AI agents can read the code. They still cannot read the reasons.
Addy Osmani's "Intent Debt" is a useful name for one of the most expensive gaps in agentic engineering: the goals, constraints, and trade-offs that never got written down.
In every mature system I have worked on, the code is never the whole truth. Some behavior exists because of a regulator. Some exists because of a customer migration. Some exists because a past failure taught the team not to be clever in one specific place again.
Humans carry that context informally. Agents do not. They see structure, names, tests, tickets, and fragments of documentation. What they miss is why a decision was made and which simplifications are forbidden. That makes agentic engineering more sensitive to missing rationale than traditional engineering ever was.
For me, this changes the documentation priority. Architecture records are no longer only for onboarding humans. They are control inputs for the tools that will modify the system next.
The next scarce artifact may not be code. It may be a preserved intent.