The fundamentals-first amplifier

The measurement and guardrail layer — CI, tests, and standards the team actually agrees on — must be made trustworthy before agents are pointed at the code. If agents cannot run tests to find out they have broken your software, do not be surprised when they break your software. If the humans do not agree what good looks like, there is a very low chance an agent produces code the team approves of.

One organization spent a year and a half systematizing flow before the AI push, then taught failure modes first and reached 94% adoption. Another made its static analysis and test suites trustworthy before any agent wrote code, and worked one area at a time. The organizations that jumped in top-down demoed features that did not match the product, because years of engineering-principle guardrails vanished in the rush.

A repository with no tests yields agents that write no tests. The practices encoded in the repo are the practices the agent adopts, which is why the guardrail layer is a precondition rather than a follow-up.

Also: the DORA 2025 report.

Assay’s whole posture is a fundamentals-first design: machine-checkable gates before autonomy in the lifecycle, briefs that carry their own Verify tables, and practice encoded in skills and written rules rather than inferred by agents from a weak-convention repository. The adoption flow installs the guardrail layer before any desk runs.

The install path scaffolds the fundamentals but does not yet assess an adopter’s existing CI and test trustworthiness before desks are turned loose on a brownfield repository. That check is the mechanism this evidence names and Assay does not have.

• Whether the summit circuit picks this up — the evidence is currently DevCon-only — and whether anyone quantifies the threshold the amplifier framing leaves open.
• Reference implementations of an adoption-readiness assessment that gates agent rollout on measured CI and test trustworthiness.