Analytics tools are putting an agent at the front door and making its answers inspectable.
The week in analytics
The sector's lead story is interface. Two data tools rebuilt their home screens around an agent, and a third made that agent's reasoning visible. The question for analytics is shifting from what the dashboard shows to whether the answer can be trusted.
A second thread is agents leaving the read-only lane: one reporting tool now launches and pauses ads, while performance data is opened to coding agents through MCP.
Leaders
MotherDuck redesigned its UI so Home opens on an Agent Chat prompt that can query, load and transform data. It also cut AI function prices by 50 to 60 percent on OpenAI models, which makes the agent cheaper to leave running.
Count shipped Reasoning Chains, tracing each claim in the agent's answer back to sources, assumptions and inferences and labelling which parts are observed. It is the sector's most direct answer to the trust problem.
Whatagraph moved from reporting to action: IQ Agents compare spend with budget, look for stalled campaigns, and use MCP to launch or pause Meta ads and create tasks. It also added OpenAI Ads as a data source.
Calibre opened its RUM, CrUX and synthetic data to coding agents through a hosted read-only MCP server on every plan, and added deployment markers to RUM charts. Performance data now sits one prompt from the code that causes it.
Grafana Loki published a 4.0 release candidate that deletes the BoltDB, Cassandra, DynamoDB, BigTable and gRPC storage backends and the Simple Scalable mode. Anyone on those backends has a migration to plan.
Wildcards
ROOT released 6.42 rc1 with its Python interface rebuilt on cppjit and Cling moved to LLVM 22, a large internal change for a scientific toolkit.
Rho added a generic ACP agent backend in 0.5.4, so agents beyond Claude can plug into an R IDE, after earlier project-scoped agent conversations.
Themes that compounded
- Agent-first home screens at MotherDuck and agent backends at Rho.
- Verifiable answers at Count rather than just fluent ones.
- MCP as the access path in Calibre and Whatagraph.
- Removals and rebuilds, not additions, at Grafana Loki and ROOT.
Watch this week
Watch whether Whatagraph's agents that act on ad accounts draw approval or audit controls in a follow-up, since Count's reasoning traces set a visible bar. Loki's 4.0 candidate may bring migration guidance for the removed backends, and Rho's ACP backend may bring more agent options into the IDE.