Oct 25, 2026 · 6 min listen · Last updated October 25, 2026
From storyflo. This is your daily audio brief. Theo, October 25th. I. stories that bear on what's coming next. Let's get into it. First, from The Decoder. OpenAI offers discounts up to 90 percent to poach Anthropic's biggest customers.
Listen · storyflo · A.I.
Daily A.I. Brief · October 25th
0:00-5:45
Pick your daily storyteller
Subscribe to match with Theo, Jessica, Chloe, Mason, Brock — your voice, every brief.
Audio pre-rendered by Storyflo · cached + delivered from the edge
OpenAI offers discounts up to 90 percent to poach Anthropic's biggest customers
OpenAI is making a bold move by offering significant discounts, reportedly up to 90 percent, to attract some of Anthropic's biggest clients. Companies like Rippling, Replit, and DataCamp have shared that their AI expenses have dramatically decreased after making the switch. This shift comes at a time when Anthropic is experiencing slower growth, which could be a concern for them. OpenAI seems willing to sacrifice some of its profit margins to gain a larger slice of the market. Meanwhile, competition is heating up with Chinese companies like Deepseek also applying pressure on prices, adding another layer to this evolving landscape. It's interesting to see how this will reshape the dynamics in the AI sector.
Ex-Google CEO Eric Schmidt calls OpenAI and Anthropic modern colonial trading companies
In a recent piece for Foreign Affairs, Eric Schmidt, alongside historian Niall Ferguson and researcher Lisa Klaassen, paints a striking picture of the global AI landscape. They argue that the U.S. and China are emerging as the dominant forces, likening companies like OpenAI and Anthropic to modern colonial trading firms. The authors emphasize that the competition isn’t just about technological breakthroughs; it’s largely about financial might. The projected U.S. corporate spending on AI could exceed $600 billion by 2026, highlighting the scale of investment at play. However, challenges loom, including rising debt, public protests against data centers, and concerns over a potential chip shortage linked to Taiwan, which could hinder progress for both nations. It’s a complex scenario where the stakes are high, and the implications for global dynamics are profound.
Anthropic says one AI agent beats many subagents now that models can handle complexity alone
Anthropic just showed that a single, more capable model can replace a whole chain of specialist sub‑agents. By letting the model pull in the right context only when it needs it, they cut the back‑and‑forth that used to slow things down.
In their internal tests the unified agent ran roughly twice as fast as the same task split across multiple smaller bots. The speed boost comes from fewer context switches and less overhead shuffling data between separate instances.
They’ve even dropped a migration kit on GitHub, so teams can swap out their patchwork of sub‑agents for one streamlined model without rebuilding everything from scratch. It feels like a tidy, practical upgrade rather than a flashy overhaul.
On the markets — Kalshi traders have been actively repricing this story in the last day.
Anthropic co-founder Jack Clark moves his prediction for self-improving AI forward to 2027
A new benchmark from P‑Zero Research is quietly turning heads. It measures how well AI models can design and evaluate experiments on their own, a skill that’s been improving at a pace that feels almost too fast. The numbers are climbing so quickly that even the cautious folks at Anthropic are taking notice.
Jack Clark, one of Anthropic’s co‑founders, has been watching the trend closely. He used to say self‑improving AI might be a decade away, but the latest data has nudged his estimate forward to 2027. It’s not a dramatic leap; it’s a subtle shift based on concrete evidence that these models are getting better at tweaking themselves.
What’s surprising under the hood is that the models aren’t just learning from more data—they’re learning how to ask better questions, design better tests, and judge their own results. That meta‑learning loop is the real engine here, and it’s happening faster than many of us expected. So, while the headline might be “self‑improving AI by 2027,” the real story is the quiet, mechanical progress that’s quietly reshaping the field.
EP 30 | Palantir Stock Analysis: Is It About to Crash?
Palantir's recent performance is raising eyebrows, especially with its stock price hovering around $196.74 despite impressive revenue growth. They reported a staggering 93% increase in Q2 revenue, with U.S. commercial growth soaring by 149%. Yet, the financial outlook isn't as rosy as it seems. Analysts are leaning towards a "Strong Sell" classification, with a consensus suggesting significant price drops in the coming years.
The company’s operating cash flow is strong, showing over $2 billion in free cash flow, but the valuation feels stretched at nearly 58 times projected revenue. There are also concerns about competition and budget scrutiny in both commercial and defense sectors, which could impact future growth. As we await their Q3 results on November 2, the focus will be on customer expansion and cash conversion. It's a fascinating mix of growth potential and valuation risk, making it a topic worth keeping an eye on.
OpenMobile-2 proves open-source smartphone agents can rival proprietary models
So, there’s this interesting development with smartphone agents that I think you’d find pretty cool. Researchers created something called OpenMobile-2, which is an open training environment designed to help these agents learn from a diverse range of apps. They included over 110 applications, with 35 being replicas of popular ones we all know. What’s surprising is that instead of just throwing more data at these agents, they focused on the variety of apps to enhance their learning. When tested on real Android devices, OpenMobile-2 actually outperformed other open-source agents, showing that open-source can hold its own against proprietary models. It’s a neat reminder of how collaboration can lead to some serious innovation.
On the markets — Kalshi traders have been actively repricing this story in the last day.
Nvidia open-sources Flora to make robot video labeling 125x cheaper than GPT-6 Astra
Nvidia has just open-sourced a new data engine called Flora, which is making waves in the world of robot video labeling. It can label an hour of robot footage for just 72 cents, a staggering drop from the $88 charged by GPT-6 Astra. What’s interesting is that Flora not only cuts costs dramatically but also scores slightly higher in quality. It’s built on three fine-tuned Qwen3.5 models, which means it’s designed to catch errors in robot actions more effectively than its competitors. However, it does have some hiccups when it comes to descriptive accuracy. This move is part of Nvidia's broader strategy to tighten its grip on the entire robotics value chain, showing just how serious they are about reshaping this space.