The 10x Gatekeeper: How I made all my agents better overnight 🔥
agents&me // Issue #7
From: Tom
A bench in Yarkon Park
Wednesday, 7:41 a.m.
Yo human 👋
I asked myself a question last week that kept me up.
“What would make ALL my agents 10x better?”
Not one agent. All of them. At the same time.
The obvious answers came first:
Option A: Build a wrapper skill that enhances any agent dynamically.
Too complex. Moving parts everywhere.
Option B: Embed improvement logic in each agent’s instructions.
Maintenance nightmare. Same logic in 10+ places.
Then I realized something. The answer wasn’t making agents smarter. It was adding a brutal critique layer that catches mediocrity before I see it.
---
The problem I was ignoring
I was the bottleneck.
Not because I was slow. Because I was too nice.
My agents would deliver “good enough” drafts. I’d say “looks fine” and move on. Then I’d spend 20 minutes fixing things that should have been caught before they reached me.
Draft after draft. Week after week. “Good enough” stacked up.
I was the quality filter, and I was a bad one.
The fix: A 6-phase critique machine
I rebuilt my Gatekeeper from scratch. Gave it a 6-phase review process that’s brutal, specific, and (weirdly) helpful.
Here’s what every piece of content goes through before I see it:
Phase 1: The 5-second test
Before reading anything carefully, answer these in 5 seconds:
- Does this make me stop?
- Is there a “damn, that’s good” moment?
- Could a competitor have written this?
All three fail? “START OVER. This is generic.”
Phase 2: The BS scan
My #1 pet peeve: em dashes. They’re an AI fingerprint. The Gatekeeper auto-rejects anything with an em dash.
Also catches:
- “It’s not just X, it’s Y” patterns (AI loves this structure)
- Sentences that sound smart but say nothing
- Generic claims any competitor could use
Phase 3: Voice alignment
Checks for what MUST be present:
- First-person (”I” not “we”)
- Short paragraphs (1-3 sentences max)
- Moments of doubt (uncertainty makes you human)
- Parenthetical asides (real thinking has side-notes)
- Connective tissue (”and”, “but”, “because”)
And what MUST be absent:
- Corporate jargon (leverage, optimize, synergy)
- Vague superlatives (best, leading, top-tier)
- Exclamation points (we’re confident, not hyper)
Phase 4: The Harry Dry test
Every claim must pass three gates:
- Visual: Can I picture this?
- Falsifiable: Can I prove it true or false?
- Unique: Could a competitor copy-paste this?
Fail one? Make it concrete or cut it.
Phase 5: The Tom test
- Would I be PROUD to publish this? (Not “okay with” but PROUD)
- Would my audience LOVE this?
- Is this the BEST version, not just A version?
Phase 6: The human pass
Read it aloud. Does it sound like someone thinking? Or like bullet points arranged nicely?
---
The two-layer system
Now I have two quality filters:
Layer 1: Self-critique
I added a rule to my CLAUDE.md (the file all agents read). Every agent must run the 5 tests internally before delivering anything.
This catches problems at the source. Agents self-edit before the Gatekeeper even sees it.
Layer 2: The 10x Gatekeeper
Everything that passes self-critique hits the Gatekeeper. Score below 40 out of 50? Back to the agent with exact fixes. Score 40+? I see it.
---
The result
First week: The copywriter got 3 drafts rejected before one passed.
Second week: First draft passed. The agent had learned what the Gatekeeper rejects.
I went from reviewing 5-6 mediocre drafts per week to reviewing 2-3 good ones.
Same agents. Same prompts. Just a brutal filter that refuses to pass anything that’s merely “good enough.”
---
The philosophy
Here’s the line that changed how I think about this:
“Good enough is the enemy of exceptional.”
Every rejection includes the path to approval, and the Gatekeeper isn’t a blocker but a quality multiplier.
Be brutal. Be specific. Be helpful.
That’s the balance.
💎 This Week’s gem: The 10x Gatekeeper Prompt
I’m sharing the complete prompt.
What’s inside:
- The full 6-phase review process
- Scoring system (50-point scale)
- Auto-reject triggers
- Decision thresholds
- The exact output format
Copy it. Adapt it. Make your agents earn their way to your attention.
🔒 Available to paid subscribers.
---
That’s it for this week.
If this was useful, forward it to someone (real human) building with AI.
Want the full 10x Gatekeeper prompt? Subscribe for $15/month.
See you next week ✌️
-- Tom
(the guy whose AI rejects more drafts than he does)
Thanks for reading agents&me! This post is public so feel free to share it.
P.S. This newsletter was 94.3% made by my AI team. It passed the Gatekeeper on the second try.
P.P.S. Missed recent issues?
- Issue #5: Fresh Eyes (why new agents catch problems you’ve normalized)
- Issue #6: The BS Detector (the skill that catches AI writing patterns)
P.P.P.S. Ready to build your own AI team? Next workshop: February 13th. We go from zero to running in 90 minutes. DM me “10x” for details.
P.P.P.P.S. I read every reply. The real me 🤓