1.
The Best Early Black Friday MacBook Air Deal: $200 Off
Well ahead of Black Friday, the MacBook Air is already $200 off, bringing down the price of the 13-inch model with an M5 chip, 16 GB of RAM, and 512 GB of storage to $1,099. This is the lowest price the MacBook Air has been at since June, when Apple permanently hiked prices across its entire Mac lineup. Amazon frequently sells this MacBook Air for $1,149, but this even lower price is usually reserved for big sales events—the last being in October during Prime Day. And now it's back. I probably don't need to expound too much on what makes the M5 MacBook Air such a fantastic, do-it-all laptop.
2.
Find Production Agent Failures Before Users Do
There’s a new approach to testing AI agents that could save a lot of headaches down the line. Instead of waiting for users to uncover failures, you can now identify issues in a simulated environment before deployment. The PlurAI Simulation system rebuilds your agent and its tools from your code, allowing simulated users to push it to the limit. It evaluates the actions taken, like ensuring a customer is verified before processing a refund. If something goes wrong, it provides the run that exposed the failure along with a fix you can verify.
Additionally, the HarnessRouter project has introduced Harness Arena, which benchmarks different harness and model combinations side by side. This lets developers see how various configurations perform under the same conditions, focusing on total cost, execution time, and task success. The results can be surprising; for instance, one configuration ran successfully at a fraction of the cost compared to another that failed multiple times.
This means you can now choose an agent stack based on real performance metrics rather than just model rankings. With the Unified Harness Protocol, once you pick a configuration, it can handle future tasks seamlessly, making the whole process more efficient. So, would you rather discover an agent failure in a simulation or in a customer ticket? It’s definitely worth considering.
3.
How AI Work Becomes Reusable. A Weekly digest Nov,4 - Nov,11
This week, the conversation around AI shifted from what models can achieve in a single run to what can actually be reused afterward. Axiom Math has made strides by formalizing a version of the Stacks Project in Lean, creating a vast library of mathematical proofs visualized like a city. This approach allows users to trace dependencies, although verifying the intent behind a formal proof remains a challenge.
We also delved into looped Transformers, which can reprocess the same layers multiple times, raising questions about efficiency and whether additional passes truly add value. The exploration highlighted that fewer parameters don’t always equate to faster performance, emphasizing the importance of adaptive stopping in model predictions.
On the practical side, we looked at the evolving landscape of AI agents, both open-source and closed-source, comparing tools across different platforms. A key takeaway is that just because an application runs locally doesn’t mean all processing is done on your machine; understanding the surrounding infrastructure is crucial.
Finally, Ksenia reflected on the "anxiety of abundance" that comes with having more AI tools at our disposal. While these agents can enhance productivity, they also create a paradox of choice, leading to more tasks and decisions to juggle. This ties back to the importance of measuring progress in nuanced ways, rather than just a binary success or failure.
4.
Inflection AI
Sean White, the CEO of Inflection AI, is steering the conversation about artificial intelligence away from the relentless pursuit of AGI. He believes that focusing on human-centric design, particularly emotional and relational intelligence, is far more valuable. Inflection’s chatbot, Pi, was trained to understand high-EQ conversations and is evolving to facilitate group interactions rather than just one-on-one exchanges. Sean argues that the current AI landscape will be more fragmented than search engines, with people using multiple tools rather than a single chatbot.
He emphasizes the importance of privacy and user control over data, drawing from his experience at Mozilla, where user data was encrypted and kept private. Inflection is also involved in the Data Transfer Initiative, ensuring users can take their data with them. Sean's vision for AI includes it working seamlessly in the background, similar to the processes on our laptops, rather than overwhelming users with numerous agents to manage.
On the hardware front, Sean reflects on the limitations of devices like AR glasses, advocating for open-source solutions to prevent obsolescence. He sees the potential for AI to enhance social interactions, even in casual settings, and believes that the future will see more AI designed for group dynamics. His collaboration with Reid Hoffman has taught him to think strategically about the broader implications of AI, pushing the boundaries of what it means to be human in this evolving landscape.
