Topic
An AI system design pattern where a top-level orchestrator agent delegates tasks to specialized subagents, coordinating their work and integrating results.
Anthropic put dynamic workflows in Claude Managed Agents beta on Oct 9, letting one run start up to 1,000 agents over its life. A run has no price of its own, so its model requests count toward the session budget, a dev.to write-up of the docs says.
Reality
Decepticon, an autonomous red-team agent built on LangChain and LangGraph, reached #13 on GitHub Trending for Python with 5,825 stars and 1,100 forks. Its write-up details how a state machine runs offensive tools end to end while leaving an audit trail.
Reality
Manus's rebuilt 2.0 platform adds Cue, an app in which each personal agent has its own email, phone number, wallet and computer. Those agents can spend within set limits. A team that allows Cue at work has to set those limits first.
Reality
GKE Agent Sandbox is generally available and publishes provisioning numbers a team can plan against. The scheduling and durability layers stacked on top of it rest on one public demo, and the deprecation argument lands there.
Reality
An engineer writing on dev.to swapped the coordinating LLM in a multi-agent system for an XState machine and typed receipts, reporting a 70% token cut. Whether that transfers depends on what your coordinator cost.
Reality
Business Insider has practitioners spending a fifth to a third of their working week supervising AI agents, work that looks like first-line management inside roles whose titles and pay bands have not moved with it.
Reality
Tencent's Hunyuan Speech team put a swappable agent model behind a conversation model that re-decides every second whether to talk, and reports best-in-test timing on Full-Duplex-Bench v3 with a task-accuracy shortfall it describes only as slight.
Reality
The desktop app creates a real git worktree for every task and launches a CLI agent inside it. The isolation is the same git command you already have. The product is the review loop built around it.
Reality
Anthropic's redesigned Projects beta enforces one ceiling, 200 new threads a day, and it has not published a usage figure for a single thread. Subscribers see what a project drew only afterwards, in the Usage tab.
Reality
shinpr has taken claude-code-workflows through 133 releases, and the recent ones delete structure the models no longer need. His session reader then found three mandatory steps in his own repository that never ran.
Reality
The 14.13.1 release captures failed agent runs as incidents, promotes them into skill fixtures, and accepts an edit only when one split improves and neither regresses, all under an enforced dispatch budget.
Reality
Factory raised $200 million to run language ports and migrations with orchestrated agents. Its projects open with a Readiness Report that checks whether the customer's repository is organized and documented well enough for those agents to work in.
Reality
Kill an orchestrator and the fleet stops; kill a coordinator and every agent keeps typing. Foremerge's maintainer submitted the tool to a list of orchestrators anyway, because no list exists for the other layer.
Reality
A dev.to post-mortem on Hermes puts the constraint plainly: workers never talk to workers. The promised comparison with LobeHub is missing from the text.
Reality
James Coombs's pipeline now re-tiers any ticket touching three or more integration layers before an agent starts, a rule that came out of one ticket filed as an error handling improvement that ran 5 hours and 17 iterations.
Reality
Anton Brilliantov spent eight parts specifying agent handoffs down to a single acceptance command. His ninth names the task shapes where those facts are still unknown, and the reading that has to happen first.
Reality
Cursor's Projects beta gives each body of work a coordinator agent on its own cloud machine. Its internal deployment shows where the review load lands, and its productivity figures arrive without a baseline.
Reality
Salesforce put Data 360, Informatica, MuleSoft, Tableau, Agentforce and Guardian under one new control plane on Thursday. The loops that run agents inside its own tools still belong to Mastra and the Claude Agent SDK.
Reality
Lauren Tan's MIT-licensed plugin routes a described task into one of 23 playbooks, but the part that decides whether any of it transfers is the per-project skill that drives the real product and recognises failure.
Reality
By Mistral's own account, agents handed one Fortran subroutine each returned working C++ that preserved the old design, and the multi-agent run stalled on hard bugs until a human operator unblocked it and reviewed the pull requests.
Reality
Build · September 1, 2026 · Confirmed2 publishers