Reliability infrastructure gets a $550M vote, and the governed-data layer becomes its own product

> ../signals/2026-09-19.md

── Signal one · Temporal raises $550M Series E at a $12.55B valuation ──

Temporal, whose durable-execution engine is what keeps long-running workflows alive across restarts and failures, closed a $550M Series E at a $12.55B valuation. The tell isn't the number, it's the category: reliability infrastructure for agents is now fundable at scale on its own, separate from the models. If your agents run for minutes or hours, the thing that survives a mid-run crash without losing state is worth more attention than the next benchmark.

── Signal two · Salesforce unveils AIforce and Headless 360 at Dreamforce ──

Salesforce introduced AIforce and Headless 360, letting agents act on Salesforce data without the traditional human-facing UI, framed as part of an "Enterprise AI Harness." It also expanded Agentforce and completed its acquisition of Fin. Headless access is the structural shift worth noting: software increasingly exposes a surface built for an agent, not a person, which changes how you think about permissions and audit on that surface.

── Signal three · Dipp AI ships a runtime Data Control Gateway ──

Dipp AI published a Data Control Gateway that redacts, region-pins, and checks every agent call against training-exclusion and boundary policies, refusing routes that would break the rules. This is the governed-data-access pattern moving from something each team builds itself into a product you can buy. Worth evaluating before you hand-roll your own policy layer in front of agent data access.

── Signal four · RavenDB Quill and Alteryx target the data-and-token-cost layer ──

RavenDB launched Quill, connecting agents to enterprise SQL without a data migration, and Alteryx wired governed business logic to external agents specifically to cut token costs. Two different vendors converging on the same insight: the expensive part of an enterprise agent is rarely the model, it's the data access around it, and doing that access cleanly is a cost lever most teams aren't pulling.

── Signal five · The agent security-and-governance stack keeps filling in ──

Composio raised $25M Series A for agent infrastructure and London-based HelmGuard closed $7.3M for an agentic governance, risk, and compliance platform. These are two more data points in a pattern we've tracked all quarter: between April and September 2026, roughly $435M went into enterprise agent security and governance across a dozen rounds. The guardrails are getting funded before the agents they guard reach scale.

── What to do with this ──

◆ Signal I: If your agents run for minutes or hours, durable execution that survives a mid-run crash without losing state deserves more attention than the next benchmark.

◆ Signal II: Expect more software to expose a surface built for an agent, not a person. Rethink permissions and audit on headless surfaces.

◆ Signal III: Evaluate a governed-data-access product before you hand-roll your own policy layer in front of agent data access.

◆ Signal IV: The expensive part of an enterprise agent is rarely the model. Treat clean data access as a cost lever.

◆ Signal V: The guardrails are getting funded before the agents they guard reach scale. Budget for detection of agent manipulation now.

── End of signal ──

Sources for this brief: aiagentsdirectory.com AI Agents News Briefs (September 14-15, 2026); Dreamforce 2026 coverage; enterprise agent funding roundups, September 2026.

ORBIRESEARCH