Part 2: Plan Your Agent from Scratch
Before touching any tools, you need to answer 4 questions.
This is the architectural thinking that separates agents that work from agents that break. Skip this step, and you'll spend weeks debugging something that was doomed from the start.
The 4 Questions Before You Build Anything
Every successful agent starts with clear answers to these questions. Vague answers produce vague agents.
“What does it do?”
List every task. Be specific. Not “social media” but:
The more specific you are here, the easier everything else becomes. If you can't describe it in one sentence, you're building two agents.
“Where does content come from?”
RSS feed? Newsletter? Manual input? Hub content? You need source material that flows in automatically.
An agent without an input source is a one-shot tool. The magic is in the continuous flow — new content arrives, the agent processes it, outputs go out. No human in the loop for the routine work.
“Who controls it?”
Email commands? Dashboard? Slack? The simpler the better.
We chose email because Peter already lives in his inbox. No new app to learn, no new tab to keep open. Send an email, get a result. That's it.
“What are the rules?”
Brand voice, character limits, forbidden words, UTM tracking, link policies.
An agent without rules produces garbage.
Rules are what turn a generic LLM into YOUR agent. Without them, you get content that sounds like every other AI-generated post on the internet.
Jenny's Architecture — A Real Blueprint
Here's how Jenny is actually built. Each component handles one job.
Source
RSS feed (aireadycmo.com/feed)
Scrapes new articles automatically
Brain
Claude Sonnet
Adapts content per platform with full brand rules
Validator
Zero-tolerance checks
No emojis, no hashtags, no corporate jargon, character limits
Publisher
Post for Me API
9 connected accounts, retry logic, batch scheduling
Scheduler
25 cron jobs
Morning, afternoon, evening, night batches + weekly reports
Control
Peter sends commands, Jenny executes and reports back
Storage
PostgreSQL
Articles, posts, analytics, partnerships, commands
You don't need to use these exact tools.
The architecture pattern is what matters. Swap Claude for GPT. Swap Post for Me for Buffer. Swap PostgreSQL for Supabase. The structure stays the same: source → brain → validator → publisher → scheduler → control → storage.
Start Small, Then Expand
Jenny wasn't built in a day. Here's the actual build progression:
Week 1
Basic content pipeline
Scrape RSS, adapt for 3 platforms, publish once a day
Week 1.5
Added email control
Pause/resume, request reports
Week 2
Multi-platform expansion
9 accounts, 6+ posts per day, multi-angle content
Week 2.5
Analytics + partnerships
Weekly reports, partnership discovery, PR automation
Week 3
Video generation + hub content
Remotion videos, 2,600+ hub items, sponsor outreach
The biggest mistake: trying to build everything at once.
Start with the content pipeline. Get that working reliably. Then add features one at a time. Each layer builds on the last. If the foundation is broken, nothing above it works.
Your Agent Planning Template
Before you write a single line of code or open any tool, fill this out:
Agent Name
Primary Task
Content Source
Output Channels
Control Method
Rules
First Milestone
Print this out. Pin it to your wall. Every decision you make during the build should trace back to these answers.
What You Just Learned
- Answer the 4 questions before touching any tool
- Study Jenny's architecture as a template — swap tools as needed
- Build incrementally — content pipeline first, everything else second
- The planning phase is the most valuable part of the entire build
Next: Define the brand voice system that turns generic AI output into content that sounds like you.