Arcade.dev | The Actions Runtime

Ship agents. Not auth infrastructure. |

[Arcade.dev | The Actions Runtime @TryArcade]

On 11/3, https://t.co/dBXpdhwjiD and @LangChain_OSS are hosting a hands-on workshop in Toronto on building agents that can pass a security review, with user-scoped authorization, policy checks, human approval, and LangSmith tracing.

Request a spot: https://t.co/u4PFxBtb0M

[Arcade.dev | The Actions Runtime @TryArcade]

@catmanyau We had considered including sample prompts to solve the "what to ask" piece. This is a good indicator that we should think about it again, appreciate it.

[Arcade.dev | The Actions Runtime @TryArcade]

The new Arcade Plugin takes you from signing up for https://t.co/dBXpdhwjiD to your first tool call in one install. Add it to Claude Code, Cursor, VS Code, Codex, and more, then sign in and ask your agent for something real, with no gateway to set up.

https://t.co/IjVingqFc6

[Arcade.dev | The Actions Runtime @TryArcade]

@SamPartee , River Stevens, @ptdamiba, and Valerie Fanelle put the research behind https://t.co/dBXpdhwR8b’s Tool Recommendation on arXiv. These are the details behind all the token cost reduction and ToolRet scoring. For the nerds: https://t.co/qiABvBrD0H

[Arcade.dev | The Actions Runtime @TryArcade]

@kamathematic @zeddotdev @DhravyaShah Tool search...with receipts https://t.co/qiABvBrD0H

[Arcade.dev | The Actions Runtime @TryArcade]

This October, we’re sharing Agent Horror Stories, real moments when an AI agent struck terror into a user’s heart. First up: @ptdamiba told his agent to “kill the clusters” so he’d stop getting billed. The billing stopped, because the agent deleted every cluster. 🎃

[Arcade.dev | The Actions Runtime @TryArcade]

“Even if you pretend that you don’t have agents running inside of your organization, you do.” @nbarbettini on the CyberBytes podcast with David Huish, from governing agents to how fast https://t.co/dBXpdhwR8b has grown since his last startup. https://t.co/YIYcgw5wKK

[Arcade.dev | The Actions Runtime @TryArcade]

If you’re in San Jose for AGNTCon + MCPCon, join https://t.co/dBXpdhwR8b and @mcpjams the night before, Oct. 21 from 4 to 7 PM, for lightning demos, drinks, and people building with agents and MCP. 21+, no badge needed. Request a spot: https://t.co/mmlzwhhajI

[Arcade.dev | The Actions Runtime @TryArcade]

https://t.co/dBXpdhwR8b released Tool Recommendation last week. Here’s the simplest way to explain it: agents searching a big tool catalog walk every aisle at Home Depot to find a hammer, and Tool Recommendation sends them straight to it. https://t.co/RoWqvYUsKH

[Arcade.dev | The Actions Runtime @TryArcade]

The https://t.co/dBXpdhwjiD team is heading to AGNTCon + MCPCon in San Jose, 10/22 and 10/23. Catch our four sessions, including a live demo where an agent tries a wire transfer and gets held for approval. Save 25% with OUTREACH25: https://t.co/Es5LYfYvSl @AgenticAIFdn

[Arcade.dev | The Actions Runtime @TryArcade]

https://t.co/dBXpdhwjiD’s new Tool Recommendation gives agents your whole tool catalog for up to 37% less in token costs, with up to 91% fewer input tokens per session on every model we tested.

Available now in Arcade Cloud: https://t.co/RoWqvYTUV9

[Arcade.dev | The Actions Runtime @TryArcade]

Thanks to @VentureBeat for the exclusive on Tool Recommendation’s launch. @carlfranzen nailed the problem it solves: “Connecting an agent to more business systems also makes its instructions larger and more expensive.”

https://t.co/Mz43NNTGAE

[Arcade.dev | The Actions Runtime @TryArcade]

Think you've got what it takes to play "Rogue Agent" and add your initials to the leaderboard? Prove it: https://t.co/1Y7GhwfomZ

[Arcade.dev | The Actions Runtime @TryArcade]

https://t.co/dBXpdhwR8b has three new teammates and a lot of confetti to clean up. Welcome to Thanh-Ha Nguyen (ex-Okta), Don Tregonning (ex-Splunk), and Alex Laughnan (ex-Asana), who all survived the confetti cannon.

We’re hiring for the next round: https://t.co/RtA8jUkfpC

[Arcade.dev | The Actions Runtime @TryArcade]

If your afternoon in SF is open, there’s still time to join https://t.co/6U6bfSuDK8, @mastra, and @elastic for a hands-on MCP for GTM workshop today at #SFTechWeek. Build a sales agent and see four ways a discount request goes wrong, plus how to catch them. Register: https://t.co/mWjswVlpXD

[Arcade.dev | The Actions Runtime @TryArcade]

Our very own @kamathematic tried to announce https://t.co/6U6bfSuDK8’s new Private Registry before Pascal finished making coffee. Open any Arcade MCP URL in your browser to see your tools, what you can access, and request anything missing, right from the page.

https://t.co/E166y4BTFj

[Arcade.dev | The Actions Runtime @TryArcade]

Thanks to the banking, insurance, payments, retail, and healthcare leaders who joined https://t.co/dBXpdhwR8b and @LangChain for dinner in Toronto. Determinism was the hot topic, since models can be probabilistic but the controls on what agents do need to be consistent.

[Arcade.dev | The Actions Runtime @TryArcade]

Traders can't approve their own trades, and in a new @Forbes piece, https://t.co/6U6bfSuDK8 CEO @TheMostlyGreat argues AI agents shouldn't approve their own actions either. An agent on a shared service account can act with more access than any one employee would get.

Read the full piece: https://t.co/oYzJQzXgSq

[Arcade.dev | The Actions Runtime @TryArcade]

Builders are already adopting MCP's July 28 spec. In https://t.co/dBXpdhwjiD's ToolBench rescan of 90,000+ open-source servers, the 5.8% using its stateless request handling score about eight points higher on average.

Full findings: https://t.co/0GG3TQvgRD