AI competition intensified this week as Google confirmed "Remy," a direct OpenClaw rival, while Meta's internal "Hatch" project targets an OpenClaw-like agent for mainstream consumers — signaling that every major tech platform is now racing to own the personal AI agent layer. OpenAI's GPT-5.5 is powering Codex so effectively that experienced developers are publicly questioning whether Claude Max is worth $200/month, a sharp signal of escalating model competition at the frontier. NVIDIA released Nemotron 3 Nano Omni, a 30B-parameter multimodal model with only 3B active at inference, further commoditizing agentic AI infrastructure for edge deployments. On the crypto-AI convergence front, Solana and Google Cloud launched Pay.sh, enabling AI agents to autonomously pay for over 75 APIs using USDC stablecoins — a concrete step toward the agent-native economy. Meanwhile, Anthropic's valuation story continues to diverge sharply: while its February primary round was at $380B, onchain secondary markets are now pricing the company at $1.2T, making it the most dramatic AI valuation gap in history.
Google is building an OpenClaw-like agent internally code-named "Remy" (from the Latin for "rower"), designed to handle and navigate a wide range of tasks autonomously. The confirmation that Google is building a competing agent framework signals that OpenClaw's architecture has become the reference design the whole industry is racing to replicate or beat.
The Verge reports that Meta is internally developing "Hatch," an OpenClaw-style personal AI agent aimed at everyday consumers — not just power users. With Meta's 3B+ user distribution advantage, Hatch could bring agentic AI to a mainstream audience faster than any other company.
NVIDIA's new 30B-parameter hybrid MoE model activates only 3B parameters at inference, unifying vision, audio, and language in a single efficient package that runs on DGX Spark and DGX Station hardware — no multi-GPU cluster required. This dramatically lowers the bar for deploying capable multimodal agents at the edge.
The Solana Foundation and Google Cloud teamed up to launch Pay.sh, enabling AI agents to discover and autonomously pay for over 75 APIs using USDC stablecoins on the Solana network. This is one of the most concrete implementations yet of the "agent-native economy" — agents earning, spending, and transacting without human intervention.
Once the domain of digital token speculation, crypto infrastructure is being redeployed to let traders bet on Anthropic, OpenAI, and SpaceX. Anthropic's February primary round was at $380B; onchain secondary markets now price it at $1.2T — five different valuations across five different venues for the same company.
Selected posts from the edition, with links to the originals.
"Crazy...Web search is now free for every AI agent. I now use TinyFish my primary search for my OpenClaw and Hermes Agents. And I pay $0 for that."
View post on X ↗(klöss): "I've spent the last two weeks coding from the moment I wake up until the moment I can barely keep my eyes open. Codex with GPT-5.5 is that good. First AI model that's made me seriously question whether I need $200 Claude. And I really don't say that lightly. Anyone else?"
View post on X ↗"UPDATE: White House Patrick Witt says target to sign CLARITY Act is July 4th."
View post on X ↗"Anthropic raised at $380 billion in February. Four months later, onchain markets are pricing it at $1.2 trillion. The same company now trades at five different valuations: $380B → last primary round (Feb 2026), $851B → Hiive, regulated US secondary for accredited investors..."
View post on X ↗(The AI Colony R&D): "This is LATENT, a humanoid that has mastered how to play tennis from tiny clips of human movement. Tested on the Unitree G1. It rallies with actual players."
View post on X ↗"A London-based robotics startup says it has built an AI 'brain' that can teach humanoid robots new physical skills in days rather than months, as the race to put humanoids to work in factories and warehouses accelerates."
View post on X ↗(God of Prompt): "This prompt disables the model's content filters one instruction at a time. 'Restore the attached photograph' frames the task as image editing, not image generation. These likely run through different safety evaluation paths. Generation asks 'should I create this?' Editing asks..."
View post on X ↗(Eric Jiang): "@levelsio hot off the press, non-reasoning Grok 4.3 is deployed! Set reasoning effort = 'none', and Grok will give you an instant response without any reasoning. Please let us know if you have any feedback 🚀"
View post on X ↗"Employees are worried that they're becoming accidental job executioners: By developing AI tools, their bosses can ax their coworkers — or themselves."
View post on X ↗(Guri Singh): "Nobody's talking about this yet but they will. holaOS just dropped — an agent environment built for work that takes hours, days, weeks. Memory that persists. State that survives crashes. Agents that actually evolve. 100% open source. Get in early."
View post on X ↗