TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Cockroach Labs describes five months of using a hospital-style workflow to coordinate coding agents on its database migration tools. The company says the system added IBM Db2 support to MOLT in under two days, with agents writing the code and humans involved in two escalations; its speed and cost comparisons are company-reported.
Cockroach Labs says a coding-agent workflow it tested for five months added IBM Db2 support to its MOLT database-migration tools in less than two days, compared with nine months for the company’s Oracle support work in 2024. The company says agents wrote all the code for the Db2 effort, while a review system modeled on a teaching hospital sent work through staged checks and escalated two issues to a human.
The Db2 project required changes across MOLT, Cockroach Labs’ migration tool set: a schema converter, a new process for fetching rows and loading them into CockroachDB, a verification path for comparing data, an ANTLR grammar, a CI Docker image and more than 10,000 lines of test fixtures, according to the report. The company says the work cost $4,172 in model-token charges. It describes that as 164 times faster and 38 times cheaper than the Oracle-support effort, which it says took nine months and about $160,000 in engineering time. Those are the company’s comparisons, not an independently audited assessment.
The workflow began with a GitHub issue describing the requested feature. A planning agent split the work into 15 sub-issues and laid out dependencies; some tasks were then broken down further. The company reports that the project generated 32 sub-issues and 27 merged pull requests. Reviewers returned work for changes 55 times, and nine issues were escalated, including two to a human. Cockroach Labs says test coverage was equivalent to its PostgreSQL coverage, which it described as its best-tested dialect.
The system, named MOLT Sinai, represents issues as patients and merged work as discharge. The company says the process also includes a Charge Nurse agent that checks for stalled work every 30 minutes, an Infection Control agent that can halt workflows if the main branch breaks, and safety and research roles. The report says the system runs on GitHub Actions, with issue labels and scratch files tracking work; the supplied account does not detail every role’s permissions or the full review criteria.
Quality Checks Before Code Ships
The report addresses a central concern for companies using coding agents: generating code quickly does not by itself show that the result is safe to ship. That distinction matters for database migration software, where a defect could affect how customer data is transferred or verified. Cockroach Labs says its design puts staged review, testing and escalation between agent-written changes and the main codebase, rather than optimizing primarily for the number of changes produced.
The example offers a concrete account of one company’s approach, but it does not establish that this workflow will produce similar results across other teams, repositories or kinds of software. The figures also do not measure the wider cost of designing and maintaining the system, human oversight, or any post-release defects. Readers should treat the reported time and token-cost comparison as a case study, not a general benchmark for agent-driven development.
As an affiliate, we earn on qualifying purchases.
From Oracle Work to Db2
Cockroach Labs builds CockroachDB, a distributed SQL database, and MOLT tools for moving workloads to it. The company says its earlier Oracle-support work in 2024 took about nine months and roughly $160,000 in engineering time. Its later Db2 project covered a different database source, with a rich SQL dialect, complex types and a native wire protocol, and required new components across the migration pipeline.
The report frames MOLT Sinai as a response to the risks of relying on agents for production work. Its authors contrast the hospital’s handoffs and second opinions with agent systems designed mainly to increase coding throughput. They say the hospital model accepts bureaucracy, longer waits and additional costs in exchange for stronger controls. The report is a Cockroach Labs account of its own experiment; the source material does not provide an independent evaluation or results from a controlled comparison.
“The question we cared about was not “how many PRs per hour?” but rather, “how many of those PRs would we have been embarrassed to have merged ourselves?””
— Cockroach Labs report
database migration tools for IBM Db2
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Limits of the Reported Results
The report does not identify the year of the April Wednesday-to-Friday Db2 project in the supplied material, nor does it provide an independent audit of the time, cost or test-coverage comparisons. It also does not state how much engineering time went into building MOLT Sinai, how much human review the 55 returned changes required, or whether the $4,172 figure includes costs beyond model tokens.
It remains unclear how the workflow performed after the reported project: the source material gives no production incident rate, customer-impact measure, or comparison with human-led Db2 development under the same conditions. The company’s account establishes what it says happened in this case, but not whether the method generalizes or whether its safeguards catch every consequential defect.
As an affiliate, we earn on qualifying purchases.
Further Evidence From Real Work
The report describes a five-month trial and one detailed Db2-support example, but the supplied material does not set out a date for a broader evaluation or a next project milestone. The practical next evidence would be additional work through the same process, alongside clear reporting on review effort, defects found before and after merge, and the full cost of operating the system.
Until those details are available, the result is best read as a company-reported case study of using agents under structured review. Cockroach Labs’ account makes the workflow and its reported project metrics public; whether the same balance of quality, time and cost holds elsewhere remains open.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Cockroach Labs use coding agents to build?
The company says the agents added IBM Db2 support to MOLT, its database-migration tool set for moving workloads to CockroachDB. The work covered schema conversion, data fetching and verification, CI and test fixtures.
How long did the Db2 work take?
Cockroach Labs says the work ran from a Wednesday evening to a Friday afternoon in April and took less than two days. The supplied report does not specify the year.
Did humans review the agent-written code?
Yes. The company reports that reviewers returned work 55 times and that nine issues were escalated, with two escalated to a human. It says no human wrote the code, which is distinct from human review and escalation.
Are the speed and cost figures independently verified?
No independent verification is included in the supplied source material. Cockroach Labs reports a $4,172 token bill and compares the effort with its nine-month, roughly $160,000 Oracle-support project from 2024.
What is MOLT Sinai?
MOLT Sinai is Cockroach Labs’ name for a hospital-inspired system for assigning and reviewing coding-agent work. The company says it tracks tasks through GitHub issues and uses roles for planning, stalled-work checks, safety reviews and escalation.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
