Posts

OpenClaw just dropped its most substantial release in months, and if you’ve been watching the agentic AI space closely, v2026.4.5 is worth your full attention. This update ships three headline features — Dreaming Memory, built-in media generation, and a prompt caching overhaul — plus a significant provider shift that reflects where the LLM landscape actually stands today. Dreaming Memory: Background Consolidation While You Sleep The biggest conceptual leap in v2026.4.5 is Dreaming Memory. Inspired by how biological memory consolidates during sleep, the feature runs background memory processing sessions that compress, link, and surface important context across long-running agent deployments. The output surfaces in a new Dream Diary UI — a timeline of what the agent “processed” overnight, complete with connection maps between memories. ...

When 512,000+ lines of Claude Code’s source landed on the open internet on March 31, Anthropic’s response was measured, careful, and clearly drafted by people who were thinking about something other than just the technical disclosure. They were thinking about the S-1. That’s the core observation driving The Register’s Kettle podcast deep dive this week — and it’s an uncomfortable one. When a frontier AI company responds to a major source leak with language calibrated for investor relations rather than security disclosure, you learn something about what’s actually being prioritized. ...

Xiankun Wu, CEO of Kuse, is exactly the kind of technologist the AI industry profiles approvingly. He built AI employees using OpenClaw. They work nonstop, never complain about timezones, and cost a fraction of their human equivalents. He deployed them. He was proud of them. His human team quietly created a private Slack channel without the AI employees. What Actually Happened According to Business Insider’s reporting, the Kuse team didn’t rebel against the AI coworkers in any dramatic sense. There was no manifesto, no confrontation. The humans simply created a separate channel — a small digital room where they could have conversations without AI involvement, without everything being logged, analyzed, and fed back into workflows. ...

There’s a rule in computer security called Kerckhoffs’s Principle: a system must remain secure even if everything about it is public knowledge. Anthropic, a company that has staked its entire identity on being “safety first,” just shipped a product that violates that principle in a way that’s almost poetic in its mundaneness. Not through a zero-day exploit or a sophisticated attack chain. Through a performance shortcut. What Actually Happens Claude Code lets operators and users configure deny rules — a list of commands the agent is never allowed to run. You can say “never execute rm,” “never run curl,” “never touch /etc/.” It’s the primary mechanism for keeping an AI agent that has shell access to your machine from doing something catastrophic. ...

Somewhere in your company’s recent strategy deck, there’s a slide about multi-agent AI systems. It probably has a diagram with six or eight boxes connected by arrows, each box representing a specialized agent — one for research, one for synthesis, one for outreach, one for quality control. It looks clean. It looks powerful. It looks exactly like the microservices architecture slides that were circulating in 2014. InfoWorld is issuing the same warning now that engineers were quietly issuing then: distributed complexity is not a free upgrade. You have to earn it. ...

For roughly three years, the AI pricing model worked like an all-you-can-eat buffet. Pay $20 a month. Use as much as you want. Get smarter over time. The math worked — barely — because most users were having conversations, writing emails, and generating the occasional image. Human-paced usage is predictable. It’s manageable. It turns out it’s also completely incompatible with the next phase of AI. The Fundamental Incompatibility Autonomous AI agents don’t sleep. They don’t pause between messages. When you set a coding agent loose on a codebase, it might invoke 500 tool calls in an hour. A scheduling agent running in the background 24/7 isn’t “one user” in any meaningful billing sense — it’s a cluster of compute consumption that no $20/month plan was designed to absorb. ...

Ant Group’s blockchain division has published the clearest vision yet of what it means for AI agents to be autonomous economic actors — not just task executors working on behalf of humans, but entities that hold assets, discover counterparty agents, and settle transactions independently. Anvita, launched by Ant Digital Technologies, is a dual-platform system designed to provide the infrastructure layer for machine-to-machine commerce. The Two Platforms Anvita Chain handles the asset side: tokenization infrastructure and settlement rails. It’s the blockchain layer that gives agents something to actually transact with — tokenized assets, stablecoins, programmable payment logic. ...

There’s a particular kind of tedium that every AI engineer knows intimately: the prompt-tuning loop. You write a system prompt, run your agent against a benchmark, read the failure traces, tweak the prompt, add a tool, rerun. Repeat this a few dozen times and you might move the needle. It’s grunt work dressed up in Python files. AutoAgent, built by Kevin Gu at thirdlayer.inc, proposes a direct alternative: don’t do that work yourself. Let an AI do it. ...

Less than a year ago, Anthropic characterized China in a regulatory filing as an “enemy nation” — one of the key justifications for why the company should be treated differently from its competitors in US AI policy discussions. This week, Chinese developers are celebrating the 512,000-line Claude Code TypeScript source leak as a detailed architectural roadmap for building the domestic AI coding agents that would compete directly with Anthropic’s products. ...

When developers talk about building AI agents that get smarter over time, they usually mean one of two very different things — and they rarely realize the ambiguity. LangChain’s Harrison Chase published a framework today that finally gives the field a shared vocabulary: continual learning for AI agents happens at three distinct layers, and conflating them leads to systems that are overbuilt for simple problems or structurally incapable of solving hard ones. ...