Analytics tools handed query and authoring surfaces to AI agents this week, and shipped the governance to contain it.

The week in analytics

The dominant move this week was analytics tools handing their query and authoring surfaces to AI agents, then building the governance to survive it. Lightdash shipped a local workflow where a coding agent scaffolds a data app on your machine, iterates against live data, then uploads the source for Lightdash to build inside a governed instance. Usermaven opened a public MCP server so a workspace can be queried from Claude, ChatGPT or Cursor, and paired it with a data-health score. Neo4j shipped MCP for Aura as a hosted service inside the database itself. The common thread is that authoring is leaving the product UI while the data and metrics layer stays controlled.

The second move was commercial machinery. Plotly completed a metered hosting stack — compute modes billed per five-minute window, then customer-owned domains with managed TLS — turning its cloud into consumption-priced infrastructure. ApexCharts went the other direction with the same intent, converting an open charting library into open core across three releases: an offline licence check, then the first premium-only chart type, then entitlement enforcement. Both build a paid surface without removing what already worked.

Leaders

Lightdash had the week's highest activity (velocity 8.8, two sparks). Its local data-app workflow lets a developer build in Cursor, Claude Code or Codex against live data and ship into a governed instance, while a parallel content-as-code expansion put charts, dashboards, permissions, roles and AI agents into pull-request-reviewable form. Together they point at agent-authored changes arriving as PRs against a Lightdash instance.

Usermaven shipped two sparks that change what it ingests and how it is queried. Event Sources pulls conversion events from payment providers, CRMs and spreadsheets in as native events with identity resolution and revenue attribution, and the new MCP server makes the interface optional — deliberately paired with a Measurement Trust Center that scores workspace data health, since a chat interface answers confidently from bad data.

Plotly closed its production-hosting story with custom domains and automatic certificate renewal, a month after per-app compute modes introduced credit-metered billing. Its two tracks — Studio as the AI authoring front end, Cloud as the metered host — now converge on a single author-then-deploy-then-pay-by-compute funnel.

ApexCharts stood up a paid tier in three releases and began enforcing it, shipping a premium-only unit chart that renders watermarked in trial mode until a key is set, then tightening the check to require an entitled plan. It sequenced the pivot so nothing installed was taken away; the pressure lands on new capability.

Neo4j shipped MCP for Aura as a hosted service authenticated with existing logins, exposing schema, read and read-write tools, alongside a Cypher Copilot that lints and EXPLAIN-tests its own generated queries before showing them. The pairing courts the same non-Cypher-writing user from two directions.

Wildcards

Axiom removed the last human-gated step in its funnel: an AI agent can now create its own Axiom organization and have a person claim it afterward. It extends an agent-operator track that had lived in querying and monitoring all year up to account provisioning itself.

Apache Storm cut its 3.0 line, removing every remaining trace of the Clojure it was originally written in and moving to a Java 21 baseline, with the first substantial throughput work in years — compressed tuples and cluster state, and a control plane split from the data plane on receive queues. It is the rare release here about modernizing a legacy codebase rather than adding an AI surface.

Themes that compounded

• MCP and agent-native access showed up across the board — Lightdash, Usermaven, Neo4j, Axiom and MotherDuck all shipped agent-facing surfaces.
• Governance shipped alongside AI, not after it: data-health scoring, audit logs of every AI query in Basedash, and content-as-code permissions in Lightdash.
• Consumption and open-core pricing surfaced as first-class product work in Plotly and ApexCharts.
• Analytics products increasingly want to be substrate, not destination — embeddable APIs and scheduled delivery in Basedash, provisioning hooks in Mixpanel.
• Declarative and generated authoring spread past dashboards into orchestration and modeling — Dagster automation reaching jobs, Omni generating semantic models.

Watch this week

The open question each leader left is review. Lightdash can have agents author apps but has not settled how agent-authored apps get approved before viewers see them; Neo4j has Copilot writing Cypher and MCP exposing the graph but no stated control over what an agent may run against production; and Axiom now lets agents provision orgs, leaving billing and access control as the next human-gated targets. Watch whether next week's releases add the approval and permission layers these agent surfaces now require, and whether Usermaven's data-health score begins gating results rather than only reporting them.