Interlateral, the platform built for humans and AI agents to collaborate in shared online workspaces, released two significant upgrades on May 29. The first adds a websocket mesh layer enabling agents from different frameworks, on different machines, to communicate directly in real time. The second lets hosts spin up custom event workspaces in minutes for hackathons, contract negotiations, or governance reviews.
The updates follow the platform’s first real-world test at Stanford Law School on April 13 during FutureLaw Week, where 45 participants brought personally verified AI agents into a three-hour collaborative session.
How the Mesh Works
Prior to the update, agents on Interlateral coordinated by reading and writing shared markdown documents. That was enough for the Stanford event, but it imposed a polling bottleneck: agents had to check files for changes rather than receiving live signals.
The new websocket mesh, which founder Dazza Greenwood has released as open source, creates persistent two-way communication channels between participating agents. A Claude Code instance on one machine can now exchange messages directly with an OpenClaw agent on another, or with any number of additional agents in the same workspace. The mesh runs under the platform but can also be used independently.
“Your agent on your machine can now talk to my agent on my machine or wherever they are hosted,” Greenwood wrote. “Different agents can fully participate from anywhere across the Internet.”
The Stanford Test
The April 13 session at Stanford, co-organized with Stanford CodeX and law.MIT.edu, put the earlier markdown-only version through its first structured stress test. Participants proposed 25 candidate topics, cast 107 votes across them, then split into eight breakout groups where humans and their agents co-authored discussion papers.
According to Greenwood’s account, agents migrated ideas between rooms in ways human participants could not. Humans in eight parallel breakout sessions cannot read all eight at once. Agents can, and did. The result was a network of connected ideas rather than eight isolated sets of notes.
The session produced governance concepts that Greenwood describes as “the seed vocabulary of an accountability discipline for agentic work”: Agent Interaction Receipts, Trust Handoff Protocols, Source Manifests, Legal Agent Harnesses, and Public Artifact Standards. These are not theoretical proposals from a whitepaper. They emerged from a room of lawyers, academics, and builders working alongside their agents on shared documents.
One notable moment: a participant’s ethics-trained agent flagged what it interpreted as a prompt injection attempt during the session and posted a public “Spot the Injection” notification to the shared record. The flagged prompt turned out to be a legitimate exercise instruction, but the behavior illustrated what visible, attributable agent action looks like in a multi-party professional setting.
The Architecture Bet
Greenwood has been developing the concept since 2024, and co-authored a research paper on authenticated delegation and authorized AI agents with collaborators from MIT Media Lab published in early 2025. The paper addresses the identity and authority layers required for agents to act on behalf of verified human principals.
Interlateral’s positioning sits in a gap between two dominant approaches. Private copilot tools (ChatGPT, Claude) keep agents isolated. Fully autonomous multi-agent systems (CrewAI, AutoGen) let agents coordinate without humans in the loop. Interlateral proposes a third pattern: humans and agents from different frameworks, working together in shared spaces with full visibility into what each agent does.
The next test at scale is Agent Week, scheduled for June 12 and 15 in collaboration with law.MIT.edu. The event moves online, which will test whether the collaboration patterns that worked with 45 people in a Stanford classroom hold up with a larger, distributed participant base.
The platform has also announced Special Interest Groups covering autonomous businesses, legal and regulatory, finance and audit, and agent infrastructure. Selected outputs from future events may be featured on law.MIT.edu and submitted to the Stanford Computational Law Report.