Agentic AIThe briefing
AI-generated illustration
Perplexity launched Teammates, a system that lets multiple agents coordinate tasks, maintain shared project state across sessions, and continue working after a user closes the browser tab. The feature is available now to web users on the Max plan.
Perplexity introduced Teammates that coordinate several agents on the same project. Agents submit changes to a shared Git-backed repository while each runs in its own isolated sandbox. The system separates persistent project state from individual execution environments so work continues without an active user session.
State is kept in Git repositories and a separate Brain memory store. Median submission latency stays under 1.7 seconds and fewer than 2 percent of commits are rejected for stale state. Sessions can be paused, resumed, or forked using Firecracker microVMs and Btrfs snapshots.
Users connect triggers from Slack and GitHub. Each agent runs in its own browser session with permissions managed through existing controls. Perplexity has not detailed how hand-offs prevent agents from exceeding their assigned access.
Teammates is released on the web for Perplexity Max subscribers. iOS support is planned but not yet available. The company has not published enterprise data-retention or audit terms for the new agent feature.
Builders evaluating agent orchestration can test whether Git-backed shared state meets their consistency needs. An illustrative scenario is a team that wants background agents to update documentation while a foreground agent handles user queries. The builder would connect a GitHub trigger, assign separate sandboxes, and measure commit rejection rate over several days. The evidence does not establish how the system behaves when two agents edit the same file simultaneously beyond the reported 2 percent rejection figure, nor does it describe conflict-resolution behavior for non-code artifacts.
Operators reviewing Teammates should verify data-handling terms before connecting enterprise Slack or GitHub workspaces. An illustrative scenario is a support team that wants agents to draft responses from past tickets. The operator would first confirm whether project repositories leave the Perplexity environment and whether Brain memory is retained after an agent is paused. The supplied evidence does not establish enterprise audit logs, data-retention periods, or whether Max-plan accounts receive the same isolation guarantees as any future enterprise tier.
Create a test workspace limited to non-sensitive repositories, connect one GitHub trigger, and run two agents on the same project for 48 hours. Record commit latency, rejection rate, and any permission warnings that appear. The results will show whether the reported sub-2-percent rejection rate holds for the specific workflow and whether additional human oversight steps are required before wider rollout.
Source material behind this briefing. Our practical implications are Ashlr’s interpretation.
This briefing was prepared with AI assistance from the linked sources. The analysis may include interpretation; consult the original sources for complete context. Cover artwork is illustrative.
Ashlr release
AI-generated illustration
Merged bridge jobs added a live resource board to Fleet and mobile parity for sessions and approvals in Phantom. Ashlr Stack gained worktree cleanup reports and command updates.
Agentic AIAshlr release
AI-generated illustration
Phantom gained agent-native fleet commands for lane targets, reassigns, and model sets. Multiple products received new documentation and regression tests.
Agentic AIThe briefing
AI-generated illustration
Genie 4 renders real-time simulated environments with physics and audio. Gemini models steer the scenes so agents can rehearse multi-step tasks before deployment. Access remains limited to trusted testers and planned Google AI Ultra subscribers.
