In short
AI coding agents write and review code quickly, and they are not accountable for when a release ships. In a firmware team tied to a silicon date, each agent needs an identity, a defined scope, human approval on merges, and a place in the plan — so that its work is held to the same date as the engineers'.
What changed?
A typical firmware organisation now runs dozens of AI coding agents. More code is produced, faster. Faster individual work does not by itself make the release earlier: dependencies decide the speed of a program.
What are the risks?
Three, mostly about accountability:
- No identity: agents act under a shared or borrowed account, so nobody can say which one did what.
- No scope: an agent can reach repositories it has no reason to touch.
- No date: agents optimise the task in front of them, not the release.
What should be in place?
Each agent enrolled in the identity provider as a named identity; credentials scoped to its repositories; code owners approving every merge; a signed record of each action; and a way to suspend any agent with one command.
How Tymeline handles it
Tymeline gives every agent an identity and puts its commitments in the same record as the humans'. When the silicon date moves, the agents' work is re-sequenced with everyone else's.