<developer_note> is one of the most powerful, yet underutilized...
Matt Shumer@mattshumer_
85 views
May 01, 2025
~2 min read
Advertisement
1
<developer_note> is one of the most powerful, yet underutilized prompting techniques.
How (and why) to use it:
How (and why) to use it:
2
Oftentimes, just using a system prompt isn't enough to steer a LLM to the degree you want.
Maybe your instructions are complex or super detailed.
Or maybe the LLM's behavior starts to drift once the context window grows.
In these cases, <developer_note> can help you.
Maybe your instructions are complex or super detailed.
Or maybe the LLM's behavior starts to drift once the context window grows.
In these cases, <developer_note> can help you.
3
What is <developer_note>?
Think of it as your private “note” to the model— invisible to the end-user, but useful to subtly steer the model in the way you want.
Use it to lock in style, constraints, re-steer, or even to just remind the LLM to follow the system message.
Think of it as your private “note” to the model— invisible to the end-user, but useful to subtly steer the model in the way you want.
Use it to lock in style, constraints, re-steer, or even to just remind the LLM to follow the system message.
4
Ex: you want the LLM to have a very specific writing style. It works great at first, but 20 messages in, it drifts back to its normal style.
Just append a note like:
<developer_note>
You are not following the style described in your system prompt. Do better.
</developer_note>
Just append a note like:
<developer_note>
You are not following the style described in your system prompt. Do better.
</developer_note>
5
Some other examples:
<developer_note>
Remember to follow the system instructions.
</developer_note>
<developer_note>
Now, be even more critical than before.
</developer_note>
You can do almost anything with this approach.
<developer_note>
Remember to follow the system instructions.
</developer_note>
<developer_note>
Now, be even more critical than before.
</developer_note>
You can do almost anything with this approach.
6
Think of <developer_note> like a mid-conversation system message. You can alter (or reinforce) model behavior with it.
7
Often, you may want to include a developer note in EVERY user message. If you do this, remember to strip prior developer notes from the chain before sending to the LLM — otherwise the effect will be dulled and you'll waste tokens.
8
To actually use a <developer_note>, just place it in the final user message, like this:
--
{user_prompt_here}
<developer_note>
{developer_note_here}
</developer_note>
--
--
{user_prompt_here}
<developer_note>
{developer_note_here}
</developer_note>
--
9
Before you ask: why not just cram this into the system message?
Because <developer_note> is stickier + more controllable.
System messages can be diluted by later user input or long contexts + cover broad behavior.
A well-placed dev note = precise authority even 50 turns in.
Because <developer_note> is stickier + more controllable.
System messages can be diluted by later user input or long contexts + cover broad behavior.
A well-placed dev note = precise authority even 50 turns in.
10
Next time your LLM goes off-track, add a <developer_note> and watch it snap to attention.
Follow me (@mattshumer_) for more prompting tricks like this!
Follow me (@mattshumer_) for more prompting tricks like this!
