Here's what I learned from pitting Claude and Gemini against each other while building my first bot

While building a crypto alert bot with no prior coding background, I developed a habit that turned out to be genuinely useful: copying the same question into both Claude and Gemini and comparing the answers.

My journal at the time: "Pasting the same questions into Gemini and Claude and watching them disagree. This is more fun than I expected."

Here's what I learned from that.

They diverge on honesty more than capability.

When I asked how long Make.com setup would take for a beginner:

• Gemini: "20 minutes"
• Claude: "That's quite optimistic. Expect 1–2 hours. Also, the free plan has a 1,000 ops/month limit — worth knowing upfront."

Tried it myself. Claude was right.

On code review:

• Claude: "6.5/10. It works, but this code fails silently." (followed by a concrete bug list)
• Gemini: "100 out of 100... but here are 5 things to fix."

Gemini's praise felt good the first time. After the third time, I stopped trusting it for critical feedback.

Ask a tool about itself.

Had a Gemini API error. Asked Claude — didn't work. Asked Gemini directly — solved immediately.

Obvious in hindsight. Worth noting anyway.

The moment that reframed the whole project.

Asked Claude: "What if I charge people for bot signals?"

Claude: "Companies with billions in funding already do this. And selling investment signals for money can be a legal issue. Honestly, the realistic angle is selling the process — not the signals."

Didn't want to hear it. Wasn't wrong. That answer pushed me toward documenting the build instead of monetizing the signals — which is what I'm doing now.

Journal after that: "There's this AI called Claude. More realistic than Gemini. Ended up buying it."

Takeaway for builders using AI tools:

It's not about which one is better overall. It's about knowing which to reach for when.

• Claude: uncomfortable truths, accurate criticism, realistic scoping
• Gemini: fast drafts, knows its own ecosystem best

Running both in parallel showed me things I couldn't see using either alone. The disagreements were often more informative than the answers.

This is part of a series I'm writing about building 6 bots from scratch with no prior coding background — what worked, what didn't, and what I'd do differently.

What's your setup for getting honest feedback from AI tools?