.
When my automation intent was submitted, the Keystore Rollup showed the zkPermission had been written successfully. At first glance, that looked like the agent was ready to go.
Not exactly.
The Keystore only records what an agent is authorized to do. The actual execution happens later, when validators pick up the intent, verify it against that stored permission, and finalize the state transition. In other words, permission storage and agent execution are separate events—not a single process.
Another detail that stood out is Newton's validator design. Validators stake NEWT, face a 14-day unbonding period, and can be slashed for invalid behavior. That means every execution is backed by economic incentives rather than blind trust.
The question I'm still thinking about is liveness. If some intents are quicker or less resource-intensive to verify than others, could validator queues naturally process those first? Not because of bias, but because of efficiency.
I'm also curious how the network behaves during heavy demand. If mainnet beta suddenly receives a large influx of automation intents across Ethereum and Base, does execution latency increase evenly, or do certain categories of intents experience longer delays?
One takeaway feels important: seeing a permission successfully written is proof that authorization exists—but it's not proof that your agent has already executed or will execute immediately. Those are distinct milestones, and understanding that distinction seems essential for anyone using Newton for the first time.