dltHub is launching the AI Workbench, an infrastructure layer for dltHub Pro that enables AI agents to not just write dlt data pipelines but also run, validate, and deploy them to production. The platform ships four toolkits: a REST API pipeline toolkit (with 9,700+ pre-built configurations), a data exploration toolkit (with validation, SQL interface, and Marimo dashboards), a data transformation toolkit (with semantic ontology mapping), and a deployment toolkit that handles workspace conversion, pre-deployment validation, and production deployment with human checkpoints. The core design principle is that agents propose, humans verify, and deterministic tooling enforces boundaries — with metadata, schemas, and runtime traces flowing continuously between steps via an MCP server. The platform separates credentials and raw data from metadata so agents can reason about data shapes without security risks.
Table of contents
I. Writing code was never the hard part Link iconII. We are building the missing infrastructure Link iconIII. dltHub Pro: where it all clicks Link iconIV. How it works in practice Link iconPropose, verify, enforce: How Agents, humans and tools do best together Link iconTrust in the agent's work is built into dltHub, not bolted on Link iconStart here Link iconSort: