Time-as-Knob Pattern¶
Type: Mechanism / pattern
Referenced from: closed-loop-agent-control
Definition¶
The Time-as-Knob pattern means giving an agent a first-class tool to control the execution pace of a target system (typically a simulation or game engine). The agent decides when and how fast time advances, rather than observing a running system and reacting to its output.
Nikita M.'s assessment: "Because the game time was a first-class knob for the agent, it turned into an unbelievably strong way of closing the feedback loop."
Capabilities It Enables¶
| Time Control | Debugging Use Case |
|---|---|
| Advance frame-by-frame | Precise debugging of state transitions |
| Skip ahead | Test specific scenarios without waiting |
| Pause | Reason about current state before acting |
| Replay | Verify that fixes actually work |
Why It's Novel¶
Most agent-tooling focuses on perception (reading state) and action (writing state). Time-as-knob adds a third dimension: temporal control. This lets the agent:
- Decouple reasoning speed from simulation speed — think for 10 seconds about a state that existed for 50ms of game time
- Compress无聊 scenarios — skip routine gameplay to reach interesting edge cases
- Extend critical moments — slow down or freeze time during debugging of specific interactions
Implications¶
This pattern suggests that agent architectures should expose execution flow control as a first-class tool, not just state read/write. For non-game systems, the equivalent might be:
- Pause/resume API execution
- Step-through debugging at the API level
- Control over retry/backoff timing in async workflows
Related Concepts¶
- closed-feedback-loop — the broader principle; time control is one of three mechanisms
- tool-mediated-game-state-control — the implementation that includes adjustClock
- model-size-agnostic-iteration — the outcome enabled by temporal control