Tech & AI Twitter Unfiltered: The Smartest Daily Chatter, Curated

Forrester started with six hundred and fifty conversational AI vendors for its latest report. Only fourteen survived the evaluation. That’s not a typo. Six hundred and thirty-six companies in the chatbot space effectively just vanished from the list of serious contenders. Last week on the show, we talked about the rise of personal AI as a kind of 'second brain'. Well, this week, that brain is getting hands, and the market is brutally deciding who gets to build them. The biggest story is that consolidation. Forrester’s new Wave report on conversational AI is a bloodbath. When you go from 650 companies down to just fourteen leaders, it’s not a correction. It’s an extinction event.

This is the market saying the era of simply having a chatbot is over. The bar has been raised, and most players couldn't clear it. We're seeing the first major shakeout of the AI boom. Now, running parallel to that, a fundamental shift in what we even want AI to do is happening. A thread from Khemaridh Hy summed it up in five words: "Chat is dead. Agents are the new default." This isn't just talk. Codex just released its "super app," a platform called Cowork designed to bring local AI agents directly into your knowledge work. This puts it in direct competition with OpenAI's ChatGPT "work" offerings, but the philosophy is different. It’s not about asking a question and getting an answer.

It's about giving a task and getting it DONE. And the tools for these agents are arriving at a blinding pace. Just this weekend, a new open-source skill called '/last30days' dropped. Imagine typing one command and having an agent read and synthesize everything on X, Reddit, YouTube, TikTok, and even academic papers on arXiv from the last month. That’s not a search query. That is a research assistant. And it’s open-source, meaning anyone can build on it. This is the ground-level view of the agent-centric world that’s emerging. Meanwhile, the hardware is trying to catch up to the software. Or maybe, the software is trying to catch up to the physical world. A new report from Pandaily clocked thirteen new embodied AI models released in June alone.

That’s basically a new one every forty-eight hours. These are the systems designed to power robots, to understand and interact with physical space. The competition is shifting from pure language benchmarks to models that can see, move, and manipulate. And finally, in the middle of all this tech, a reminder of the people driving it. A popular thread this week highlighted a profile of Daniela Amodei, the co-founder of Anthropic. She gets less public attention than her brother and co-founder Dario, but the piece paints her as the key operator, the one who actually makes the massive, complex machine of a frontier AI company run. It’s a critical reminder that behind every model, every breakthrough, there are human leaders making thousands of consequential decisions.

So what's the thread connecting a market report, a new app, an open-source tool, and a CEO profile? It’s a transition from passive to active AI. From conversation to capability. Let's dig deeper into that, because this is the week the change became undeniable. The idea that "chat is dead" sounds like hyperbole. It's not. It’s a redefinition of the job to be done. For the last few years, the primary interface for AI has been a blinking cursor in a text box. You ask, it answers. It was revolutionary, but it was also limited. You were the one who had to take that answer and DO something with it. You had to copy the code, paste the email draft, check the facts. The AI was a brilliant, talkative intern.

But it was still an intern. The shift to agents is about promoting that intern. Codex's new Cowork platform is the first major shot across the bow in this new war. While ChatGPT and others have been adding "actions" and "plugins," Codex is framing its entire product around agents from the ground up. These are local agents, meaning they can have more secure access to your files, your emails, your data, without necessarily sending it all to the cloud. They’re designed not to just answer your question about a report, but to write the first draft of the next report using the data you gave it. This is a move from an AI you converse with to an AI that works alongside you—and for you.

It’s a colleague, not a search engine. This is where that new '/last30days' skill becomes so powerful. Think about what it represents. An agent equipped with this can answer a prompt like, "What was the consensus among AI researchers on X and YouTube about the new Fable 5 model in the past month, and how did the market on Polymarket react?" To do that yourself would take hours. You’d have dozens of tabs open. You'd be scrolling endlessly, trying to piece together signal from noise. An agent with this skill does it in minutes. It ingests, reads, understands, and synthesizes across completely different platforms and content types. This isn't just a powerful tool. It changes the scope of the questions you can even think to ask.

You can operate at a level of meta-analysis that was previously reserved for dedicated analyst teams. And now, it’s a single command line. So let’s connect this back to that Forrester report. The one that cut down 650 vendors to just 14. The companies that were culled were the ones that perfected the chatbot. They made them sound more human, answer questions more accurately, and integrate with a knowledge base. All necessary things. But they were perfecting the last generation's model. The fourteen survivors—the Microsofts, the Googles, the OpenAIs, and the emergent leaders—are the ones who are building agents. They understand the goal isn't a better conversation. The goal is a better outcome.

The goal is automation of cognitive work. What does it take to survive that cut? First, you need a world-class foundational model. That’s table stakes. But that’s not enough. Second, you need a platform that allows that model to take actions—to access APIs, to run code, to read files, to browse the web. Securely and reliably. Third, you need an ecosystem of developers building skills and tools for those agents, like '/last30days'. You can't build it all yourself. And fourth, you need the trust of users and enterprises to give these agents access to the data and systems required to do meaningful work. That is the razor-thin tightrope the report was talking about. It’s not just about chatbot accuracy anymore.

It’s about building a trusted, capable, extensible platform for automated work. Most of those 636 companies were building products. The 14 survivors are building platforms. There's a universe of difference. So when you see these threads on X, this isn't just tech chatter. This is the live documentation of a paradigm shift. We are moving from the era of Large Language Models to the era of Large Action Models. From an AI that knows, to an AI that DOES. And this week, the difference became crystal clear. This all sets up a fundamental change in how you're going to interact with technology over the next eighteen months. The debate is no longer about which chatbot gives the cleverest answers.

The question is now, which agent can you trust to get the job done? The one that can schedule your meetings, draft your reports, and summarize your industry's entire online conversation from the last month. The rise of embodied AI, with a new model every two days, shows where this is headed. First, the agents master the digital world of text and data. Next, they are given eyes and hands and are pointed at the physical world. The same principles of breaking down a high-level goal into a series of executable steps apply, whether the goal is "summarize these documents" or "assemble this piece of furniture." The eulogy that one developer wrote for agents wasn't for the agents themselves.

It was a eulogy for the old way of doing things. A eulogy for the friction, the manual labor, the cognitive overhead of tasks that are now ripe for automation. We are at the very beginning of this transition, and the playing field is being set right now. The companies that survived Forrester's cut aren't just the winners of the chatbot wars. They are the initial contenders for building the next operating system. An OS built not on files and folders, but on goals and agents. The conversation is over. The work has begun.