5.
The Jev Revolution: What Engineers Are Already Building with System One Models
At the AI Engineer Code Summit, the spotlight was on system one models, which are faster and more reliable than the traditional reasoning models we often hear about. Diogo Almeida, the CEO of TypeSafe AI, introduced Jev, a model that’s already making waves since its launch in September. He emphasized that while LLMs can be unpredictable, Jev offers what he calls “reliable intelligence,” urging engineers to integrate these models into their code for smarter software.
Chinmay Goyal from Glean shared how they use Jev alongside LLMs, highlighting its flexibility without burdening users with technical details. Meanwhile, the team from UiPath showcased Jev’s efficiency in ranking document chunks in mere seconds. Scott Roberts from Workday demonstrated how Jev can enhance user interfaces by predicting actions, not just text.
The overarching theme was the potential of system one models to reshape software development, much like the early days of the internet. Almeida's call to “automate everything” resonated with many, suggesting we’re on the brink of a new wave of innovation in AI.
6.
9to5Mac Daily: November 11, 2026 – New Apple TV feature, more
Apple's just rolled out a new feature for Apple TV that’s really interesting. They’ve added a more intuitive way to manage your viewing preferences. Now, the system learns from your habits and suggests shows based on what you actually watch, not just what’s trending. It’s like having a personal assistant for your entertainment choices, which could make scrolling through options a lot less tedious.
In other news, there’s been a focus on improving the HomePod experience. Users can now control smart home devices more seamlessly, and the sound quality has received a notable upgrade. It’s all about creating a more integrated ecosystem, making it easier to manage everything from music to lighting with just your voice.
Lastly, Apple’s also been making strides in privacy. They’ve introduced new features that give users more control over their data, allowing for more granular settings on app tracking. It’s a move that reinforces their commitment to user privacy, which is becoming increasingly important in today’s digital landscape. So, if you’re an Apple user, these updates could really enhance your experience.
7.
Nvidia CEO Jensen Huang may be ousted after second affair revealed, reports claim
Nvidia CEO Jensen Huang came under scrutiny this week after he admitted to an affair that The New York Times linked to a suspected Chinese intelligence asset. It now appears that this may not have been his only affair, a second strike which may see him ousted from his position leading the multibillion-dollar chipmaker. The New York Times reports that Huang had an extramarital romantic relationship with his coworker Kimberly Powell, who currently serves as Nvidia's vice president of healthcare.
8.
Hackers stole AI credentials to get $740,000 in free compute; Railway picked up the bill
As AI model training and distillation increase, AI tokens are becoming a piping hot currency for threat actors. Cloud hosting provider Railway found out the hard way; it has confirmed a near 26-hour battle against attackers targeting AI provider credentials to sell to AI ‘transfer stations’ for use in model distillation. The crooks generated about $740,000 in fraudulent AI usage charges using the stolen credentials and potentially had access to up to 100,000 workspace credentials.
9.
What to Do if a Bug in Your AMD GPU Is Wearing Down Your SSD
AMD is working on a software update to fix a driver issue that causes excessive, continuous data writes to solid-state drives across its Radeon GPU lineup. First reported on sites including Tom’s Hardware and Neowin, the bug has been traced to two pieces of software: AMD External Events Client Module and Utility, known as atieclxx.exe or atiesrxx.exe under Windows processes. The background service handles monitoring functions, but as some users have noticed, it’s also writing a large amount of data to SSDs, as much as 432GB in 24 hours, according to Neowin.
10.
United Airlines Bans AI Agents From Booking Flights On Behalf of Travelers
The carrier also said that using an AI agent to, for example, check in for a flight could violate federal regulations, including rules requiring travelers to verify in person that their luggage does not contain banned substances. "As a result, United is taking steps to restrict access -- directly within our digital platforms and indirectly via travel agencies and aggregators -- to prohibit use by unauthorized, third-party personal AI agents or assistants," said the airline, which is updating its booking and ticketing policies to reflect the new rule.