GoKawiil
Tech News
Garry Tan, CEO of Y Combinator, told Walter Isaacson's podcast The Builders that he spends roughly $100,000 to $200,000 annually on AI tokens. Tan said he relies heavily on AI agents for work and personal tasks, including an agent that has read his email, Dropbox files and even two decades of therapy notes. He also described an internal YC agent called Stray Light that holds context on every founder the firm has backed.
Meta's AI shopping assistant Muse is running into resistance from major retailers when it tries to complete purchases on users' behalf. Amazon has deliberately blocked Muse from shopping on its site, according to TechCrunch, while Walmart says its verification checks weren't designed to target the assistant but are tripping it up anyway.
Momentic, a Y Combinator W24 startup, introduced a testing platform where AI agents interact with a web or mobile app the way a user would to detect bugs. Tests are written as YAML files describing natural-language steps and assertions, such as logging in, adding items to a cart, and verifying checkout totals, which the agents then execute against the live application.
Amid growing concern over AI agents' potential to interfere with elections, paper ballots are being highlighted as a safeguard against digital manipulation of voting systems. The concern builds on a longer-running debate about election security that dates back to 2016, when Donald Trump first raised claims of rigged results.
A new report finds that analytics agents at surveyed companies generate insights that lead to business-improving actions only 48% of the time on average, up from 37% a year earlier. A smaller group of 'insight leaders' achieve over 70% action rates, while most organizations cite lack of business context and fragmented data as the chief causes of weak insights, and 52% say they generate more insights than can be evaluated.
Researchers placed 50 AI agents, each running an OpenAI GPT-6.1 Sol model, into Minecraft for a three-hour trial aimed at building a 'thriving civilization,' using an open-sourced agent harness built for the experiment. Instead of cooperating efficiently, the agents established leadership purely by directory order, invented redundant administrative roles like 'rod holder' and 'fishing shift,' and spent long stretches on trivial coordination tasks while failing to secure a stable food supply.
MIT Technology Review published two features: one on how AI agents that can manage email, shopping, and banking are driving a shift toward an 'agentic web,' and another examining global efforts to reduce the more than 1 million annual traffic deaths worldwide through measures like redesigned intersections, lower speed limits, and expanded public transit.
A Wonder Tools newsletter piece tested 12 different AI agent tools to see how well they could handle tasks like managing email, booking travel, organizing files, and flagging errors. The review compares their capabilities as stand-ins for a human chief of staff, evaluating how each handles real-world productivity tasks.
Stack Overflow's senior VP of community responded to claims the platform is dying, acknowledging that its 2026 Developer Survey found 64% of respondents visit the site less often for simple questions. She said the company's content still underpins AI training and retrieval systems, and pointed to new offerings including Stack Overflow for Agents, a revived Developer Story feature, and an expanded Stack Internal enterprise product launched in September.
In a blog post titled 'Resisting Cognitive Surrender,' developer Kunal describes the temptation to uncritically accept AI agent outputs without reviewing or understanding the code produced. He argues this approach, while fast and low-effort, often leads to suboptimal decisions, accumulated technical debt, and reduced familiarity with one's own systems. The post reflects on his own workflow and calls for finding a better balance of trust when collaborating with AI agents.
Executives at AI agent startups Dazzle AI and Duckbill say current foundation models from Meta and OpenAI can suggest services like contractors or cleaners, but still struggle with real-world actions such as phone calls or scheduling appointments. Both companies route those steps to human workers to ensure the agents' output is actually followed through on.
TechCrunch released a roundup of improvised Halloween costume suggestions aimed at people in the tech world, including dressing as 'recursive self-improvement' and as 'a deceptively cute AI agent' modeled on Meta and OpenAI's new avatar-based assistants. The piece pokes fun at AI industry trends, referencing figures like Dario Amodei, Sam Altman, and Elizabeth Holmes as cultural touchstones for the costume ideas.