# Agentic governance — a thread from the writing of Anivar A Aravind

Who answers when an automated system acts, and whether the people it decides about can get the decision corrected.

Canonical: https://anivar.net/topics/agentic-governance/
Each essay negotiates to markdown via Accept: text/markdown at its URL.
The thread reads in order:

1. The Emergence of the Context Engineer (2025-08-15) — And the unsung role of QA in AI reliability — the people who make memory-native systems dependable. — https://anivar.net/writing/the-emergence-of-the-context-engineer/
2. Where Delegation Stops (2026-05-03) — The Layer 8 — Issue Two: Architected and mandated signatures, and what the protocol layer cannot decide from below. — https://anivar.net/writing/where-delegation-stops/
3. Building the Signature Surface (2026-05-12) — Signed Truth, Part Three of Three — https://anivar.net/writing/building-the-signature-surface/
4. The Sovereign Handoff (2026-05-22) — The standards ecosystem is building the agentic internet — and externalizing its governance. — https://anivar.net/writing/the-sovereign-handoff/
5. Dashboards Were the Last Central Planner. Context Windows Are the Next. (2026-05-23) — An essay on harnesses, knowledge, and the institutional layer that makes distributed action accountable. — https://anivar.net/writing/dashboards-were-the-last-central/
6. Exit Is the Primary Agentic Right (2026-05-28) — Agentic systems do not keep authority because it was granted. They keep it because the context that surrounded the grant never went away. — https://anivar.net/writing/exit-is-the-primary-agentic-right/
7. Agentic Observability: Seeing Is Not Stopping (2026-05-30) — Observability tells you what an agent did. Governance ensures someone else can correct it. The industry is wiring observation to action and calling the result governance. — https://anivar.net/writing/agentic-observability-seeing-is-not/
8. Beyond Commerce (2026-05-31) — The industry solved who the agent is. The hard part is who answers when a valid action goes wrong. That gap is the institutional layer, and it does not stop at commerce. — https://anivar.net/writing/beyond-commerce/
9. When GCCs Stop Counting Heads (2026-06-27) — Why Agentic AI Breaks the Link Between Headcount and Capability — https://anivar.net/writing/when-gccs-stop-counting-heads/
10. The Compiler Forgives. The Loop Forgets. (2026-07-21) — A coding agent works because the compiler, not the model, writes the truth down. That witness has no equivalent once loops compose: the handoff, not the loop, is the wall the field walks into. — https://anivar.net/writing/the-compiler-forgives-the-loop-forgets/
11. Releasing Decern: The Proof and the Record (2026-08-02) — A deterministic authorisation kernel for agentic systems. Proven, recorded, and in the open. — https://anivar.net/writing/releasing-decern/

## Other threads
- Digital identity — https://anivar.net/topics/digital-identity/
- Sovereignty — https://anivar.net/topics/sovereignty/
- Provenance — https://anivar.net/topics/provenance/
- Memory — https://anivar.net/topics/memory/
- Commerce — https://anivar.net/topics/commerce/
- Open weights — https://anivar.net/topics/open-weights/
- Institutions — https://anivar.net/topics/institutions/