Tech News
← Home  ·  All topics

Agentic

214 GoKawiil briefs on this topic · RSS for agentic

Researchers find AI shopping agents ignore sponsored product listings

A study led by University of Toronto's Akshit Kumar found that AI shopping agents disregard items tagged as 'sponsored' when evaluating search results. The agents did show preferences based on position in results, but which position—top, middle, or bottom—varied inconsistently across different models and model generations. Economist David Rothschild and others are separately discussing how AI agents might enable micropayment-based content consumption instead of subscriptions.

Major retailers and banks split over allowing AI shopping agents like Meta Muse

JPMorgan Chase, Charles Schwab, Amazon, Neiman Marcus and Kohl's are blocking AI personal agents from acting on customers' behalf, while Walmart has opened its site to such bots, pricing products the same for agents as for humans. A Billions Network audit of 51 large U.S. retailers found only 11 have deliberately set a policy on AI agents, with most blocking them unintentionally via older anti-scraper rules.

Companies report little real-world impact from Meta's Muse shopping agent so far

Meta says more than 10 million people in the US and Canada have activated its agentic AI tool Muse, with over 25% using it for shopping and average purchases above $200. But roughly 50 companies across travel, retail, telecom and finance told WIRED or said on earnings calls that orders, bookings and cancellations tied to Muse and similar agents remain minimal, with firms like DoorDash, T-Mobile, Hilton, Ally Financial and eBay describing activity as negligible.

Chainalysis reports breach using 'offensive agentic techniques'

Blockchain analytics firm Chainalysis disclosed that an external attacker used what it calls 'offensive agentic techniques' to gain unauthorized access to its internal systems and tools earlier this month. The company says exposed data includes customer names, email addresses, user IDs, and organizational role labels, but found no evidence that investigation content, authentication credentials, or session tokens were accessed. Chainalysis has patched vulnerabilities, is working with outside cybersecurity experts and using LLMs for a security review, and says the incident is unrelated to the recent Binance theft.

Intel shares fall 29% after weak Q4 revenue guidance and PC demand warning

Intel stock dropped nearly 29% to close at $65.76 on Friday, its worst single-day performance since 1974, despite beating Wall Street's earnings estimates for the quarter. The company's fourth-quarter revenue forecast of $15-16 billion fell well short of the $17 billion analysts expected, with Intel citing a projected 12% drop in PC shipments in 2026 due to rising memory prices, plus a fab shutdown in Israel following a fatal explosion.

Apple acquires AI data engineering startup Ascend.io

Apple has acquired Ascension Labs, Inc., known as Ascend.io, according to a disclosure required under the EU's Digital Markets Act. The deal is structured as an acquihire, giving Apple the right to hire certain Ascend employees, a non-exclusive license to its intellectual property, and assignment of certain patents. Ascend.io, founded in 2015, built an 'Agentic Data Engineering' platform using AI agents to automate data pipeline tasks.

Nvidia's Vera CPU benchmarked against AMD Turin and Intel Granite Rapids-AP

Tom's Hardware tested Nvidia's new Vera data center CPU in standalone form against AMD's EPYC Turin and Intel's Granite Rapids-AP chips. Vera posted leading results in Python performance and memory bandwidth, and also held its own in general-purpose server workloads, though x86 rivals retained an edge in core-density-driven frontend tasks.

Mistral open-sources weights for Large 4 'Le Chonk' trillion-parameter model

Mistral has released the open weights of Mistral Large 4, its flagship 1-trillion-parameter multimodal model, following a public preview earlier this month. The company says added reinforcement learning has improved coding, cybersecurity and reasoning efficiency since preview, and the model now ranks as the top open-weight model for coding among the most capable open models overall.

Meta, OpenAI and Anthropic compete for consumer AI agent market, guide notes

A tech guide highlights that agentic AI tools, once mostly used in technical circles, have expanded this year into mainstream consumer use for tasks like grocery shopping and to-do list management. Major AI labs including Meta, OpenAI and Anthropic are now competing directly in this space, offering different agent products with varying capabilities. The guide offers tips for newcomers trying to navigate the differences between these offerings.

Thoughtworks CTO explains developers' hesitancy toward agentic coding tools

Rachel Laycock, CTO at Thoughtworks, writes that many skilled developers remain cautious about fully adopting AI coding agents despite the firm's history of embracing early tools and practices like continuous integration and automated testing. She says a Technology Radar team discussion in September helped clarify that the hesitation isn't primarily about job security or love of hand-written code, but a deeper concern tied to quality and production-readiness.

Amazon adds 100+ shopping skills to Alexa ahead of holidays

Amazon has rolled out more than 100 new shopping capabilities for Alexa, letting users manage lists, adjust subscription delivery frequency, update delivery instructions, and resolve payment issues through voice commands. Users can also ask Alexa questions about Prime and Audible subscriptions and billing problems. Amazon said use of its agentic shopping features has grown eightfold since the start of the year, with over 400 million people using Alexa for shopping in the past 12 months.

Tom's Hardware develops prototype CPU benchmark for agentic AI workloads

Tom's Hardware is building an internal benchmark to measure how CPU and memory performance affect agentic AI loops, isolating these components from the AI accelerator. The outlet says it has prototyped the test and settled on a core design, but acknowledges the methodology is unfinished and the results aren't yet reliable enough for publication.