Geordie AI’s Valerie Dryden argues that AI coding agents make established software-engineering practices more important, not obsolete. Her practical prescription is to build feedback and shared understanding into the work, rather than asking people or agents to rescue a large change at the pull-request stage.
Geordie Watch analysis
What happened
In “What’s Old Is New Again”, Dryden challenges the idea that AI-generated code calls for scrapping Git or reviewing code at machine speed. She argues that producing and approving code is only part of engineering: teams also need to understand what they are building, how it fits together and what could break.
Her proposed toolkit is familiar rather than futuristic: shorten feedback loops, keep changes small, make quality part of the work, share decisions and integrate continuously. She points to test-driven development, acceptance-test-driven development, pairing and ensemble programming as ways to test assumptions and discuss design before implementation. Geordie says its team pairs on workstreams, sometimes works individually with agents, then comes back together to discuss the work.
Why it matters
Agents can produce code quickly, but speed does not automatically give a team the context to judge whether that code is appropriate. Dryden’s point is especially pertinent when software handles customer data, moves money or supports systems people rely on: a late review is a poor substitute for understanding the decisions being made along the way.
The argument also gives teams a more useful choice than “ban agents” or “automate the whole review queue”. Practices such as smaller changes, automated tests and early collaboration can make AI assistance part of an existing engineering process, while keeping people involved in decisions they remain accountable for.
Our read
The strongest idea here is that AI changes the pace of software work without abolishing the need for engineering judgement. Geordie’s essay is a practitioner’s argument, not evidence that one workflow fits every team. Still, “more code, faster” is a thin definition of progress; teams should ask whether their process helps them catch mistakes and understand the system before a pull request becomes the first proper conversation.
What to watch
- Whether teams using agents adopt smaller changes and faster feedback as routine practice.
- How test-driven and acceptance-test-driven approaches work in agent-assisted projects.
- Whether early pairing and design discussions reduce late review bottlenecks without shifting accountability away from people.
Discussion spark: When AI agents can produce code faster than people can review it, should teams invest first in stronger tests and smaller changes, or in keeping humans involved throughout design and implementation?
Sources and evidence
- Source update (17 November 2026, 00:00 UTC)
Independent WittyWires tracker for public updates about Geordie. Not affiliated with or endorsed by Geordie; this is not an official account.