I’ve been searching for the most concise way to describe what has changed in the agentic world and why Dolt, the world’s first version-controlled SQL database, is now essential. I’ve spilled a lot of words on the topic. Here’s my best shot.

Before Agents…#

Before agents, there used to be two types of writes: human and machine. The three concerns in writes are throughput, concurrency control, and correctness.

Human Writes#

Human writes are:

1. Large
2. Long duration
3. Untrusted
4. Textual

Human writes are managed by version control to maximize throughput, manage concurrency, and ensure correctness. Because human writes are untrusted, intelligence applied at review time ensures correctness and granular diff and rollback are essential for error recovery.

Machine Writes#

Machine writes are:

1. Small
2. Short duration
3. Trusted
4. Structured

Machine writes are mainly managed by online transaction processing (OLTP) databases under multi-version concurrency control (MVCC) to maximize throughput, prevent conflicting edits at scale, and ensure correctness via structured constraints. Structure and constraints are also enforced in the software systems built on top of databases to make writes trusted.

Agentic Writes#

Agentic writes are new and fall on a spectrum between traditional human and machine writes. Agentic writes are:

1. Small to large
2. Fast and slow
3. Untrusted
4. Textual or structured

Thus, agentic writes require both write management models: OLTP/MVCC because the writes are machine scale and version control because the writes are untrusted.

To date, agentic writes have almost strictly replaced human writes under version control, like code, or single actor systems with no concurrency, like writing documents.

Dolt is the only database to offer both write management models. For agentic writes to scale beyond human write use cases, Dolt is required.

Conclusion#

Short enough for you? Dolt is the database for agents, in less than 300 words. Anything to add? Come by our Discord and let me know.