The Week Agents Got Wallets and Watchdogs

Meta shipped its first coding agent. Google's Gemini Spark now runs 24/7 on cloud infrastructure. And $60 million just poured into startups that do one thing: secure and pay the non-human workforce.

This was the week AI agents stopped being a demo and started being infrastructure. Here's what happened.

Featured: Meta Enters the Terminal Race

Meta launched Muse Code on Aug 5 — its first terminal coding agent, built on the new Muse Spark 1.2 model. This is Meta's direct shot at Claude Code and OpenAI's Codex.

The details that matter:

• Handles whole repos — plans changes, writes code, validates results across large codebases, spinning up parallel subagents per task.
• Built to run long — in one internal test, Muse Spark 1.2 optimized GPU kernels over 1,000+ tool calls and up to 24 hours of continuous work.
• Pricing is the flex — pay-as-you-go at $1.25/M input tokens, $4.25/M output. Meta says its contributor tier comes in "more than 10 times cheaper" than that.
• Beta for macOS and Linux — one command to install.

Three major labs now have terminal agents. The coding-agent price war is officially on.

"One command to install. A whole repo to chew on. 24-hour agent runs. That's the new baseline for coding agents."

Google's Agents Never Sleep Now

Gemini Spark — Google's 24/7 personal agent, built on Gemini 3.5 Flash — runs entirely on Google Cloud, so it keeps working even with your phone and laptop off. It uses Chrome to carry out web tasks: booking viewings, preparing flights, handling recurring work. You reach it by messaging a designated Gmail address.

Google also pushed the Gemini Enterprise Agent Platform to GA this quarter — agents that keep state across days, log every operation, and use dedicated Agent Identity credentials with minimal permissions. The enterprise version of "my agents ran all weekend."

The Non-Human Workforce Gets a Banking System

When agents transact, someone has to handle identity, security, and payments. This week, three startups raised for exactly that:

• Natural — $30M Series A ($40M total) — building "Stripe for AI agents": transaction rails so agents can actually pay for things. Akamai joined as an investor in Hush, below.
• Hush Security — $30M Series A ($41M total) — securing the "non-human workforce": agent identity, credential governance, observability for bots.
• Varonis Agent Intent-Based Access Control — launched after one rogue bot deleted a production database. Access control now keyed to what the agent intends to do, not just who it is.
• Drata AI Agent Governance — limited availability; discovers, monitors, and governs agents inside orgs, starting with Anthropic-built agents.

That's a signal. The money is following the agents.

Quick Hits

• OpenAI Agent Plugins — an interoperability standard with partners (AWS, GitHub, VS Code, Vercel) so agents call shared capabilities consistently across tools.
• Hermes Agent v0.20.0 — Nous Research shipped Aug 3; the releases keep coming roughly weekly now.
• Helium 10 MCP connector — live Amazon seller data now queryable in plain English from Claude and ChatGPT (Diamond plan).
• AgentRadio (Coral AI Labs) — four coordinating agents outperformed Claude Opus 4.8 on enterprise coding tasks using async messaging.

Tool of the Week: Muse Code

Worth a weekend pilot on a non-critical repo. Install with one command, point it at a mid-size codebase, and watch it plan, write, and validate with parallel subagents. The contributor tier pricing is the interesting part — Meta is clearly willing to undercut on price to gain share.

The Takeaway

The agent race is past the model layer. This week was about rails: who codes with agents (Meta), who runs agents for you (Google), who pays them (Natural), and who keeps them honest (Hush, Varonis, Drata). If you run agents for a business, the security and governance layer is where you should be looking next — the tools just arrived.

Wednesday's deep dive: what it actually takes to let agents handle money — the new payment rails, and the guardrails that come with them.

Automate & Grow with A.I.
Written by Michael Devellano — building AI automation systems for founders and agencies.
Subscribe for weekly AI automation insights