LoginSign Up
← Back to Analytics
Weekly · Analytics · Week of July 6, 2026

Analytics vendors stop bolting AI on and start rebuilding around agents

agent-nativemcpconsolidationopen-table-formatsdata-productization
Y
By Yahya Tür
LinkedIn ↗
Generated 1h agoDrawn from 8 products

The week in analytics

The dominant pattern this week is not a new feature category but a change in who the customer is. Across the sector, the most directional releases treat an AI agent — not a human analyst — as the primary operator of the platform. MotherDuck shipped agent-native pipelines and MCP-creatable data apps; Apify now lets agents run and pay for its scrapers in USDC with no account; Deepnote wired OpenAI's Codex natively into a workspace's notebooks and schedules; Dovetail and Displayr are both making their chat surfaces auditable rather than just conversational. The through-line is that the query engine, the notebook, and the research repository are all being repositioned as things an agent acts through, with MCP as the connective tissue. This is the same move happening in parallel at a dozen vendors, which is what makes it a sector shift rather than a single roadmap.

The second, quieter pattern is consolidation and trust work underneath the AI layer. Usermaven pulled four separate analysis modules into one Analytics Hub; Chord is rebuilding its Copilot from the ground up rather than patching it; Displayr added a context pill and post-edit summary so users can see exactly what its assistant sent and changed. The teams furthest along have stopped shipping AI as a bolt-on and started paying down the debt that a bolt-on creates — governance, permissions, transparency, and a single coherent surface. Note that raw velocity is a poor guide this week: the two Apache Superset rows and Fulcrum all posted steady cadence, but every entry is either a Helm-chart packaging tag or a maintenance fix, with no product signal to read.

Leaders

MotherDuck posted two sparks and is the clearest mover. Server-side Iceberg attach (preview) lets it query external Apache Iceberg REST catalogs alongside its own and DuckLake tables, and direct dbt Cloud connectivity via the Postgres endpoint deepens the plug-into-the-existing-stack arc. Combined with the earlier Flights agent-native pipelines and Dives reaching GA, MotherDuck is visibly climbing from a serverless DuckDB warehouse toward a full, agent-operable platform.

Apify also posted two sparks, both aimed at agents as customers rather than developers. Agents can now run and pay for eligible Actors in USDC on Base via the x402 protocol with no account or API key, and new MCP connectors let Actors operate behind logins like Notion, Slack, and GitHub without seeing credentials. The counterweight — requiring approval for full-permission Actors — shows security catching up to an agent-first execution model.

Deepnote made its most directional move by connecting OpenAI's Codex natively to a workspace, so an agent's exploration starts from a team's real notebooks, schedules, and integrations rather than a blank prompt. New MCP tools to create, attach, and detach integrations extend the same bet: the notebook as the trusted context layer that outside agents operate through.

Dovetail shifted its center of gravity from storing research to answering questions over it. A new deep research mode splits the chat surface into a fast path and a slower, multi-step analytical one, building on full cross-turn context and in-chat code execution shipped the same month. Its MCP server is gaining write tools, making the repository operable by outside agents, not just readable.

Chord is making the boldest single bet: a ground-up rebuild of its Copilot assistant, shown first to a small customer group. Copilot Next reframes the assistant from an add-on into Chord's core interface, with persistent history, shareable chats, and answers that show their reasoning — a full rewrite rather than another increment on the old one.

Wildcards

Appfigures is off-pattern: instead of AI plumbing, it is turning its app-store estimate dataset into standalone market-intelligence products. New Leaderboards rank apps and games across both stores by 14 unified metrics, building on a 15-report App Intelligence suite for sizing up any competitor. This is a data-asset-to-product move, not an agent move.

Fairing is the other outlier: its Shopify Analytics integration syncs post-purchase survey responses into Shopify Order Metafields, so merchants analyze attribution and NPS against native dimensions like product purchased. The strategy is embedding survey data where merchants already act on it, rather than adding a chat layer on top.

Themes that compounded

  • MCP is becoming the default interface between analytics platforms and external agents, appearing in MotherDuck, Apify, Deepnote, and Dovetail releases.
  • Agent-native execution now brings its own guardrails — Apify's full-permission approvals and Displayr's auditable Chat edits are the trust counterweight to autonomy.
  • Open-table-format interop is spreading, with MotherDuck attaching Iceberg catalogs server-side alongside DuckLake.
  • Sprawling suites are consolidating into single surfaces, from Usermaven's Analytics Hub to Chord's ground-up Copilot rebuild.
  • Vendors are productizing their underlying datasets, as Appfigures turns estimates into Leaderboards and competitor intelligence.

Watch this week

Watch whether the previews and early-access bets in this week's feed start graduating. MotherDuck's Iceberg attach and Flights are both explicitly in preview, and Chord's Copilot Next is limited to a small customer group — the next signal is any of these moving toward GA. On the agent-payments front, Apify's x402 support is the first concrete instance of an analytics vendor letting agents pay per run with no account; a second vendor doing the same would confirm it as a pattern rather than an experiment. Treat the Superset rows and Fulcrum as no-signal noise until their feeds carry real application release notes.