It’s almost impossible to keep up with the flood of AI news. A few examples:

• Anthropic released a browser-based coding environment, called Code. It’s similar to OpenAI’s Codex—it lets you write code in the browser, and integrates with GitHub. It feels like there will be less and less oxygen for smaller code-generating platforms like Lovable and Replit, as we discussed on this week’s podcast.
• Anthropic also released “Skills.” This is an approach to coding with AI and extending their chat platform—you should read one to get a sense of how it works. It’s sort of a notebook that mixes prose and code, guiding an LLM to achieve a task. 
• OpenAI released a whole new browser, called Atlas. Like so many of their products, it doesn’t quite work—following the prompt to search for restaurants in my area, it suggested the very fancy Le Bernadin, then also showed me restaurants in Houston. It’s like Perplexity, I guess. Or Atlassian’s Dia.
• OpenAI is paying bankers $150/hour to train ChatGPT to replace junior banking analysts. Mostly they’re using Excel.
• Anthropic is doing something similar around life sciences.
• Ali Baba got 82% more processing out of its GPUs. They did it by…programming more intelligently.
• DeepSeek released an open LLM that is good at OCR and apparently represents a path to dramatically expanding context windows.
• Amazon plans to replace 500,000 jobs with robots. That’s about 75% of operations.
• Nvidia released a desktop computer. It’s the size of a Mac Mini and can do a petaflop.

To me, it feels like three things are happening at once:

First, upstart companies are broadening their offerings, but they’re already unbelievably huge, so it feels like Godzilla swishing its tail around. Anthropic is nerdy and is trying to nail down coders, and their products vibe with a lot of engineers. That’s bad news for a lot of smaller vibe-coding tools, though. OpenAI is aggressively trying to move into the consumer product space, with offerings like “shopping” and “a browser,” because they want to compete with Microsoft and Google. 

Since literally everything else is cheap compared to LLM development, this creates a really wild situation where their competitors—like Perplexity—are totally reliant on them, while OpenAI can simply stand up another 30-person team and go “make me a browser,” which they then give away on top of ChatGPT. It’s unfair, but that’s the risk of building your business on someone else’s API. Of course, Google already has a browser, and OpenAI is building on top of that. There’s a lot of mess ahead.

Second, software development acceleration is well underway, and LLM companies are looking for more white-collar low-hanging fruit to pick off. A product like Claude Code, or Copilot—they’re finding their market and their identity. But a lot of jobs are programming-adjacent; junior analysts at banks are basically spreadsheet programmers. 

It’s not surprising that OpenAI is nudging along a “financial model” generation path in its products. They’re obviously going to keep fanning out to finance and engineering-adjacent roles, trying to replace the less experienced humans in those positions. Anyone with “analyst” in their title who isn’t a therapist is up for inspection: Business analysts, risk analysts, actuaries, data analysts, policy analysts. If you use Excel, SPSS, R, or MATLAB, they may be gunning for you here.

Finally, platforms are moving fast at scale. This one is kind of meta, but it’s pretty surprising to me. Microsoft Windows is a platform. They release a new version every couple years, and they usually give it a number, like “95” or “11.” (Notwithstanding “Vista” and “XP.”) It costs zillions of dollars and takes thousands of people to run, and every time they put out a new version, people upgrade computers and IT departments grumble; it requires tons of effort, and marketing and things break. 

Meanwhile, Amazon has a platform that it uses to run factories, which seems to be updated all the time. That software makes the robots scoot, but when they want to make them scoot smarter, they release SCOOTFIX hot pack #5000, and now they can hire 50,000 fewer humans made of meat. 

The LLM companies themselves are also moving rapidly. DeepSeek does quant trading, has a lot of national pride—and uses the open-source ecosystem to keep everyone paying attention by releasing products that five years ago would have made MIT professors cry, but is now just a normal weekday. OpenAI and Anthropic are very large platforms, and they’re shipping at unbelievable velocity for their scale, not least because they’re willing to ship broken stuff and fix it later.

This sort of all-in massive platform rapid development and deployment by multiple tech megacorps at once doesn’t have a lot of precedent. Google is huge, but it took 20 years to get to monopoly status. Social networks grew quickly, but they were primarily about communication—they weren’t seeking to replace people at banks. You know those Instagram videos where thousands of drones form an image of a dragon over some industrial city in China? That’s the velocity we’re headed towards, in every industry in every way.

Last week, based on a chat, ChatGPT offered to run an agent for me: It would summarize research papers for me every Monday. I said sure. It didn’t notify me the next time it did this work, or even send me an email—it just updated the chat with some new content and put that chat at the top of a list. There’s a clear path to a future where ChatGPT is running lots of little agents on my behalf and pinging me when they do work; it’s also getting quite good at suggesting topics to further a conversation. They want all of my computer time, and there’s a non-trivial chance they could have it. It’s an era of huge bets and huge risks—and for us little guys, it can be unsettling. We just have to avoid Godzilla’s tail.