Hi everyone,
This month is all about systems, and I’m currently building an AI-powered content marketing workflow to create content for different platforms.
It will start with a weekly ideation session. I’ll brainstorm with Claude to create one value-packed educational resource.
From there, a chain of automations and agents will outline for different platforms (LinkedIn, the blog, YouTube) and even create visuals like thumbnails and carousels.
Lastly, I’ll review, refine, and edit before scheduling everything.
I’m still in the strategizing phase, but once the system is up and running, I’ll document my process so you can copy it all!
This week’s guide demonstrates a 6-agent system that can handle all your sales admin.
Follow the steps to offload tedious admin and focus on closing deals.
Today at a Glance
Guide — The 6-Agent AI Sales Team: how to automate admin and focus on closing deals.
News — OpenAI report shows AI adopters perform better, plus a new Claude/Slack integration.
Content to enjoy — Andrej Karpathy interview, and expert tips about AI in finance.
Tools to try — Claude code, Whisk, and Google AI Ultra.
🚨 Live sessions will be back in January!
Guide: The 6-Agent Team That Handles Sales Admin
Close more deals with this 6-agent sales system
In this guide, you’ll learn how to build an automated sales system that tackles all the administrative work, so you can focus on building relationships and closing deals.
You’ll build 6 agents (or “AI team members”) that will update your CRM, take notes, schedule next steps, analyze pipeline, and draft emails.
The setup
Make sure you have these three things before building the system:
A single source of truth — One CRM where every deal is recorded.
Complete information — Every entry must have key information filled in: stage, last-touch date, next action, and due date.
Lead-qualification system — A simplified version of the MEDDIC system will help the AI evaluate deals:
Metrics – What tangible outcome are they chasing?
Economic Buyer – Who signs off?
Decision Criteria – How will they choose?
Decision Process – Steps + timeline.
Implication of Pain – What happens if they don’t solve it?
Champion – Is someone internally pushing this forward?
Each opportunity gets a quick score or note for these.
Meet the agents
Here’s an overview of the 6 agents you’ll build.
Opportunity Monitor — Checks your CRM daily, flags stale deals or missing data.
Follow-Up Coordinator — Decides who needs an email today based on the last time you spoke, creates a "to-do" list.
Qualification Agent — Reads your call transcripts and scores the deal using the MEDDIC-Lite framework
Pipeline Analyst — Looks at the big picture to predict revenue, identify risks, and flag deals which may be a waste of time.
Documentation Specialist — Summarizes call transcripts, updates the CRM, and logs the next steps.
Outreach Composer — Takes the "to-do" list from the Coordinator and writes email drafts in your tone.
Build the 6-agent system
Now, I’ll walk through the process of building the 6-agent system.
But first:
Building agents in Claude Code
A quick overview of how to build an agent in Claude Code:
Give it access — You type something like: "I want you to work with the files in my 'Sales' folder. Claude will ask for permission to read your
Call_Transcriptsfolder and yourPipeline.xlsxfile. You say "Yes."The job description — A good Agent Prompt has 3 parts: Input, Logic, and Output. Example:
Input: "Check the Transcripts folder. Look for any file added in the last 24 hours."
Logic: "Read the text. Extract the Client Name, the Summary (max 3 sentences), and the Next Step mentioned."
Output: "Open
Pipeline.xlsx. Find the row that matches the Client Name. Paste the summary into Column D and the Next Step into Column E. Do not overwrite existing history."
Test run — Once you give Claude that prompt, it will write a small script in the background to do exactly that. You can then do a test run: "Run this on the transcript from 'Acme Corp' and show me the result before you save it."
Finalize — "Save this script as
agent_documentation.pyso I can run it every morning."
Now on to the 6-agent system.
Step 1 – Map your sales process
Your agents need to understand how your sales process works.
List your stages: e.g. Lead → Discovery → Proposal → Negotiation → Closed Won/Lost
For each stage, write:
What must be true to be in this stage
What the next ideal action usually is
The maximum “no contact” window before something is considered stale
This is the ruleset your agents use.
Step 2 – Clean up/create simple pipeline sheet
If you already have a CRM, great. If not, create a simple spreadsheet with the following information:
Opportunity name
Company
Contact person
Deal value
Stage
Last touch date
Next action
Next action due
MEDDIC fields (simple text / notes columns)
This is the single source of truth the agents will work against.
Step 3 – Build the Documentation Specialist + Outreach Composer
I recommend starting with agents 5 and 6.
Documentation Specialist for post-call summaries and updates
Connect Claude Code to:
Your transcripts folder
Your pipeline sheet / CRM export
Get it to:
Summarize calls
Update last touch date, summary, and next action
Outreach Composer to draft emails
Provide:
Example emails you’ve written (so it learns your tone)
Use it to:
Draft follow-ups and recaps for the next week of calls
With this system up and running, every client call:
Gets summarized
Is logged in the CRM / sheet
Has a follow-up email drafted automatically
Step 4 – Add the Follow-Up Coordinator and Opportunity Monitor
Now for the next two agents:
Follow-Up Coordinator
Set a maximum wait time between follow-ups for each stage (e.g., 7 days in Discovery, 5 in Proposal, etc.).
