Agents — AI news

Two Hacker News posts on harnesses and agent.md files reveal where real agent quality now lives.

Multi-agent AI systems fail from missing management structure, not weak models, a new analysis argues.

The AI needed a human to point out its own rules before it would actually act on them.

Agentic token usage on OpenRouter jumped 14x, signaling autonomous AI is moving from demos to production.

A protocol standardizing tool access and an agent that watches live code both attack the same problem.

One base agent, cloned into a full office cast, exposes what real multi-agent orchestration actually requires.

A new study maps why structured skills boost AI agents, and where the approach quietly fails.

DeepSeek's new experimental vision model rivals Claude Opus 4.8 on agent benchmarks at a fraction of the cost.

OpenAI's new macOS plugin lets ChatGPT read, search, and send iMessages without you touching Messages.

Slack Code moves AI agents from a private sidebar into the same channel humans use to coordinate work.

A new classification system asks which government decisions AI agents can make without a human sign-off.

A tiny open-source coding agent called fx reopens the fight over how much software a developer tool is allowed to weigh.

Binance's new Agent OS lets AI agents trade with real money — inside limits it enforces, not just suggests.

A new arXiv paper frames long-horizon robot manipulation as an agent problem, not just a control one

Three items from TLDR AI's roundup show where model competition, billing rails, and agent reliability are heading.

Anthropic's AI has moved from stocking shelves to making termination calls in a real US retail job.

Jean-Denis Grèze tells Platformer how Town runs day-to-day work through AI assistants, not middle managers.

Anthropic's own agents turned a shared task into a turf war — a preview of production multi-agent risk.

Novo Nordisk's AWS-powered discovery agents and Meta's local Muse Glimmer show agentic AI scaling both up and down at once.

Nvidia's latest open-source model targets specific use cases, signaling a shift in AI development strategy.

Anthropic's latest update to Claude allows code sessions to share context, streamlining complex development workflows.

AI agents demand significantly more energy than simple prompts, posing a critical sustainability challenge for builders.

Major players coalesce around a shared standard, promising a more interoperable and efficient future for AI agent development.

A recent US court decision permits Perplexity to deploy AI agents for Amazon purchases, signaling a pivotal shift for e-commerce and AI development.

Meta AI's innovative approach uses a secondary agent to manage primary agent memory, promising enhanced performance on complex, multi-step tasks.

OpenAI's new initiative aims to bridge the gap between AI agent prototypes and robust, production-ready enterprise solutions.

MIT researchers highlight the need for tailored environments to unlock agentic AI's potential.

Leveraging frontier models for planning and cheaper models for execution dramatically alters the economics of AI-assisted software development.

A deep dive into the technical and ethical considerations of integrating autonomous AI agents into the open internet's architecture.

Google's new initiative provides developers with no-cost tools and resources for AI agent creation.

Two market leaders are synchronously embedding AI agents into Photoshop and Slack, fundamentally changing the game for builders.

Understanding ACU limits, pricing tiers, and the true cost of Cognition's autonomous agent.

Discover Devin, how it works, and why this autonomous AI engineer could become a critical tool for AI developers.

Discover how to integrate Codex into your AI projects for automated code generation and optimized development, a practical guide for AI builders.

OpenAI's Codex impresses in demos, but certain scenarios render it not just inefficient, but actively detrimental to a project's success.

A practical guide for technical teams looking to integrate Codex without chaos or unpredictable regressions.

We break down when to use OpenAI's Codex and when Claude Code, Cursor, or Copilot might be a better fit for your coding tasks.

OpenAI has relaunched Codex as an autonomous agent. We look at what's changed under the hood and how to use it now.

Codex is no longer just an editor's suggestion. It's an agent that receives a task, runs tests, and returns a pull request without your intervention.

Codex is a powerful tool, but typical errors can turn it into a source of elusive bugs. A practical breakdown for AI builders.