AI coding tools have compressed application development from days to hours, yet analyst reports find that AI accelerates coding by 30 to 40 percent while overall team productivity rises less than 10 percent when downstream steps remain manual. More often than not, the database is what slows things down. VMblog met with Sandeep Parikh, Chief Product Officer at GemDB, to hear more about GemDB’s transactional object database where the code is the schema, and their newest product GemDB Code.
VMblog: What problem are AI coding tools exposing?
Sandeep Parikh: Roughly 57 percent of software changes require a database update, and in regulated enterprise environments, getting a schema change approved can take up to two weeks. That mismatch existed before AI coding tools, but it mattered less when code moved at a human pace. Code is now moving faster than ever, and the database migration pipeline has not kept up.
VMblog: How does GemDB’s architecture eliminate that constraint?
Parikh: GemDB is object-native, so when you create a Python object and commit it, the database persists it as a Python object with full ACID transactional guarantees. There is no serialization to rows, no translation to JSON documents, no ORM mapping. Because the code is the schema, when an AI coding agent modifies a data model, that modification is already the database change. No separate DDL script, no migration to orchestrate, no production database to take offline. We call this agentic schema evolution. Adding new many-to-many relationships, which would typically require join tables, foreign key definitions, and backfill scripts in a relational database, requires only adding a field to an object.
VMblog: How does ACID compliance work in GemDB?
Parikh: Every operation in GemDB is transactional by default, with no transaction mode to enable and no performance penalty to evaluate. Its ACID semantics are built into the architecture rather than layered on top. The engine has been running mission-critical workloads at JPMorgan Chase, UBS, and other financial institutions where transactional consistency is a regulatory requirement.
VMblog: What is GemDB’s production track record?
Parikh: The underlying engine was spun out of VMware 13 years ago and has been running in continuous production since 1986. Emergence, a permanent capital partner, acquired GemDB in April 2026 to fund new language runtimes and AI capabilities. Lifeware, an insurance technology company, runs its entire operation on GemDB. As founder Massimo Arnoldi has said, they never had to flatten a life insurance contract into rows and columns, so their developers spent their time building insurance logic rather than fighting with the database.
VMblog: What is GemDB Code, and how can developers try it?
Parikh: GemDB Code is a free VS Code extension that packages the database engine, a Jupyter Notebook kernel, and an MCP server for natural-language querying. The database downloads and configures automatically on first activation. It ships with Ice Cream Brain Freeze Insurance, a guided demo that walks developers through object persistence and schema evolution using an AI coding agent. What makes the exercise memorable is what never fires. No migration script, no downtime, no second step. GemDB Code is available now on the VS Code Marketplace.
VMblog: What can VMblog readers expect in 2027 from GemDB?
Parikh: We will continue to invest in making GemDB easy for developers building with AI agents and modern tooling. We will be making improvements to our observability stack to integrate with OpenTelemetry & Prometheus, enhancements for deploying & operating GemDB at scale, and building more support for continuous code & schema updates to production systems. Expect more tutorials, best practices, and agent skills to support increased production velocity. More to come!
Sandeep Parikh is Chief Product Officer at GemDB. GemDB Code is available as a free download on the VS Code Marketplace. For more information, visit gemdb.com.
