TECH Signal 512 2 feeds carried it
LLM agents prompt nine new engineering practices to extend Joel’s Test
↗
A proposed Shelley Test adds nine agent-centric questions to Joel Spolsky’s original twelve for modern software teams.
The original Joel Test remains a quick litmus test for team health. Adding agent-specific questions reflects the shift from human-led workflows to LLM-supervised ones. Teams that adopt these practices will trade manual review cycles for automated, adversarial checks and continuous deployment loops.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Agentic code review replaces human peer review with adversarial LLM subagents.
Continuous deployment is now supervised by LLM agents that monitor logs and metrics.
Observability tooling must be SQL-queryable and integrated with coding agents for triage.
THE READ
What the cluster adds up to.
The Shelley Test extends Joel Spolsky’s 2000 checklist with nine new questions tailored for LLM agents. These questions cover agentic code review, continuous deployment, integration tests, observability, model access, merge queues, tooling agility, workflow discussion, and agent legibility. The shift is from human-centric practices to workflows where agents perform or supervise each step.
Adopting these practices requires infrastructure changes. Teams need to integrate adversarial LLM subagents for code review, deploy agents like Athena for continuous deployment, and ensure observability tools are SQL-queryable. The cost is not just tooling but also cultural: engineers must trust agents to make deployment decisions and triage alerts without human intervention.
The practices stop working where agent capabilities plateau. If models cannot reliably detect subtle bugs or integration issues, adversarial reviews and continuous deployment loops will fail. Similarly, if observability data is incomplete or noisy, agents like Sisyphus will misclassify alerts. Teams must also ensure their merge queues complete in under three minutes to avoid bottlenecks.
The Shelley Test reflects a broader trend: LLMs are not just augmenting engineers but replacing entire workflows. Human oversight is still required, but the role shifts from active participation to supervision and adjustment. Teams that adopt these practices early will gain faster feedback loops, but they must also accept the risk of agent errors and the need for rapid iteration on tooling.
Written by elseif from the cluster below · checked for specifics the sources never containedTHE CLUSTER