SandboxAQ Switch, Google agent billing, and BugHerd MCP ship today

Three launches today put agents into the places teams already work: shared chat rooms, cloud billing limits, and website feedback boards that can close their own tickets under review.

SandboxAQ released Switch, free software for putting AI agents into the collaboration tools companies already use. It turns channels in Slack, Microsoft Teams, Discord, Telegram, and Mattermost into shared “rooms” where humans and agents keep the same context, history, participants, and rules. Teams bring their own agents from Claude Code, Google ADK, LangChain, OpenAI tooling, and similar frameworks through MCP, APIs, and adapters. Switch ships no models of its own; the server runs on customer-controlled infrastructure, with a desktop Switch Console for everyday operations. 12

The license is Apache 2.0 with the Commons Clause: free for enterprise, personal, and product use, but you cannot sell Switch itself as a hosted service whose value comes mainly from Switch. Start with one room and one agent on a non-destructive workflow, and watch who can still act when people leave the channel.

Google Cloud shipped expanded billing flexibility and cost controls for agent workloads across Gemini Enterprise and developer tools such as Google Antigravity and Android Studio. Project-level spend caps can pause agent API calls at a hard monthly limit, with email alerts at 50%, 80%, and 100%. Early anomaly detection flags AI spending deviations and points at the top three driving SKUs. A pay-as-you-go consumption edition for Gemini Enterprise is available for select customers and rolling out more broadly soon, with no upfront base subscription fee. Deferred execution pricing for eligible agent workloads is still coming soon, with Google claiming up to half the inference cost during off-peak windows. 3

Treat the shipped caps, overage controls, and anomaly tools as usable today. Treat deferred pricing and the broad PAYG rollout as still arriving. Set one project cap before you scale background agents, or the first runaway loop will set it for you.

BugHerd launched an AI suite built around BugHerd MCP, plus similar-task detection, auto-title generation, and auto-tagging. BugHerd says the MCP connects Claude, Cursor, ChatGPT, VS Code, Windsurf, and similar tools to every feedback task. An agent can read the board, triage website feedback, make changes in a codebase, CMS, or design tool, mark the task done, and post a summary comment for review. 4

These are company press-release claims, so verify the install guides and write permissions on a staging site before letting an agent close client tickets. Keep a human review gate on the summary comment: marking done is not the same as shipping correctly.

Taken together, today’s launches are about shared rooms, spend limits, and review loops. The useful next step is not a bigger agent fleet. It is one room, one budget, and one human checkpoint you can inspect.