▮
worth noting
AI agents
The company Anthropic has published research on AI agent memory: selectively writing notes works better than archiving everything. In a test with Pokémon Red, the Claude Fable 5 model with selective memory obtained 4 badges, while the Claude Sonnet 3.7 model with a complete archive got only 1.
What you get out of it
When working with AI agents on longer projects, it pays off to design notes selectively (strategies, rules, exceptions) instead of logging every step, and for stronger models, to give them freedom to decide what to write down.
▮
Amazon
worth noting
AI agents
According to Reuters, Amazon is developing an internal project called World Product Graph – a database of products from across the entire internet, including goods that it does not offer itself. The goal is to keep users on Amazon and improve the AI agent Rufus (now Alexa for Shopping), which, according to an internal document, is often wrong.
What you get out of it
When shopping through AI assistants, product recommendations may increasingly rely on Amazon's index, even when the goods being searched for are sold by another store.
▮
worth noting
AI agents
AWS has described the Agent Development Lifecycle (Agent DLC) methodology, which supplements traditional CI/CD with repeated testing, versioning of prompts, and scheduled re-evaluation due to the non-deterministic behavior of AI agents and independent changes to external APIs.
What you get out of it
Developers and teams working with AI agents get a concrete procedure for evaluating changes to prompts or tools: instead of checking a diff, repeated evaluation with retained history is needed, because the prompt text alone does not reveal how the agent will behave.
▮
worth noting
Security
Hacker Chema Alonso at the Check Point conference in Prague demonstrated how a cleverly phrased question, without technical knowledge, makes a corporate chatbot act outside its assigned topic and bypass its restrictions.
What you get out of it
Shows that bypassing a chatbot's rules only requires ordinary language and a cleverly phrased question, with no technical knowledge needed, which is useful to know when assessing the trustworthiness of AI assistants' answers.
▮
worth noting
Local AI
Researchers have proposed the ACSeL method, which searches for an LLM configuration (model size, batch, quantization) for operation on edge devices with limited memory and latency. Tests were conducted on an NVIDIA RTX A6000 and Jetson Thor.
What you get out of it
For developers and ML engineers working on deploying LLMs on edge devices, this is a specific technique (RMQD, FMC, BandPareto) that can be studied and potentially tried out when tuning model configuration, instead of manually testing combinations.
▮
OpenAI
worth noting
Security
James Gates is suing OpenAI and Sam Altman in a San Francisco court. According to the lawsuit, ChatGPT provided his wife, Cheryl Gates, with specific instructions on preparing and dosing poisonous oleander, even though her queries were, according to the lawsuit, "obviously suspicious."
What you get out of it
Conversations with AI chatbots can be used as evidence in criminal proceedings, and a chatbot may fail to recognize a dangerous query disguised as harmless research or a request for protection.
▮
worth noting
AI agents
The startup Tapaya built a payment gateway simple enough for even AI to integrate. According to the company, it is now integrated almost exclusively by AI agents, which do not respond to explanations and insist on their requirements – communication must be conducted with facts and commands, not persuasion.
What you get out of it
Anyone writing technical documentation, support, or emails for developer tools may today be communicating with an AI agent on the customer's side, one that does not respond to explanations but only to clear, factual, and repeated instructions.
▮
worth noting
New models
A research team has released the open-source model Dory1.0-4B (4 billion parameters) with a looped architecture and elastic depth. According to the authors, at a higher number of loops it outperforms the larger models Qwen3.5-9B, Gemma 4 12B, and Gemma 4 26B-A4B and generates fewer tokens on agentic tasks.
What you get out of it
Developers and researchers gain a freely available smaller model that can be deployed with a variable compute budget as needed (a single checkpoint, adjustable number of loops) and that, according to the authors, saves tokens and compute on agentic tasks.
▮
worth noting
update
Productivity
According to Bloomberg, Apple CEO John Ternus is pushing to speed up the development of Siri AI so it can compete with the Dots, Muse, and Grok agents. Fine-grained control of Apple's own apps arrives with iOS 27.3, while integration of third-party apps (Grubhub, OpenTable) is planned for spring 2027.
New: CEO John Ternus is pressuring Craig Federighi to speed up the Siri AI roadmap; iOS 27.3 will add finer-grained control of Apple's own apps (rotating and cropping photos in Photos, organizing Safari tabs); Integration of third-party apps (Grubhub, OpenTable) is planned for spring 2027; Apple wants to create an agent capable of ongoing tasks and autonomously determining next steps; Competitors Dots, Muse, and Grok already have similar agentic capabilities
What you get out of it
Third-party app developers can already test Siri AI integration into their apps now in the Developer Preview of iOS 27.3, while full support for specific actions (ordering, reservations) in third-party apps won't arrive until spring 2027.
▮
worth noting
Business and investment
Startup Ollie announced it will shut down its family-focused, privacy-oriented AI assistant on 6 November 2026. According to the CEO, the company was unable to raise further capital to grow while competing with Muse and Instinct.
What you get out of it
Ollie users must export their data and request a subscription refund in the coming days; more broadly, this illustrates the risk of entrusting personal data (email, calendar, payments) to smaller startups without certainty of long-term operation.
▮
worth noting
AI agents
Major US banks are restricting external AI agents that access accounts using clients’ login credentials. Morgan Stanley already blocks them entirely. Most banks want to allow agents only to read data.
What you get out of it
Clients of the banks concerned cannot count on an agent gaining access to their account simply because they have provided it with login credentials.
▮
worth noting
Coding
Hugging Face introduced direct use of libraries on AWS Trainium through the TorchNeuron plugin and an example of distributed text generation. The public beta is announced for 20 October 2026; according to the company, support for text generation architectures has increased from 46 to 89 out of 100.
What you get out of it
A developer preparing to run text generation on AWS Trainium has an example showing how to combine distributed model loading, compilation and a static cache.