On this page
Anyone who works with an AI assistant every day knows the feeling: the AI is doing the legwork, yet you still end up exhausted. What wears you out is rarely the hands-on execution, which the AI already handles, so much as the steady stream of multiple-choice questions popping up in the chat—big and small decisions lining up for your sign-off until your attention is completely drained.
Lately I’ve been using a simple trick to fix this: stop rethinking every problem from scratch, and have the AI actively record the decisions you’ve already made. Next time it needs a call from you, have it look up past precedents first and attach a recommended course of action, so all you have to do is nod or tweak the answer.
See the 3-month 1-on-1 program →
Borrowing common-law precedent and Ming dynasty draft slips
This workflow borrows from two old institutions.
The first is the doctrine of precedent in common law: when a new case comes up, you check how similar cases were ruled in the past instead of inventing a new standard every time. The second comes from how Ming dynasty emperors handled memorials to the throne. When a memorial reached the emperor’s desk, it didn’t arrive as a bare document. Pasted to it was a small slip written by the Grand Secretariat with a recommended ruling, known as piaoni (a draft rescript). The emperor read the slip, approved it if it made sense, or wrote a different instruction if he disagreed.
Bring those two ideas into your daily AI workflow, and the division of labor between you and your assistant gets much lighter right away:
- Log precedents at the end of a chat: Have the AI actively save the decisions you made during the session into topic files in its memory space as you wrap up.
- Attach a draft slip before asking you to decide: Whenever the AI needs a decision from you, have it check those memory files for similar precedents first and give you a recommendation based on them; you can either approve it or go a different way.
Put these two steps together and you’ll “cut down attention drain dramatically.” When your AI looks up precedents and attaches a draft slip for you, you can “attend to ten thousand affairs of state without ever flagging.”
Copy-paste instructions for your AI
You don’t need to write a single line of code. Just copy the two prompts below to your AI assistant, or drop them into its operating rules, and you can start right away.
Ask the AI to file your decisions at the end of each conversation:
At the end of every conversation, please proactively record the decisions I made during this chat into the topic files in your memory space.
Ask the AI to check precedents and attach a recommendation before asking for your call:
Whenever you need me to make a decision, please check your memory files first for any similar precedents and give me a recommendation based on them; I can then either agree or choose a different option.
One key reminder when setting this up
One detail matters a lot when you set up this workflow: do not stuff these decision logs into the always-on config file that loads automatically with every single chat.
Why keep them out of the always-on config? Because your preferences depend heavily on context, and any given type of problem only comes up once in a while. There is no reason to burn your context budget on every single turn just to carry preferences you only need occasionally.
Keeping these precedents in “an on-demand space,” where they take up zero always-on quota and are only read when filing decisions at the end of a chat or looking up a similar question, “is the smartest move.”
Appendix
If you use an AI agent tool that supports project config files and a memory directory, the always-on config file mentioned above corresponds to Agents.md, while the on-demand memory topic files correspond to the individual topic files under the memory directory. Routing decisions made during chats into memory topic files preserves the context of your past calls without bloating your always-on config file and eating into your quota.