The Startup Ideas Podcast

Host: Greg Isenberg
Episode: How I code with AI agents, without being 'technical'
Date: January 7, 2026

Episode Overview

In this episode, Greg Isenberg dives into how non-technical people can code using AI agents, inspired by Ben Tossell’s widely-shared guide on the topic. Greg breaks down Ben’s workflow, experiments, and mindset shifts, illustrating how leveraging AI agents and command line interfaces (CLI) unlocks new creative and technical possibilities—even for those without traditional programming skills. The episode is packed with practical tips, lessons, and motivational insights for aspiring builders, makers, and anyone curious about the modern “vibe coding” era.

Key Discussion Points & Insights

1. The Rise of “Vibe Coding” with AI Agents

• Ben Tossell’s Experience: Spent “3 billion tokens in four months” coding entirely through AI agents via the terminal, without ever writing code himself. (00:33)
• Ben describes himself as a “vibe coder”—someone who doesn’t code traditionally but uses AI to get things built. Greg points out that this resembles the no-code movement, but leverages much more robust tools.
• Quote:

  "I don't read the code, but I read the agent output religiously. And in doing so I picked up a ton of knowledge around how code works, how projects work, where things fail and where they succeed." – Greg summarizing Ben Tossell (01:40)

2. What Ben Has Shipped Using AI & Agents

• Personal site: Built an interactive website styled like a terminal CLI.
• Social tracker: Automated feed to track social and GitHub mentions.
• Factory Wrapped: Internal analytics tool, later adopted by Ben’s company.
• Crypto signal tracker: Predicts sentiment and aggregates dynamic data.
• Droidmiss Project: 12 games/experiments built in 12 days on various AI-related themes.
• AI video demo system: AI-operated screen recording/directing/editing tool.
• Telegram bot & various CLIs: Tools for coding and deploying from anywhere, including from a mobile phone or Slack.

3. Embracing the CLI Over No-Code or GUI Tools

• CLI-First Approach:
  Ben “uses CLI exclusively, terminal over web interfaces always. It’s just more capable as a general agent and I get to see it work.” (09:43)
• Greg emphasizes moving from using friendly web UIs to using the terminal for maximal power—even as a non-coder.

4. Ben’s Workflow with AI Agents

• Idea > Project Spin-up: Starts everything in Droid, Factory’s CLI agent.
• Rapid Dialogue: Talks to the model several times for context before switching to “spec mode.”
• Spec Mode:

  “I'll basically question a bunch of things—like, I don't understand what this is or why we need that over this. Can't we do it this way?” (13:25)

• Iterative Building: Watch agent code creation, intervene on errors, iterate.
• Learning by Building Ahead: Gaps and bugs are seen as learning moments.

5. Agents MD: “ReadMe for AI Agents”

• Concept: A file (Agents MD) in each project repo acts as an instruction manual for AI agents, with standardized setup steps and context.
• Used by 60k+ open source projects.
• Ben customizes his Agents MDs for each repo to streamline future sessions and maximize agent effectiveness. (18:10)

6. Testing & Coding on the Go

• Installs the Droid GitHub app in every repo; uses pull requests and tags Droid to review/fix code with custom prompts.
• Remarkably, Ben codes from his phone—triggering fixes and editing on the go, even at a restaurant. (23:00)
• Slack channels are spun up for each project, letting Ben “fire off” ideas and tasks to agents directly.

7. Unlocks & Technical Skills Learned by “Non-Technical” Builders

• Bash Commands:
  Rediscovered while automating changelogs; now central to Ben’s workflow.
• CLIs over MCPs:
  Prefers command line tools over complex multi-component platforms for simplicity and transparency. (29:50)
• VPS (Virtual Private Server):
  Used to keep bots, trackers, and experiments running always-on—deploying, syncing, and experimenting freely.