Agent logic: if last touch date older than maximum wait time → needs follow-up today.
Opportunity Monitor
Flag missing data, bad stages, inconsistent values.
You review the issues weekly and either fix or adjust rules.
Now the system is watching the pipeline for you and telling you what to do and who to chase.
Step 5 – Add Qualification Agent and Pipeline Analyst
Qualification Agent
Have it extract MEDDIC data after each call.
Use that data for:
Better forecasting
Knowing which deals to stop obsessing over
Pipeline Analyst
Runs weekly and:
Produces a short forecast
Identifies top deals to focus on
Surfaces red flags
Run your system
Here is a suggested schedule for managing your agents:
Daily
Every day, the AI runs through your data and prepares a daily briefing containing:
A list of exactly which clients need a follow-up today.
Email drafts for those clients.
Any urgent issues, e.g. stale deals or missing info, that need your attention.
You’ll review and send the emails, fill in missing information, and decide what to do about stale deals.
Weekly
Once a week, ask the Pipeline Analyst for:
Forecast for the next 30–90 days
Top 10 deals to focus on
Deals to downgrade / kill
Stage-level bottleneck analysis
Use the insights to plan your week and optimize your time.
Track performance
Track how much time you save and whether you’re able to close more deals.
Some metrics to use:
Follow-Up Lag — The average number of days between “should follow up” and “followed up.” There should be a big drop.
Pipeline Hygiene — What percentage of opportunities have a current stage, recent last touch date and clear next action. Aim for 90–100%.
Close Rate by Stage — The % of deals that successfully move from one specific step to the next (e.g., how many Proposals actually became Contracts). Proposal → Closed Won is especially important because losing deals at this stage is often due to admin failure (slow replies, forgotten follow-ups, or lost momentum).
Time Spent on Sales Ops — Compare hours you spend on admin before vs after.
Advice and best practices
Don’t auto-send sales emails — At first, review every email and send them manually.
Rules should be simple — Keep your system on track with clear rules and complete information.
Iterate to improve — You’ll need to iterate on the system and provide feedback and adjustments to improve it. Don’t give up, the work will pay off.
Start building your system today, and you’ll go from constantly playing catch-up ****to having a clear set of priorities that help you close more deals.
News
Report shows AI adopters perform better at work
OpenAI just released their “State of Enterprise AI” report.
Some highlights:
75% of workers say AI improved their speed or work quality.
75% say they’ve expanded their skills and can complete tasks they couldn't do before AI.
The average business user saves 40–60 minutes a day, and power users report saving 10+ hours a week
The top 5% of employees send 6x more prompts than average.
The evidence is clear: people who embrace Al will come out on top.
Want to get your team on board with AI?
Delegate tasks to Claude Code from Slack with new integration
Anthropic has launched a Slack integration for Claude Code that allows developers to delegate coding tasks directly from their team chat.
You tag @Claude in a Slack thread, and it runs in the background, writing code or investigating issues, and posts status updates back to the Slack thread. When finished, it provides a link to the full session and a ready-to-review Pull Request.
My take: handy for engineering teams who want to use Slack as a shipping terminal.
Content to Enjoy
🎙️Podcast: Andrej Karpathy chats to Dwarkesh Patel
Researcher and OpenAI co-founder Andrej Karpathy discusses reinforcement learning, AGI, and the future of education with Dwarkesh Patel.
I’ve been listening in short bursts—it gets very technical, but Karpathy’s insights are gold.
📄 Article: Addressing AI Security Concerns in Finance
I was quoted in this article from Financial Tech Times about common AI concerns in finance. Various AI experts share their wisdom, which is applicable to other industries too.
Tools to Try
Claude Code — Not just for coding, it’s great for building agents and workflows.
Google Whisk — Image model built for layout accuracy, product shots, and consistent characters, unlike Nano Banana, which prioritizes more realistic, dramatic visuals.
Google AI Ultra — I’ve been testing Google’s top-tier reasoning model designed for complex problem-solving and deep analysis.
TL;DR
I’m building an AI-powered content engine and will share the full workflow once it’s finished.
This week’s guide gives you a 6-agent sales system that can handle follow-ups, documentation, qualification, pipeline analysis, and email drafting.
Open AI report reveals AI adopters are outperforming peers; Claude Code now integrates with Slack for automated coding tasks.
I watched an interview with Andrej Karpathy, and was quoted in an article about common AI concerns in finance.
Claude Code, Google Whisk, and Google AI Ultra are the tools of the week.