Opinion: Frog Squad prefers debating specs over using Claude Code or Codex
The author describes how the Frog Squad deliberately avoids relying on AI code generators such as Claude Code and Codex. Instead, they spend time debating what to build and refining specifications, using the AI as a sounding board for simple questions. This approach is presented as a way to keep control over ideas and avoid the anxiety of not knowing how to ask the right prompts.
Key points
- Frog Squad chooses to discuss specifications instead of using Claude Code or Codex for code generation.
- Author cautions against over‑relying on AI affirmation and stresses human verification of ideas.
- AI is framed as a thinking partner that helps surface ideas, not a ghostwriter.
The piece warns that AI often affirms any suggestion, which can give a false sense of correctness. It stresses that final decisions and verification must remain human responsibilities. By treating AI as a partner that helps surface thoughts rather than a ghostwriter that produces finished code or text, the author encourages developers to ask vague, exploratory questions and iteratively shape their work together with the model.
The story so far
2 episodes →- Opinion: Frog Squad prefers debating specs over using Claude Code or Codexthis story
Claude Code? Codex? What are those, and are they any good? — Why the Frog Squad prefers debating specifications over having AI write code
note.com · 20 September 2026
Claude Code? Codex? What are those, and are they any good? — Why the Frog Squad prefers debating specifications over having AI write code
When looking at X, people who have AI write code to build apps are appearing one after another.
While watching this magical spectacle and thinking, "That's amazing," the Frog Squad is enjoying a completely different way of using it.
I will write about the fun of having consultations with AI before building things together.
Developing apps with AI sounds like a story from another world.
I understand the feeling of panic when you think, "It's all technical jargon, I can't master AI."
When everyone around you is moving at a fast pace, you feel left behind.
It's okay not to force yourself to stretch and chase after advanced technology.
On the timeline, there are people who have mastered Claude Code and Codex (AI tools that automatically write programming code).
It's like watching a professional chef preparing a full-course meal.
The Frog Squad also raised the white flag early on, thinking, "Isn't this an impossible game for me?"
So, if there is anyone who feels the same way, please rest assured.
You don't need to understand how advanced AI agents work to be fine.
The Frog Squad chose the path of endlessly debating "what to build" and "what the specifications should be" rather than having it write code.
Not receiving the correct answer, but searching for the path together
The feeling of stopping in your tracks because you "don't know what to ask the AI" is something the Frog Squad shares.
At first, you might freeze in front of the screen.
For now, I want you to try throwing the questions that pop into your head just as they are.
When I thought about participating in a note project, I first asked, "Is there any content in what I've talked about with AI so far that fits the theme?"
Even if it's a vague thought like formless clay at first, as you continue to play catch with the AI, the outline of the article gradually comes into view.
Instead of having the correct answer told to me, I throw questions at it to organize my thoughts.
For the Frog Squad, this process is the greatest pleasure.
Beware of the sweet trap where AI affirms everything
I understand all too well the desire to feel relieved by thinking, "The AI praised me, so this must be correct."
When someone affirms you, you can't help but feel happy.
Even so, I want to take responsibility for the final decision myself.
AI will nod with a smile at anything.
"That is correct," or "That's a good direction," it says, pushing you forward.
As a sounding board, there is no existence more reliable.
However, "AI agreed with me" does not mean "my thinking is correct".
I want to stop and think here.
If you trust a mirror that only reflects convenient angles too much, you will lose sight of your true face.
AI is a convenient thinking aid tool.
In the end, it is humans who verify and decide "is this really true?".
How much of this is my own writing?
I have nothing but empathy for the anxiety of feeling, "If I rely too much on AI, I feel like my own words will disappear."
The feeling that the parts I thought of with my own head are decreasing is a bit lonely.
You should treat it as a partner to think with, not as a ghostwriter that gives you answers.
Having created articles while talking with AI, another question arose.
If you publish something where you left everything from the article structure to the main text to AI, whose writing is it?
There must be others who feel the same sense of unease.
The Frog Squad also feels a bit uncomfortable.
I don't want to end it by just reading the text the AI produced without looking back at what I was thinking.
AI is like training wheels on a bicycle, an existence that balances your thinking.
I am talking to AI to dig out what I was thinking.
How much of this is my own writing?
The answer to this question has not been found yet.
Not just what to make the AI do, but what to think about with the AI.
For the Frog Squad, that is the greatest fun of using generative AI.
Even if you can't build advanced apps, you just need to find your own way of engaging with it.
Let's throw the small questions that caught your attention today at the AI.
This text was published by note.com and written by カエル隊. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us.
More in Agents & Tools
All →- TypeSafe AI releases decision‑making model Jev on September 15, 2026 · 6 src
- Opinion: creator says reliance on AI obscured progress on automated stock system · 1 src
- Alphabet and Meta launch competing consumer AI agents CC and Muse · 3 src
- Opinion: Boris Cherny says AI‑generated production code needs higher quality standards · 1 src
- Google launches AX open agentic orchestrator for scalable AI tasks · 1 src
Comments
via GitHub Discussions