8. The New Layer of Abstraction

• The old “no-code” layer was about drag-and-drop; today, it’s about mastering how to instruct and collaborate with AI agents.
• Quote:

  “What I need to learn is actually how to work with an AI agent: How can I prompt it well, how can I make sure it has the right context... How do I improve my system over time?” – Greg summarizing Ben Tossell (39:05)

9. Borrowing from the Community

• Follows actual programmers like Peter Steinberger for inspiration:

  “He just talks to the model, lets it do its thing, and doesn’t really worry about extra slash commands, subagents, or skills.” (42:10)

• Clones OSS projects, simplifies them, and experiments as his main mode of learning.

10. Learning Mindset & Psychological Permission

• “Ever-Patient Over-the-Shoulder Programmer:” Ben sees AI agents as a constant, judgment-free expert to ask “silly” or beginner questions.
• Learning Through PRs (Pull Requests): Watches and learns by inspecting PRs from experienced engineers.
• Key message: “No piece of software feels unattainable.” (53:00)

11. Why Coding Is Different Now

• The old path was years of incremental learning to become a “real” programmer. Now, being fluent in systems thinking and tools like Zapier, Webflow, and CLI AI agents can get you shipping real products, fast.
• Mindset shift: “Git clone it and play. Explore, have fun, ask dumb questions, learn quickly.” (58:36)

12. Failing Forward and Embracing Play

• Greg and Ben stress that experiencing errors and failures is the fastest way to learn—give yourself permission to play, be curious, and learn in public.
• Quote:

  “The way to learn about code is just to build ahead of your capability and fail forward.” – Greg Isenberg (01:06:29)

13. Just Get Started: Pick a Tool, Stick With It

• Don’t get paralyzed by too many choices (Droid, Cursor, Google’s Anti Gravity, etc.).
• Pick one agent/CLI that clicks for you and double down.
• “Any tool or feature I think I’m missing, I’ll have a crack at building myself.” (01:13:20)

Notable Quotes & Memorable Moments

• On learning by doing:

  “Every little hiccup, bug or issue you run into, question it.” (01:10:05)

• On AI as the ultimate tutor:

  “The model knows everything. It’s your ever patient, over-the-shoulder expert programmer.” (01:00:45)

• On the explosion of software:

  “We’re going to see 10,000x if not more... Most of it will be bad, horrible, horrible, horrible AI slop. But some of it will be incredible.” (01:03:50)

• On mindset:

  “Give yourself permission to go and do it… treat it like just a learning experiment, a sandbox for fun, [and] you’re going to come out the other side more powerful.” (01:15:30)

Timestamps for Key Segments

• 00:00 – Greg introduces Ben Tossell’s story and gives context for non-technical AI coding.
• 03:05 – Ben’s shipped projects using AI agents.
• 09:43 – CLI-first approach and why it matters.
• 13:25 – Workflow: dialogue with agents, spec mode, iteration.
• 18:10 – The role of Agents MD in project success.
• 23:00 – Coding and maintaining projects on the go with Droid, GitHub, Slack, and Telegram bots.
• 29:50 – Choosing CLIs over multi-component platforms, lessons from Bash, and VPS.
• 39:05 – The programmable layer of abstraction in today’s AI coding world.
• 42:10 – Learning by watching and remixing others’ open source projects.
• 53:00 – “No piece of software feels unattainable”—overcoming impostor syndrome.
• 58:36 – On exploration, play, and the new philosophy of building.
• 01:06:29 – Failing forward as the prime approach to learning.
• 01:10:05 – Actionable advice: start with something simple, question bugs, learn by doing.
• 01:13:20 – Just pick a tool—don’t over-optimize on platforms or features.
• 01:15:30 – Wrap-up: building, failing, and having fun.

Conclusion & Takeaways

Greg delivers a highly motivational and practical guide for anyone looking to start building with AI agents without extensive technical know-how. The key message: the barrier to software creation is lower than ever. Embrace AI agents, learn iteratively by building, and trust that playful exploration and failing forward are the best ways to advance. Being “non-technical” is no longer a block—it may even be your greatest asset in the new world of agent-powered creation.

Final message:

“Build, fail forward, and keep shipping. Treat it all as a big learning experience, and you’ll become way more powerful than you think.” (01:15:30)

For more startup ideas and practical tips, subscribe to The Startup Ideas Podcast and share this episode with someone curious about building with AI!