Fetching from the wire…
Markets2026-09-26 · source-backed
The September 24 deal brings in the Substrate hypervisor, the Architect Kubernetes platform and AGX sandboxes. Substrate moves long-running workloads between machines, so voice, video and physical-AI agents survive host failures, spot reclamations and cross-zone moves without dropping the session. CEO Russ d'Sa compares it to telcos handing calls between cell towers. An agent SDK company buying the runtime layer underneath its agents is a pattern I expect to repeat.
Each link below shares sources, entities, or timing with this story.
The bug hit query() calls using hooks, can_use_tool or SDK MCP servers. When a subagent finished just before the turn's result arrived, the SDK closed stdin early and the model reported the tool as refused. It now waits for the CLI to report idle via session_state_changed, cap...
v2.1.0 adds three set-once surface-identity headers, with the append-only X-Mem0-Client carrying name/version per layer outermost first, so a plugin calling the SDK reports the whole chain rather than only the last speaker. The stack is bounded by dropping whole entries rather...
If you wrote an MCP server before July, it's on a protocol shape the maintainers have already removed. Not deprecated-with-a-migration-window. Removed from the spec. MCP lead maintainers David Soria Parra and Den Delimarsky published an updated roadmap on August 22, and the re...
Released yesterday, one day after 0.21.0, v0.21.1 adds model call timeouts, run-scoped sandbox working directories, network-disabled Docker sandboxes and Modal sandbox resource options (GitHub). It also fixes the core honoring exact call-approval decisions and rejecting partia...
The releases promote AG-UI from RC to stable, fix stateless replay of reasoning-paired tool calls, and add explicit prompt cache breakpoints for GPT-5.6 models. Harness agent, mode/todo providers and tool approval graduated out of experimental, plus an Azure Cosmos DB semantic...
Version bumps are usually boring. This one isn't, because of what it concedes. The OpenAI Agents SDK hit v0.13 with an any-LLM adapter, breaking the assumption that you'd be running OpenAI models inside OpenAI's framework. Requesty's SDK comparison has the details: alongside t...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.