The Agent OS and the End of CDC

Omnigen provides a common API over multiple coding agent harnesses, with built-in server, sandboxes, and collaboration. This tackles the fragmentation teams face when building internal AI tools—each group reinvents sharing, session management, and security. Databricks already saw five internal agent frameworks emerge before deciding to standardize on one open platform.

The security model introduces contextual, stateful policies that go beyond simple allow/deny lists. For example, an agent can read confidential documents and publish to a website, but not both in the same session. Budget caps per sub-agent session add spend control. This shifts the security usability trade-off in favor of both safety and flexibility.

LTAP (a play on HTAP) writes OLTP data from Postgres directly into the data lake as columnar Parquet, using idle storage-fleet CPUs for transcoding. This eliminates fragile CDC pipelines, making transactional data instantly available for analytics without impacting the primary database. The initial prototype worked with no performance penalty by leveraging existing infrastructure.

The open-source approach aims to create network effects around integrations and libraries, similar to Spark. However, the evidence is early: LTAP is a prototype, and Omnigen’s community adoption is just beginning. The real test will be whether enterprises adopt these conventions or continue building bespoke tooling.