stages. This usually involves one or more legacy systems, tends to require deterministic results (no room for ambiguity), and is often directly tied to high-value business outcomes.

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching6 items

2026-04-28

Gemini launches Agentic Trading: Enabling Claude and ChatGPT to place orders directly on regulated exchanges and execute strategy trading

Similarity 140%關鍵字 claude/chatgpt同分類 zh

2026-04-24

The EU is pressuring Google to open core Android features to ChatGPT and Claude, signaling the start of the AI gateway war.

Similarity 140%關鍵字 claude/chatgpt同分類 zh

2026-05-26

Base Launches MCP Gateway Letting Claude and ChatGPT Execute Onchain DeFi Actions

Similarity 130%關鍵字 claude/chatgpt

2026-05-26