The context pipeline: federated data, live mirrors, graph layers, and expertise over MCP
Starburst pushes federation over migration, Snowflake mirrors Postgres live, Neo4j launches a context engine, and Daloopa pipes expert interviews into agents over MCP. The agent stack's new bottleneck is context plumbing.
- Starburst advocates data federation over batch ETL and centralized warehouses for agent workloads, with governance enforced deterministically at the query engine, per the company. Self-reported m:…3601
- Snowflake made Postgres data mirroring generally available: object-storage-based change data capture with change batches pushed every 15 seconds, Iceberg change logs auto-refreshed every 30 seconds, and exactly-once processing, per the company. Self-reported m:…3443
- Neo4j launched its Context Engine, part of its graph intelligence platform, to create a shared understanding of business data across systems for enterprise AI agents. Self-reported m:…2733
- Daloopa integrated Heron AI Events — AI-conducted expert interviews with former employees, suppliers, customers, and industry insiders, typically following market-moving events — into its Model Context Protocol server. Self-reported m:…2605
- Analysts can access the expert interviews in Claude, ChatGPT, or other AI platforms connected to Daloopa's MCP and pair them with audit-ready fundamental data to pressure-test a thesis; access is opt-in at the company level. Self-reported m:…2605
Query it where it lives
Starburst argues enterprise data was never built for AI. "AI workloads require real time, cross-domain context that traditional batch ETL pipelines and centralized data warehouses fail to deliver efficiently" — so the company's prescription is data federation: querying distributed datasets in place, with a hybrid approach that hits high-throughput analytical tables in Apache Iceberg while federating to operational databases when live source context is required. m:…3601
The failure mode Starburst names is sharp: "Autonomous AI agents magnify schema ambiguity, generating confidently wrong answers when disconnected from explicit business definitions and domain boundaries". The proposed fix is certified data products — tables packaged with metric definitions, schema grains, domain ownership, and documentation — and governance enforced deterministically at the query engine, with server-side access rules, row filters, and column masking applied before execution. m:…3601
Built on Trino, Starburst says it runs high-concurrency federated queries across object stores, relational databases, and warehouses in a single SQL statement — eliminating the cost, delay, and security exposure of centralizing data first, in the company's telling. m:…3601
Mirroring Postgres
Snowflake has made Postgres data mirroring generally available. Instead of intermediate replication infrastructure, data mirroring uses only object storage as the intermediary, with change batches pushed from Postgres every 15 seconds and Iceberg change logs auto-refreshed in Snowflake every 30 seconds, the company says. m:…3443
The engineering claims: mirrored tables follow Postgres transaction boundaries for internal and cross-table consistency, changes are captured and processed exactly once, mirrors can snapshot over a million rows per second on larger instances, and $live views combine mirrored tables with pending changes for low-lag access. Mirroring from Snowflake back to Postgres is coming soon, the company says. m:…3443
For agent builders, the pitch is a live operational mirror without a replication stack to babysit — the kind of plumbing that turns a warehouse into a context source. m:…3443
A shared layer for business meaning
Neo4j launched its Context Engine, part of its graph intelligence platform, to create what the company calls a shared understanding of business data across systems — aimed at trustworthy enterprise AI and better agent performance. m:…2733
The bet is the graph as a knowledge layer: business entities and their relationships, resolved once and shared across systems, rather than re-derived by every agent from raw tables. Whether one shared layer can hold a whole enterprise's meaning is the open question. m:…2733
Expertise delivered over MCP
"Daloopa's MCP now includes Heron AI Events from Heron Intelligence" — AI-conducted expert interviews with former employees, suppliers, customers, and industry insiders that follow market-moving events, most often earnings releases. Analysts can access them in Claude, ChatGPT, or another AI platform connected to Daloopa's MCP. m:…2605
The use case is pairing qualitative commentary with Daloopa's audit-ready fundamental data: analysts can compare model assumptions against what experts are seeing in the market, pressure-test a thesis, or review issues experts flagged the previous quarter to prepare targeted questions for management. Access is opt-in at the company level, with transcripts then available to all users at the firm. m:…2605
It is a different kind of context pipeline — not rows and schemas, but judgment, delivered where the agent already works. m:…2605
What is still unresolved
- Performance and cost figures from Starburst and Snowflake are vendor claims; no independent replication appears in the evidence.
- No independent evidence shows whether federated queries or graph context layers measurably improve agent outcomes.
- Daloopa's expert-interview coverage depth is not quantified in the evidence.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.