Inngest Deferred Functions, Trigger.dev AI SDK 7, and n8n's AI Assistant Experiment
June 1–7, 2026
Inngest launched Deferred Functions for fire-and-forget background work, Trigger.dev reached v4.5.0 release candidate 5 with AI SDK 7 support and a new tools option for chat agents, and n8n shipped multiple patch releases across its active branches plus an experimental AI Assistant workflow preview.
Inngest Deferred Functions: fire-and-forget with typed payloads
The most significant new capability this week came from Inngest. Deferred Functions let you spin off independent background work from inside any running function by calling defer("some-id", { function, data }). The parent function keeps executing immediately — the deferred run gets its own retries, concurrency, and step state, and its payload is fully typed via a Standard Schema. This is a cleaner pattern than spawning a separate function via the API or using a queue primitive: you stay inside Inngest's execution model, and the deferred work inherits the same observability and tooling.
Inngest also shipped a pair of server releases. v1.26.0 promotes the API CLI to a top-level beta command (npx inngest-cli@latest api --help) and adds support for parsing span attribute data from Langfuse, which is useful for teams that already export traces to Langfuse and want to correlate them with Inngest runs. v1.25.0 added a v2 REST endpoint for listing function runs by event ID, plus more Open Inference attributes. The v1.24.0 release from earlier in the week surfaced deferred runs in the run details UI and fixed several queue edge cases, including quarantining partitions for deleted accounts.
Trigger.dev v4.5.0 release candidates: AI SDK 7 and chat agent tools
Trigger.dev continues to iterate fast on its v4.5.0 release candidates. The latest, v4.5.0-rc.5, adds AI SDK 7 support: the ai peer dependency range now includes v7, and both the chat.agent and chat surfaces work against v7's ESM-only build. On v7, you need @ai-sdk/otel installed separately. The rc.4 release that preceded it introduced a tools option on chat.agent — declaring your tools there threads them into the SDK's internal convertToModelMessages, so each tool call is properly represented in the conversation history sent to the model. The rc.3 release from Monday laid the groundwork with general improvements to the AI SDK integration.
Meanwhile, the Trigger.dev runs page got a live update: statuses change in place, a banner flags new runs, and parent tooltips now show child-run breakdowns. This is a quality-of-life improvement for anyone monitoring long-running workflows.
n8n: patches across the board and an AI Assistant experiment
n8n shipped a steady stream of patch releases across its active branches this week. The most interesting change is in v2.25.3: an experimental AI Assistant empty state that shows workflow previews when no workflows exist yet. It's behind a feature flag, so you'll need to opt in, but it signals where the product's AI-assisted authoring is heading.
The v2.25.2 release fixed a bug where external agent channels weren't correctly using the user ID for episodic memory, and the v2.25.1 release that kicked off the 2.25 line earlier in the week added a validate_node_config MCP tool for per-node validation — a useful addition for anyone building custom MCP server integrations. The v2.24.0 release from Tuesday added decompression output size validation in the Compression Node and fixed a code-signing issue for the macOS desktop app.
Security fixes landed across older branches too: v1.123.52 rejects unsafe property tokens in in-isolate $jmespath expressions, and v1.123.51 quiets VM expression engine logs on the happy path. The v2.25.4 and v2.23.4 patches are dependency bumps with no user-facing changes, and v2.22.6 and v2.23.2 both fixed a licensing paywall appearing on the Insights page for users who already had a license.