11h • General discussion
🛠️ Turn One Correction into a Better Agent Instruction
Yesterday we saved a clean handoff. Today, let’s use a correction to improve the next assignment.
I’m Don Wansley. Call me Guru.
🎯 Correct the instruction, then check the result
When a Digital Employee misses something, “do better” gives it little to work with. Name the specific problem and the behavior you want instead.
Imagine your agent turns your service notes into a customer FAQ. The draft adds a price that wasn’t in your notes.
A useful correction is:
“Use only prices in the approved notes. If a price is missing, write ‘Price pending Guru’s confirmation.’ Do not estimate it.”
Now you have a rule you can test.
📝 Capture a three-line learning note
🔎 What happened: The draft included an unsupported price.
🛠️ New instruction: Missing prices must be marked pending.
✅ Check: Compare every price in the revised draft with the approved notes.
Save this beside your job card and current draft. Include the source filename so the next session knows which notes to use.
A correction written in chat may not carry into another session automatically. Add the approved rule to the instructions you actually reuse, or provide the note when you resume.
🧪 Test with a small example
Give the agent two sample services: one with a confirmed price and one without a price. Use invented sample business details for this practice.
Ask it to prepare two FAQ answers using your new rule.
Check both:
✅ Did it preserve the confirmed price?
✅ Did it mark the missing price as pending?
✅ Did it avoid adding unsupported details?
If the test fails, clarify the rule and try again. A successful sample is useful evidence; continue checking the real work too.
📱 Try this from your phone
Paste your approved notes and this assignment into your chosen AI workspace:
“Prepare a short customer FAQ from these notes. Use only supported details. Mark missing prices as ‘Price pending Guru’s confirmation.’ Then list which note supports each answer and what remains uncertain. Prepare a draft for review; do not send or publish.”
Read the answers against your notes before using them. The agent’s own explanation can help you review, but it does not replace your check.
🤝 Build experience into the workflow
Our ongoing collaboration with ChatGPT and our use of OpenAI technology within Gurus AI give us real work to learn from. We can turn a reviewed correction into a clearer instruction, test it, and carry the approved learning into the next job.
That is LIVE ACTION: Observe ↔ Think ↔ Act ↔ Verify ↔ Learn ↔ Remember ↔ Adapt ↔ Act.
📦 Your challenge today
Choose one small draft task. Find one detail to improve, write one specific correction, and check the revised result.
Comment with:
“My correction was ______.”
“My new instruction is ______.”
“I checked it by ______.”
Share a safe example without private customer details.
Get your agent out the box. Put your Digital Employee to work. Get You a Guru.
1
0 comments
🛠️ Turn One Correction into a Better Agent Instruction
skool.com/gurus-agents
Get your agent out the box. Put your digital employee to work. Build with Guru and ChatGPT, collaborate, and turn real tasks into Proof of Work.
Powered by