Stash: An Open Source Memory Layer Shared Across All Agents
The open source project Stash aims to solve agents' amnesia: providing an independent memory layer so any AI agent can remember users and context the way Claude and ChatGPT do.
AI News and Industry Updates
"I Canceled Claude": A Detailed Breakup Note
An emotionally restrained but clearly itemized cancellation statement: the author lists three reasons for canceling their Claude subscription—shrinking quota, quality swings, and unreachable support—each backed by records.
The Many Voices of DeepSeek V4's Launch Day
On DeepSeek V4's launch day, the HN thread itself is worth archiving: hands-on tests, skepticism, geopolitical anxiety, and price astonishment all mixed together—a slice of industry sentiment.
Anthropic Publishes a Claude Code Incident Postmortem
Anthropic released a postmortem of a recent series of Claude Code problems, with a candor rare among AI vendors—worth reading as an engineering document.
GPT-5.5 Released
OpenAI releases GPT-5.5. More than another round of benchmark refreshes, what's worth watching about this release is the product direction it reveals: the model is doubling down on agents and long tasks.
Rating Data Reveals the "AI Design Smell" in HN Submissions
Someone analyzed the rating data of projects submitted to HN and found that projects bearing the hallmarks of AI-generated design score systematically lower—"design slop" is being recognized and penalized by the community.
ChatGPT Launches Workspace Agents
OpenAI adds workspace agents to ChatGPT: able to access a team's documents, tools, and context, and stay resident to do work. The target is clearly the heartland of enterprise software.
ChatGPT Images 2.0 Released
OpenAI rolls out ChatGPT Images 2.0, with controllability and consistency of image generation as the headline. The competitive logic on this track has shifted from "drawing stunningly" to "editing obediently."
Meta Reportedly Records Employees' Mouse and Keyboard Behavior
Reports say Meta has started collecting fine-grained operational data on employees—mouse tracks, keystrokes, and the like. At a company that built its business on data collection, the lens has finally turned on its own people.
Atlassian Enables Data Collection for AI Training by Default
Atlassian was found to have enabled the option to train AI on customer data by default, turning enterprise data in Jira and Confluence into raw material, and the old "consent by default" trick provoked public anger.
Changes to GitHub Copilot Individual Plans
GitHub adjusted the plan structure of Copilot's individual edition, with changes to both quotas and billing. The pricing of AI coding tools is broadly sliding from monthly flat-rate to metered.
Blacklist or Not, the NSA Uses Claude Mythos Anyway
Reports say that despite Anthropic being placed on the Pentagon's blacklist, the NSA is still using its Claude Mythos model. The government's internal split over AI suppliers is out in the open.
Showing 12 of 12 stories