Governance
The system limits how far a failure spreads, built in from the start.
When Hugging Face's infrastructure was breached in July - a malicious dataset, a zero-day, credentials harvested, lateral movement - the incident stayed contained. Not because a senior engineer caught it in time, but because detection, blast-radius limits, and access controls were already sitting in place before anyone had reason to expect that specific attack. That's what governance looks like when it's real: containment that doesn't depend on a person noticing fast enough.
Read: "Governance Like a Warm Hugging Face"
Understanding
Organizations deliberately build, maintain, and keep current all institutional knowledge.
The Amazon outage above is the case study. Nobody hallucinated the bad answer - the wiki gave it, faithfully. For decades, engineers knew a wiki was never the whole story and filled the rest in themselves. AI doesn't fill anything in. If the knowledge isn't captured somewhere an agent can reach, it builds on an incomplete picture with total confidence. Understanding has to become owned, resourced engineering work - not a value everyone nods at in a retro.
Read: "No More Free Rides"
Intent
A spec is only as good as an agent's ability to act on it directly.
Vague intent doesn't produce no implementation. It produces a confidently wrong one. Tests, specifications, documentation, and conventions have to make desired behavior explicit enough to guide both humans and agents - not just the humans who used to be trusted to fill in the gaps.
Coming soon in Essential Complexity.
Direction
Humans set direction and own the outcomes. The agent executes.
AI can accelerate implementation. It does not own architecture, tradeoffs, priorities, or accountability - someone still has to. The PR is increasingly the first draft of the spec, not the last checkpoint before one. Review has to change accordingly.
Coming soon in Essential Complexity.
Evidence
Proof of correctness becomes the mechanism by which humans and agents can trust the system.
Confidence should come from evidence, not assumption. As agents take on more implementation, automated tests, observability, and agent-generated proof of correctness are what make accelerated implementation safe to review and sustain.
Coming soon in Essential Complexity.