How I code with AI agents, without being 'technical'
The Startup Ideas Podcast
Workflow: context, spec mode, autonomy
Greg explains Ben's process: feed context, enter spec mode, question plans, then let the agent run and iterate.
In this episode, I’m breaking down a guide from Ben Tossel on how you can actually build with AI agents without being technical. I walk through what he’s shipped as a “non-technical” builder, why he lives in the terminal/CLI, and the exact workflow he uses to go from idea → spec → build → iterate. We also talk about the meta-skill here: treating the model like your over-the-shoulder engineer/teacher, and using every bug as a learning checkpoint. The takeaway is simple: pick a tool, ship fast, fail forward, and build your own system as you go.
Ben’s Article: https://startup-ideas-pod.link/Ben-Tossell-Article
Timestamps
00:00 – Intro
01:04 – What Ben Has Shipped
03:21 – The Workflow: Feed Context → Spec Mode → Let The Agent Rip
07:52 – His Agent Setup
08:56 – Coding On The Go
10:07 – Things to Learn
13:33 – The New Abstraction Layer: Learning To Work With Agents
14:33 – Learning from Others
16:15 – Use The Model As Your Teacher (Ask Everything)
18:13 – Contributing to Real Products
19:13 – Why this is Different
21:31 – Asking Silly Questions
24:00 – Beyond “Vibe Coding”: A New Technical Class
24:43 – Vibe Coding is a game
27:12 – Fail Forward + Permission To Build And Throw Things Away
28:16 – Pick One Tool, Minimize Friction, Keep Shipping
Key Points
I don’t need to be a traditional engineer to ship—I can learn by watching agent output and iterating.
The terminal/CLI is the power move because it’s more capable and I can see what the agent is doing.
“Spec mode” works best when I interrogate the plan like a philosopher instead of pretending I understand everything.
agents.md becomes my portable instruction manual so every new repo starts clean and consistent.
The fastest learning path is building ahead of my capability and treating bugs as checkpoints—fail forward.
Numbered Section Summaries
The Thesis: Non-Technical Doesn’t Mean Non-Builder I open with Ben’s core claim: you can ship real software by working through a terminal with agents, even if you can’t write the code yourself—because you can read the output and learn the system over time.
Proof: What He’s Actually Shipped I run through examples Ben built—custom CLIs, a crypto tracker, “Droidmas” experiments, an AI-directed video demo system, and automations that keep projects moving even when he’s away from his desk.
The Workflow: Context → Spec Mode → Autonomy High Ben’s process is straightforward: talk to the model to load context, switch into spec mode to pressure-test the plan, link docs/repos for exploration, then let the model run while he watches and steers when needed.
http://agents.md/ The “Readme For Agents” That Follows You Everywhere I explain why agents . md matters—one predictable place to tell your agent how you want repos structured, how to commit, how to test, and what “good” looks like so each session gets smoother.
Coding On The Go: PRs, Issues, Phone, Telegram, Slack We get into the real “agent native” behavior: install the GitHub app, work via pull requests and issues, tag the agent to self-fix, and even push changes from your phone—plus using Slack as a one-person “product” with an agent in the loop.
Learning The Primitives: Bash, CLIs, VPS, Skills I cover the building blocks Ben’s learning: bash commands and repeatable terminal workflows, preferring CLIs over MCPs to save context, and using a VPS + syncing to keep projects always-on.
The Mindset Shift: The Model Is The Teacher The real unlock is treating the model like your patient expert—ask everything you don’t understand, bake “explain simply” into your agent instructions, and close knowledge gaps as they appear.
Fail Forward, Pick One, Keep Shipping I end on the playbook: build ahead of your capability, treat it like play, give yourself permission to throw things away, and stop tool-hopping—pick one system and go deep.
The #1 tool to find startup ideas/trends - https://www.ideabrowser.com
LCA helps Fortune 500s and fast-growing startups build their future - from Warner Music to Fortnite to Dropbox. We turn 'what if' into reality with AI, apps, and next-gen products https://latecheckout.agency/
The Vibe Marketer - Resources for people into vibe marketing/marketing with AI: https://www.thevibemarketer.com/
FIND ME ON SOCIAL
X/Twitter: https://twitter.com/gregisenberg
Instagram: https://instagram.com/gregisenberg/
LinkedIn: https://www.linkedin.com/in/gisenberg/
The AI-powered Podcast Player
