I categorize agent building into 2 modes:
Mode 1: Chief of Staff - I typically don't ask a CoS to tell me the strategy that's going to save the company.' More like, 'go to Chipotle and get me my food while I think about what's going to save the company.
It’s checklists, prep, formatting, and research gathering
Mode 2: Persona Automation - I’m meticulous about recording the meeting, taking notes... and I do that two, three, four, maybe after three or four or five times.
Now I have a Persona, a replacement with decision making, judgment, and pattern recognition capabilities
I have operated in Mode 1 for months and realized that ultimately building a CoS isn’t my North Star.
Automation with domain expertise in GTM and commerce vertical is.
I broke down everything I do into 4 layers:
Layer 1: Information Processing
Research prospects. Draft proposals. Track pipeline. Generate reports.
This is what most people automate first. I did too. My agents handle about 70% of this now. Prospect research went from 2 hours to 5 minutes. Proposals from 4 hours to seconds.
But this is the easy part. This is labor, not judgment.
Layer 2: Pattern Recognition
"This deal feels off."
"She's the real decision maker, not him."
"Lead with ROI, not vision, for this buyer."
"They'll close in Q2, not Q1."
20 years of reading rooms, reading people, reading between the lines, and establishing a good judgement of characters.
None of this is encoded anywhere. It lives in my head. And if it stays there, it dies with my schedule.
Layer 3: Relationships and Trust
The handshake. The follow-up text. The conference dinner is where the real conversation happens.
This may never be fully automated. But it can be dramatically augmented.
Layer 4: Strategic Vision
"We should pivot to this market."
"This is the next big opportunity."
This is the last thing to automate.
This is Steve Jobs territory with sharp intuition.
Most people are still operating in Mode 1.
Mode 2 is the holy grail:
where the agents prep the call,
tell me what to say,
score the prospect,
decides on pass or no pass,
explain the why,
read my call transcript,
flag specific word choices,
and update the workflow and framework based on contextual feedback.
Every decision I make now, I ask one question:
"Could an agent do this? If not, what would it need to know?"
That question changes everything about how we build at THE OTHER GROUP.
