Check out the conversation on Apple, Spotify and YouTube.
Brought to you by:
Miro: The innovation workspace
Jira Product Discovery: Build the right thing
The AI PM Certificate: The #1 AI PM certificate
Today's Episode
You use ChatGPT.
But being an AI-powered PM means also using AI agents.
In my slack poll, only 2% of you said you use AI agents for productivity.
So I want to break that down and make it dead clear: 1) why you should use AI agents and 2) how you should build them.
So in today’s episode, I’ve brought in Jacob Bank, former Director of PM at Google (Gmail, Calendar) and now CEO of the AI agent builder company Relay.app.
He shares all his secrets - his 12 agent EA, his 40 agent marketing team, and his agent to synthesize agent updates.
I hope you enjoy.
(If you liked my episode with Lindy CEO Flo Crivello, you’ll love this.)
Your Newsletter Subscriber Bonus
For subscribers, each episode I also write up a newsletter version of the podcast. Thank you.
Today’s is the ultimate guide to AI agents for PM productivity:
How to Build a $19/Month Exec Assistant
Advanced Topics in AI Agent Building
When to use workflows
When to have human-in-the-loop
When to choose which model for a task
How to Build a 40-Agent Marketing Team with 0 employees
1. How to Build a $19/Month Executive Assistant
Every PM should copy Jacob’s 12-agent "executive assistant":
It replaces what most executives pay $2,000+ monthly for human assistants with 12 different agentic workflows.
A few interesting things he does with the individual agents:
His competitive intelligence tracker is so realtime that when competitor Gumloop introduced a lower-priced plan, Jacob got immediate Slack notification with pricing details and comparison to previous plans
Jacob's meeting briefing generator creates personalized dossiers 30 minutes before every meeting. It actually analyzes past emails, previous meeting notes, and LinkedIn profiles to give context-aware briefings
His follow-up drafter took him from 25% completion rate to 95%+ by automating the grunt work
You should steal his template and start there, and then keep up-leveling each of your EA agents over time.
2. Advanced Topics in Building AI Agents
The difference between toy demos and production agents comes down to three critical design principles Jacob discovered through trial and error.
2a. When to Use Workflows vs "True" Agents
The reality is most successful "AI agents" are actually sophisticated workflows with AI steps. True autonomy works only for simple, low-stakes tasks or expert prompters.
Predefined workflows outperform autonomous agents for most business use cases.
Workflow approach: Step 1 → Step 2 → AI Step 3 → Step 4 (you control the flow)
Agent approach: "Monitor my brand on Reddit with these tools. Good luck. Go."
Workflows succeed 95%+ of the time; autonomous agents often fail unpredictably.
2b. When to Have Human-in-the-Loop
Don’t leave everything to agents. Have a human in the loop.
Jacob uses a simple matrix to decide when humans should review agent outputs:
High AI capability + Low stakes: Fully autonomous (competitor pricing tracking)
High AI capability + High stakes: Human review required (customer emails, LinkedIn posts)
Low AI capability + Any stakes: Heavy human involvement needed
This prevents the common mistake of either micro-managing everything or letting agents run wild on important communications.
2c. When to Choose Which Models
Use different AI models for different tasks based on empirical testing:
GPT-4o: Individual research and dossier creation
Claude Sonnet 4: High-quality writing tasks
Gemini: Large PDF parsing (longest context window, cost-effective)
As he said, "Anything I tell you about specific models will be wrong in 3-4 months." The skill is rapid model evaluation and swapping, not memorizing current rankings.
3. How to Build a 40-Agent Marketing Team (0 Employees)
Jacob's most viral LinkedIn post showed his "org chart" of 40 AI agents organized by marketing channel:
For instance, the traditional webinar process is product marketer + email marketer + webinar specialist + PM + months of advance planning.
Jacob's process? Create Google Calendar event → Show up → AI handles everything else.
He can go from 4-person team coordinating for months to one person creating a calendar event.
How Do You Deal with So Many Agents?
So how does he deal with all the updates from the agents?
More agents that synthesize and prioritize.
It’s “agents all the way down.”
Newsletter digester: Daily 5 PM summary of all newsletters with 3-sentence summaries and drill-down links
Scheduled cadence: Competitive research on Fridays, content ideas on Saturdays, support summaries on Mondays
Natural workflow integration: Ad-hoc agents (YouTube → LinkedIn post) trigger during normal work flow
The meta-insight: Use AI to manage AI-generated information. The goal isn't fewer notifications but smarter prioritization.
Key Takeaways
Where to Find Jacob
BTW, I gave a talk at BYU on how to become a PM as a college student. Here are the slides (another subscriber bonus):
Related Content
Podcasts:
Newsletters:
P.S. More than 85% of you aren't subscribed yet. If you can subscribe on YouTube, follow on Apple & Spotify, my commitment to you is that we'll continue making this content better.
