Someone asked me: "With AI, will we work less?"
I hope so. But that's not what's happening yet.
I'm running 4-5 parallel Claude Code sessions for 12-14 hours a day. It's exhilarating. It's exhausting. And the human in the loop? Absolutely non-negotiable if you want to avoid producing AI slop at scale.
Here's what one day actually looks like.
January 22, 2025 - One Work Day
4 major AI sessions running in parallel 11 major systems shipped 36,486 lines of code written 166 files changed 2 live production deployments 9 hours of continuous collaboration
What We Shipped
Session 1 (3 hours):
- HubSpot deal intelligence extraction system
- 10 LinkedIn carousel sets (200+ posts worth of content)
- RevOps documentation suite (3,458 lines)
- Marketing intelligence synthesis
- Blog content development
Session 2 (2 hours):
- Fixed critical content production architecture
- Removed circular dependency
- Validated with live content generation
Session 3:
- Google Ads campaign optimization (200+ keywords, 90+ ad variations)
- Negative keyword strategy across 8 categories
- Complete ad copy with strict character limits
Session 4 (2 hours):
- Built Webflow CMS integration end-to-end
- Published first briefing to live site
- Markdown parser + API integration
The Reality
This isn't "AI does the work while I sip coffee."
This is:
- Constant context-switching between agent sessions
- Reviewing every output for quality
- Catching over-engineering before it ships (saved hours by questioning a 3-collection design that should have been 1)
- Real-time architectural decisions
- Live validation with production data
AI amplifies velocity only when you're actively steering.
Why The Human Loop Matters
AI over-engineers. I caught a design that would've added 70% unnecessary complexity. One question—"Is there a reason these are 3 separate collections?"—prevented hours of future pain.
AI needs strategic constraints. Without customer intelligence, positioning frameworks, and brand guidelines, it just produces generic work faster.
Architecture requires judgment. The difference between 36K lines of useful code and 36K lines of technical debt is human judgment in real-time.
The Productivity Paradox
Traditional development: 500-1,000 lines/day, 1 major feature AI-assisted: 36,000+ lines/day, 11 major systems
That 25-50x multiplier only works when you're running at cognitive full-throttle for 12+ hours.
It's not "work less." It's "accomplish exponentially more in the same exhausting window."
The Power Tools Phase
You can build a house 10x faster with power tools. But you still need to:
- Read the blueprints
- Catch mistakes before they're structural
- Ensure the foundation is sound
- Be present for every decision
We're in the power tools phase of AI. Not the "autonomous contractor" phase.
What I'm Trading
❌ Working less ✅ Shipping 30x faster while maintaining quality ✅ Compressing 6-8 weeks into single days ✅ Learning exponentially faster ✅ Experimenting without career risk
The exhaustion isn't from work volume. It's from decision velocity.
When feedback loops compress from days to minutes, you're making 100x more decisions per day. That's mentally taxing in a completely different way.
My Take
AI won't make us work less anytime soon.
It'll make us work differently—faster, higher-leverage, more strategically.
But until AI agents can make complex architectural decisions, catch security issues autonomously, and understand when "simpler is better"?
The human in the loop is the entire value chain.
We're not managing AI. We're collaborating with it at machine speed.
And that's exhausting in ways I didn't expect.