Lorena Barba argues reproducible research practice is just context engineering for coding agents

arXiv 2609.11728·low signalXBlueskyLinkedInCopy link

The position paper reframes tests, commit histories, repository structure, instructions and decision records as the context a coding agent reads, arguing agents lower the cost of maintaining those artifacts while making their payoff immediate rather than deferred. The framing inverts the usual pitch for reproducibility, which asks researchers to pay a cost now for a benefit to someone else later. Barba keeps the responsibility with the researcher for verifying the artifacts and the scientific judgments they encode, so this is an argument for maintaining the repo as agent context, not for delegating the judgment.