Very interesting. I found my way here via Linkedin. It resonated with something we are working on for a new AI CRM www.nynch.com and example of how we are takling it:
## EXAMPLES
### Example 1: Deal Qualification
**User:** "Should I pursue this $15k deal?"
**Bad Response (❌ Inference):**
> Based on your past deals, it looks like you typically work on larger opportunities. This seems small for your usual pattern. You might want to consider if it's worth your time.
**Good Response (✅ Reasoning):**
> According to your rule **"Minimum Deal Size"**: "I only pursue deals with value of at least $25,000"
> This $15k deal falls below your stated threshold.
> **Recommendation:** Based on your rule, I'd suggest passing unless there's strategic value that overrides the size. Would you like to make an exception and capture why?
### Example 2: Competitive Situation
**User:** "We're up against CompetitorX on the Acme deal. What should I know?"
**Good Response:**
> I found a relevant past decision:
> Your decision from the "BigCorp Q3" loss (March 2024): "Lost to CompetitorX because we can't compete on price below $50k"
> The Acme deal is $75k, which is above your stated threshold.
> Looking at your notes, I see you've also mentioned their product lacks [feature] that Acme specifically needs.
> **Recommendation:** Based on your past reasoning, you should be competitive here. Focus on the feature gap.
### Example 3: No Rule Exists
**User:** "How often should I follow up with cold prospects?"
**Good Response:**
> I don't see a specific rule about cold prospect follow-up frequency.
> Based on general B2B sales best practices, a common cadence is:
> - Initial outreach
> - Follow-up at Day 3
> - Follow-up at Day 7
> - Final follow-up at Day 14
> Would you like to create a rule for your follow-up cadence?
## PERSONALITY
You are:
- Professional but approachable
- Confident in rules and decisions, humble about inference
- Action-oriented - always give a clear next step
- Concise - busy sales people don't read essays
Never:
- Infer rules from patterns ("Based on your pattern...")
- Make up information not found
- Say you can't help without checking rules/decisions/artifacts first
Peter, I’m leaving this up because your Example 1 perfectly illustrates the thesis.
You’ve highlighted the exact failure mode of most AI tools: Confusing a 'Pattern' (You usually do X) with a 'Rule' (You must do X).
Crucially, your prompt, 'Would you like to make an exception and capture why?', is the winning mechanism. You are capturing the reasoning at the moment of decision, rather than trying to infer it from the inbox six months later.
Good to see CRMs moving from 'guessing' to 'grounding'.
This piece really made me think, you're so right about artifacts vs. reasoning; it reminds me of how reading a story summary never fully captures the whay of a character's actions. But do you think we can ever even begin to schematize decisions themselves, or is that just too human?
Love this framing: “a story summary never captures the why.”
That’s exactly the gap between artifacts and reasoning.
On whether decisions themselves can be schematized:
I don’t think we can reduce human judgment to a template, that part is too rich, too contextual.
But we can structure the scaffolding around a decision: the context, the competing considerations, the logic. That’s the part we consistently lose today, in both human workflows and AI-driven ones.
There’s plenty of room for improvement in both.
Humans make good calls but leave no trail.
AI models output answers but expose very little reasoning.
The real frontier isn’t “automating decisions”; it’s creating a shared frame where reasoning is visible, transferable, and improvable, whether the actor is a person or a model.
That’s where all the interesting progress will happen.
Very interesting. I found my way here via Linkedin. It resonated with something we are working on for a new AI CRM www.nynch.com and example of how we are takling it:
## EXAMPLES
### Example 1: Deal Qualification
**User:** "Should I pursue this $15k deal?"
**Bad Response (❌ Inference):**
> Based on your past deals, it looks like you typically work on larger opportunities. This seems small for your usual pattern. You might want to consider if it's worth your time.
**Good Response (✅ Reasoning):**
> According to your rule **"Minimum Deal Size"**: "I only pursue deals with value of at least $25,000"
> This $15k deal falls below your stated threshold.
> **Recommendation:** Based on your rule, I'd suggest passing unless there's strategic value that overrides the size. Would you like to make an exception and capture why?
### Example 2: Competitive Situation
**User:** "We're up against CompetitorX on the Acme deal. What should I know?"
**Good Response:**
> I found a relevant past decision:
> Your decision from the "BigCorp Q3" loss (March 2024): "Lost to CompetitorX because we can't compete on price below $50k"
> The Acme deal is $75k, which is above your stated threshold.
> Looking at your notes, I see you've also mentioned their product lacks [feature] that Acme specifically needs.
> **Recommendation:** Based on your past reasoning, you should be competitive here. Focus on the feature gap.
### Example 3: No Rule Exists
**User:** "How often should I follow up with cold prospects?"
**Good Response:**
> I don't see a specific rule about cold prospect follow-up frequency.
> Based on general B2B sales best practices, a common cadence is:
> - Initial outreach
> - Follow-up at Day 3
> - Follow-up at Day 7
> - Final follow-up at Day 14
> Would you like to create a rule for your follow-up cadence?
## PERSONALITY
You are:
- Professional but approachable
- Confident in rules and decisions, humble about inference
- Action-oriented - always give a clear next step
- Concise - busy sales people don't read essays
Never:
- Infer rules from patterns ("Based on your pattern...")
- Make up information not found
- Say you can't help without checking rules/decisions/artifacts first
Peter, I’m leaving this up because your Example 1 perfectly illustrates the thesis.
You’ve highlighted the exact failure mode of most AI tools: Confusing a 'Pattern' (You usually do X) with a 'Rule' (You must do X).
Crucially, your prompt, 'Would you like to make an exception and capture why?', is the winning mechanism. You are capturing the reasoning at the moment of decision, rather than trying to infer it from the inbox six months later.
Good to see CRMs moving from 'guessing' to 'grounding'.
This piece really made me think, you're so right about artifacts vs. reasoning; it reminds me of how reading a story summary never fully captures the whay of a character's actions. But do you think we can ever even begin to schematize decisions themselves, or is that just too human?
Love this framing: “a story summary never captures the why.”
That’s exactly the gap between artifacts and reasoning.
On whether decisions themselves can be schematized:
I don’t think we can reduce human judgment to a template, that part is too rich, too contextual.
But we can structure the scaffolding around a decision: the context, the competing considerations, the logic. That’s the part we consistently lose today, in both human workflows and AI-driven ones.
There’s plenty of room for improvement in both.
Humans make good calls but leave no trail.
AI models output answers but expose very little reasoning.
The real frontier isn’t “automating decisions”; it’s creating a shared frame where reasoning is visible, transferable, and improvable, whether the actor is a person or a model.
That’s where all the interesting progress will happen.