AI agents become first-class principals across the development stack, from secrets to API governance to CMS.

The week in development

The development platform layer spent this week formalizing what it means for an AI agent to be a legitimate entity with identity, permissions, and a time budget. HashiCorp Vault agentic IAM reached general availability, letting agents request short-lived credentials under policy-enforced identity. Speakeasy started tracking every agent with its own history page, and shipped session quarantine so a policy violation can freeze an agent immediately. The infrastructure is moving from treating agents as service accounts toward treating them as principals with distinct lifecycles.

The second pattern was agentic loops closing feedback cycles that used to require human handoffs. Rollbar Resolve now reads a production error, finds the fix in your codebase, runs your tests, and opens a pull request — no human touches it until the PR review. Workato extended its Genie agents to 30-minute task windows, unlocking deep research and large document generation that were previously cut off mid-run. The floor on what an agent can complete unattended is rising.

Leaders

Sanity made its most significant architectural decision this week: the Context v2.0.0 release moves Insights to a dedicated Context app and deprecates the Studio plugin, formally splitting the product into Studio (human authors) and Context (AI agents and MCP). This is a product bifurcation — the same content, two distinct surfaces, each optimized for its audience. Sanity's MCP server hit v2.31.0 in the same week, adding local schema editing and exact dataset names.

HashiCorp Vault agentic IAM reached general availability, meaning AI agents can now operate under the same short-lived credential model that Vault applies to infrastructure. Agents get scoped, time-limited credentials, and the audit trail covers both human and agent access in the same log. The GA designation matters here: enterprise customers needed production-ready status before building on this.

Rollbar shipped Resolve in public beta — the agent reads a production error, finds the fix in the codebase, runs the test suite against it, and opens a pull request. Error tracking has always been about surfacing what broke; Rollbar is now positioning the product as the thing that closes the loop. The difference between observe-and-alert and close-the-loop is substantial for on-call workflows.

Speakeasy shipped two notable agent governance features: per-agent identity pages that give every agent its own history alongside human users, and session quarantine that freezes an agent outright when a risk policy is violated. Existing violation handling denied individual events; quarantine stops the session. Both features reflect a shift toward treating the agent session as the unit of governance, not the individual API call.

Workato extended Genie agents to 30-minute task windows and shipped an Agentic Headless API in open beta, letting teams deploy Genies outside of Slack, Teams, or Workato GO into any surface via API. The time extension matters for workloads — deep research and large document pipelines that previously timed out can now complete. The headless API decouples the agent from the chat interface that most enterprise automation tools currently require.

Wildcards

Rivet shipped Dynamic Apps, a library for deploying apps that users generate through AI — powered by V8 isolates, scales to zero, retains state across sessions, and is designed for multi-tenant deployment where each user's generated app is isolated. This is infrastructure for a category that barely exists yet: user-generated applications as a product pattern. It's early, but the V8 isolate approach is technically sound for the security boundary required.

QuestDB shipped both 10.0 open source and Enterprise 4.0 in the same week. QuestDB 10.0 introduces QWP, a binary protocol that unifies ingestion and Arrow reads in a single columnar client. Enterprise 4.0 adds cold storage tiering, QWP support, and hot failover without restart. Two major versions in one week is unusual cadence for a database; it suggests either coordinated release management or a long-running feature backlog clearing simultaneously.

Themes that compounded

• Agent identity formalized at the infrastructure layer: HashiCorp Vault IAM GA and Speakeasy per-agent pages both treat the agent as a named principal with an audit trail, not a service account.
• Agentic loops extended their time horizon: Workato 30-minute Genies and Rollbar Resolve represent automation that completes a full workflow rather than a single step.
• CMS and content platforms are bifurcating their surfaces: Sanity Context v2.0.0 is the clearest example, splitting the product explicitly for humans vs. AI.
• No-code builders are gaining direct AI model integration: WeWeb added OpenAI, Anthropic, and Google Gemini calls directly to backend workflows.
• Cloud platform completeness progressed: Aiven added self-service Azure BYOC, closing out the three-cloud story.

Watch this week

Rollbar Resolve is in public beta — watch adoption friction: whether the agent's PR quality is high enough that reviewers merge rather than close, which determines whether this becomes a workflow change or a curiosity. Speakeasy session quarantine is new enough that the policy configuration surface will surface edge cases quickly. And Rivet Dynamic Apps is entering a space where the primary challenge is not technical isolation but product discovery — whether developers find the deployment primitive useful before the user-generated-app pattern becomes mainstream.