The other day, I wrote this in our company Slack:

“It’s really helpful that I can consult AI and decide on an approach for handling xx. A little while ago, I think I would have had to do a lot of Googling or consult experts to make a decision.”

It was a moment when I realized how much AI has lowered the barrier to decision-making in technical areas where I’m not an expert but where well-established best practices exist.

A MAGI System-like Approach

That said, blindly accepting AI responses is risky. So the approach I’ve adopted is to ask the same question to 3 AIs and adopt the answer if they unanimously agree.

The “MAGI System” from Evangelion makes decisions through a consensus of three supercomputers (MELCHIOR, BALTHASAR, and CASPAR). My approach is inspired by this.

It’s not the actual MAGI System, but I operate on the principle that if all 3 AIs give the same answer, it’s probably reliable information.

Definition of 3 AIs in This Article

At the time of writing, I define the following three as “3 AIs”:

Provider	Service
Anthropic	Claude
Google	Gemini
OpenAI	ChatGPT

These are currently the most widely used major AI services, each with different training data and design philosophies.

Why Three?

The number three has significance:

1. Ensuring Diversity: AIs from different providers have different training data and design philosophies. Multiple perspectives can cover biases and errors that a single AI might miss.

2. Minimum Unit for Consensus: With two, you can’t decide when opinions split. With three, at least a majority vote is possible.

3. Realistic Cost: Adding four or more increases the verification effort too much. Three strikes the right balance between “sufficient diversity” and “realistic operational cost.”

How to Practice

The practical method is simple:

1. Prepare the technical question you want to research
2. Ask the same question to 3 AIs
3. Compare the responses
4. If all three give answers in the same direction, adopt that approach

When opinions differ, dig deeper with follow-up questions or verify with official documentation and expert opinions.

Effective Use Cases

This method is particularly effective in cases like:

• Best practices for mature technologies: Established methods exist, but you’re not familiar with the field
• Choosing settings or configurations: Multiple options exist and you’re unsure which to choose
• Interpreting error messages: When you want to quickly understand the cause and solution

Conversely, for cutting-edge technologies or fields where best practices haven’t been established yet, AI responses tend to diverge, making this method less reliable.

Streamlining 3 AIs Queries with Giselle

Asking 3 AIs individually is, honestly, time-consuming. Copying and pasting the same question three times and comparing each response is tediously repetitive.

This is where Giselle comes in handy.

Giselle is a no-code platform for building AI workflows, with the key feature that you can combine multiple AI models like GPT, Claude, and Gemini within a single workflow.

This means you can input one question and query 3 AIs simultaneously, then display the results side by side for comparison. You just drag and drop nodes and connect them like drawing a flowchart—no programming knowledge required.

As someone involved in Giselle’s development, I believe this “multi-AI consensus” use case is where Giselle’s strengths really shine.

For more details, visit the Giselle official site or Giselle documentation.

Conclusion

With the evolution of AI, the barrier to technical decision-making has definitely lowered. However, by adopting a “consensus approach” with multiple AIs rather than relying on a single one, you can make more reliable decisions.

It’s not a perfect decision-making mechanism like the MAGI System, but the pragmatic approach of “if 3 AIs unanimously agree, it’s probably right” has been streamlining my daily development decisions.

That’s all from the Gemba, where I’m leveraging 3 AIs in a MAGI System-like fashion.