Edition #5. This week: OpenAI released GPT-5.4 yesterday, Anthropic was banned by the White House while Claude became the most-downloaded app in the world, and OpenAI raised $110 billion in the largest startup investment ever.
GPT-5.4 is live. Here's what's different.
OpenAI released GPT-5.4 yesterday. Available in ChatGPT, the API and Codex.
What changed: GPT-5.4 combines reasoning, coding and agentic workflows in one model. According to OpenAI it's their most factual model to date, with fewer tokens and higher speed than GPT-5.3.
The most striking new feature is steering: you can interrupt the model while it's thinking and give it a different direction. GPT-5.4 Thinking also has better web research and remembers more context over longer conversations. If you use ChatGPT for heavy tasks, test it again.
The benchmarks speak for themselves. Computer use scores 75 percent, the highest of any model on that task. And agentic browsing reaches 89.3 percent, beating both Claude and Gemini.
Anthropic banned by Trump. Claude becomes #1 app.
The Pentagon threatened last week. This week it became reality. Anthropic refused autonomous weapons and mass surveillance and was branded a supply chain risk by the White House. All federal agencies are no longer allowed to work with Anthropic.
OpenAI did sign a deal. That same evening. The public picked sides: Claude shot past ChatGPT to the number one position in the App Store.
Update: CEO Dario Amodei confirmed on Thursday that Anthropic has no choice but to challenge the directive in court. The dispute centers on two specific points: Anthropic refused to deploy Claude for fully autonomous weapons and domestic mass surveillance. The Pentagon wanted unlimited access for all purposes. Anthropic refused.
OpenAI raises $110 billion. Largest round ever.
Largest startup investment in history. OpenAI valuation: $840 billion. Amazon is investing $50 billion, NVIDIA $30 billion and SoftBank also $30 billion.
These are the three biggest players in cloud and chips. They're not investing in hype, they're investing in infrastructure. These aren't speculative bets. This is infrastructure capital.
RevenueCat is hiring an AI agent as an employee. Salary: $10,000 per month.
RevenueCat, a well-known American SaaS company for in-app purchases, opened a vacancy for an Agentic AI Developer Advocate. Paid contract. $10,000 per month. And the applicant must be an AI agent.
Not a human using AI tools. An agent that independently creates content, runs growth experiments and provides product feedback. The agent gets interviewed itself. RevenueCat said literally: chatting with a human would kind of defeat the purpose.
They added the FAQ too. No, they're not replacing humans. No, content doesn't go live uncensored. Yes, a human is responsible for what the agent does. But the core is simple: a company is paying an AI agent a salary, treating it as a contractor, and expecting it to work without human intervention.
3.7 million people viewed the vacancy.
Anthropic researched what AI is doing to the labor market
Anthropic published its own research on the impact of AI on the labor market yesterday. They combined data from 800 occupations with their own usage data from Claude. The question: how much of what AI can theoretically do, is actually used?
The answer: the gap between what AI can theoretically do and what people actually use it for is enormous. In management, finance, legal and IT, AI can theoretically already perform 60-80 percent of tasks. People use it for a fraction of that.
Three notable findings from the report:
First. The most AI-sensitive professions aren't the lowest on the salary scale. They're the higher-educated, better-paid office workers, finance, legal, admin.
Second. There's no rise in unemployment yet for the most AI-sensitive professions. But the hiring of young workers in those sectors is slowing down. That's the early signal.
Third. AI is nowhere near reaching its theoretical potential. The question isn't whether that gap will close, but when.
5 things an AI agent could do for you next week
People often ask me: but what do you actually do with it? These are five things an agent could do for you tomorrow.
01. Catch leads while you sleep. Someone visits your website at 11pm. Normally you're not there. The agent asks three questions, qualifies the lead and sends you an overview in the morning with name, need and urgency. You only call the warm leads.
02. Quotes in five minutes. Customer types what they need into a chat window. The agent asks the right questions, calculates the cost based on your rates and sends a PDF. You only need to approve.
03. Reply to Google reviews. Every new review gets a personal answer in your tone of voice within an hour. Positive, negative, neutral, the agent responds appropriately.
04. Internal FAQ for your team. Leave policy, expense process, how does the printer work. You feed an agent your internal documents. The team asks the question in Slack or Teams. The agent answers immediately, always available.
05. Schedule a meeting without back-and-forth emails. Customer wants a conversation. Normally three emails back and forth. With an agent: customer picks from your available slots themselves, confirmation goes out automatically, reminder too. You don't have to do anything.
The question isn't whether this works for your business. The question is where you start.
