Two weeks into my 100-day AI journey, I got invited onto the Verbos podcast to talk about something I was still figuring out: what does it actually mean to be a "100x developer" in 2025?
This conversation with Kasper happened at exactly the right moment — when I was still riding the wave of discovery, bumping into code that spiraled into chaos, learning to pair Bolt with V0, and starting to understand that AI development isn't about letting the AI do everything. It's about knowing when to step in, when to let go, and how to think differently about the entire workflow.
2025-01-09 · 1:03:16
Episode Chapters
- Introduction to AI and Development (0:00)
- Silicon vs. Carbon Developers (5:51)
- Developing with AI Tools (12:04)
- React and AI Combined (21:24)
- The Low-Code/No-Code Revolution (26:52)
- Integrating AI into Development Tools (35:08)
- Productivity and Creativity (41:28)
- Interaction Between AI Agents (46:43)
- Funding and Rapid Idea Production (54:03)
- Closing and Future Visions (1:00:02)
Highlights: The Breakthroughs We Discussed
1. Silicon vs. Carbon Developers (5:51)
The concept I started posting about on LinkedIn: developers made of silicon (AI) versus those made of carbon (us, humans). It's not about replacing one with the other—it's recognizing that we're now working alongside something fundamentally different. Kasper picked up on it immediately, asking if someone using AI becomes a "cyborg." I wasn't sure about the terminology, but the idea landed: we're entering a phase where developers need to think about what type of agent to deploy and when.
2. The Productivity Cliff and How to Navigate It (9:37)
I talked about a curve I'd seen showing AI productivity—it spikes up fast, then crashes down hard as you realize you've built a house of cards you don't understand. The breakthrough for me was realizing: "If you use that first spike to get 80% done and have the skills to finish the final 20%, you win massively." That shifted how I approach AI-assisted code from "let it all rip" to "use it strategically."
3. Bolt and V0: Picking the Right Tool (14:42)
I explained how Bolt.new works—it's Claude with a system prompt, running in a web container so you see changes instantly. And V0 (from Vercel) is optimized for Next.js components specifically. The key learning: "Accept that AI generates best in the frameworks it's seen most in training data." Rather than fighting to convert everything to what I prefer, I started using each tool where it excels. Bolt for full-stack, V0 for React components. Both.
4. Building My Own System Prompt (18:22)
When I looked at Bolt's open-source code on GitHub, I realized something that clicked: "There's no magic. It's just Claude with a system prompt." That moment unlocked everything. I could build my own custom agents for specific tasks—a project manager agent, an accessibility checker agent, a framework-specific coder. The building block is surprisingly simple if you understand prompting.
5. Context Windows and Smart Filtering (30:50)
I built a CLI tool called ccon (copy context) to automatically collect all relevant files and send them to the AI with full context. But context windows explode fast. The insight: instead of sending everything, I'm working on intelligent filtering—only include template files for Django projects, only views.py, only the method signatures. Massive token savings, better results.
6. The Real Advantage: Iteration Speed (41:28)
"The thing that really changed for me is iteration speed." I used to spend days on a landing page, obsessing over animations and design details. Now I can sketch something in minutes with Bolt, validate if the idea works, and move on. For weekend projects, that's revolutionary. For customer work, it means more time on what actually matters—architecture, security, testing—instead of the boilerplate.
7. Two-Agent Flows for Large Projects (46:43)
For big codebases, I'm experimenting with chaining agents: a design agent that figures out requirements and writes a spec, then hands it off to a coding agent with just that spec (not the entire repo). Reduces token usage, keeps each agent focused, and mirrors how human teams actually work. Kasper asked the right question: does a "master agent" pick which agent to use? Still exploring that.
8. Ideas Win Now, Not Execution (54:03)
A colleague told me they can now produce funding proposals in 10 days that used to take 3 months. The shift isn't "more money for the same ideas"—it's "now the good ideas actually win because anyone can execute quickly." Before, whoever invested in the process (mastered the grant writing game) won. Now, faster feedback means better ideas survive. That's a fundamental change in how competition works.
9. The Last Generation of Traditional Developers (59:05)
"I think we're the last generation of developers who had to learn bits, bytes, memory management, C++ from age 14 just because we liked it." Future developers won't need that foundation—they'll start with a different set of skills entirely. AI handles the plumbing; they'll focus on judgment and architecture. Not good or bad, just different.
10. Building in Public, Building Faster (1:00:42)
I launched POC Graveyard (poc-graveyard.com) in roughly an hour: buy domain, deploy, ship. Pre-AI, that sounded exhausting. Now I can throw ideas at the wall and see what sticks without the ceremony. For someone who loves building but hates deployment friction, this changes everything.
Where This Connects
This conversation sits right at the beginning of my AI development journey. If you want the full story:
- 1-1: Ten Websites for Two Dollars — The LinkedIn post that sparked this whole thing
- Related: AI and React Together — Deep dive into tool pairings (coming soon)
- Related: Building with Agents — How I structure multi-agent workflows (coming soon)
The episode itself is in Danish, so you'll hear the full conversation with all the tangents, questions, and moments where Kasper pushed back on my thinking. That's where the real value is—not in my conclusions, but in how we got there.
One of the most surreal moments: my first podcast appearance about AI coding, in the middle of my first serious AI experiment. Recording this at the end of December, not yet knowing how far this would go. Grateful to Kasper for asking the right questions and calling out the over-hype moments. The skepticism helps.