You're Asking the Wrong Question About AI. Here's the Right One.

A common question I get when I'm teaching is: "Which AI model should I use for this workflow?" Whether it's for building a marketing campaign, drafting a technical document, or analyzing customer feedback, everyone wants to know the single "best" tool for the job. But this question, while well-intentioned, is the root cause of countless failed automations and frustrating results.

The problem isn't the AI models; it's the question we're asking them to answer. We are trying to assign a massive, complex process to a single tool and hoping for a magical outcome. This approach often leads to models that stall, loop endlessly, or produce low-quality, hallucinatory output.

The key to unlocking the true potential of AI isn't finding one model for your entire workflow. It's about a fundamental mindset shift: you must learn to identify the right model for each specific task within that workflow.

1. Your Workflow Isn't One Job—It's a Dozen "Lego Bricks"

The first step to effective AI automation is to stop thinking in terms of broad workflows and start focusing on their "atomic units": the individual tasks. A workflow, like writing a product requirements document (PRD), isn't one monolithic job. It's a sequence of smaller, distinct tasks—like Lego bricks—that snap together to create the final product. These tasks can include cleaning data, finding context, inferring missing pieces from a pattern, reasoning, transforming a format, checking for correctness, producing an artifact, and handing it off to the next step.

Trying to assign an entire multi-step workflow to a single AI model is like asking one person to be a researcher, writer, editor, and designer simultaneously. It's an inefficient and unreliable way to work. The unit of work is simply too large and too vague. For AI to produce reliable and accurate results, you have to be honest about the individual steps involved.

"...if you want predictability, if you want repeatability, if you want high quality and high consistency, then what you need is to think in terms of the task."

2. The "Best" AI Model Doesn't Exist. You Need a Team of Specialists.

For anyone doing serious work with AI, picking just one model is not a viable strategy. Different models have unique strengths and are better suited for different types of tasks. The most effective approach is to assemble a "team" of specialist AIs and deploy the right one for each specific job.

Consider the workflow of creating a Product Requirements Document (PRD). Instead of asking one model to "write a PRD," we can break it down and assign each task to a specialist:

• Synthesizing customer stories: Gemini 3 is especially effective at synthesizing insights from video, such as recordings of customer interviews.
• Studying the UI: Gemini with nano banana is effective for analyzing a user interface to identify where a new feature might go.
• Aligning with the roadmap: ChatGPT 5.1 in thinking mode or pro mode is well-suited for reasoning about the relationship between a new idea and an existing product roadmap.
• Constructing the final document: Opus 4.5 can take all the previously generated inputs and construct the final, polished PRD.

This "team of specialists" approach isn't limited to general-purpose LLMs. It can also include truly specialized tools built for a single purpose. For instance, a dedicated tool like ChatPRD exists for a reason and can be a powerful addition to your toolkit for certain steps. By leveraging the specialized capabilities of each model for each atomic task, the quality of the final outcome is significantly higher than what any single generalist model could produce on its own.

3. The ROI of AI Isn't Linear—It's Exponential

Behind the question "Which model should I use?" is often a more practical one: "Where should I spend my $20 a month?" While it's tempting to find a single, budget-friendly subscription, this mindset limits your potential. For serious AI work, the return on investment follows an exponential curve.

A single $20/month plan might give you 2x the value for your investment, which feels like a win. However, investing more—perhaps $100 a month across multiple, more powerful models—and using them strategically by picking the right model for each task doesn't just give you a linearly higher return. The speaker emphasizes that when you invest more, your "return on investment is not linearly higher it is exponentially higher."

This exponential return comes from two key factors: higher usage limits and access to better intelligence. Premium plans unlock the full power of the models, allowing you to tackle more complex tasks more effectively. There is also a powerful correlation effect: the people willing to pay more are typically the ones who know how to use AI more effectively, creating a feedback loop that drives massive value.

4. True AI Fluency Comes From "Fingertip Feel," Not Theory

So how do you know which model to pick for which task? The answer isn't found in a manual or a cheat sheet. It comes from hands-on practice and developing what can only be described as a "fingertip feel."

Developing this intuition is a straightforward, practical process:

1. Give multiple models the same real-world task. Don't use theoretical examples; use the actual work you need to get done.
2. Directly compare the outputs. Place them side-by-side and see which model performed better.
3. Be brutally honest in your assessment. Use simple, clear judgment: "this sucks, this doesn't suck, this sucks less."

This process of deliberate practice is non-negotiable for achieving AI fluency. By consistently testing different models against your specific tasks, you will rapidly develop an intuitive sense of which tool is right for which job. There is no shortcut to this expertise.

Conclusion: From A Vague Wish to a Precise Command

The path to mastering AI in your work begins with a simple but profound shift. Stop asking a single AI to fulfill a vague wish ("handle my workflow") and start orchestrating a team of specialist AIs to execute a series of precise commands ("complete these tasks"). By breaking down your processes into their fundamental components and matching each piece with the right tool, you move from frustrating unpredictability to consistent, high-quality results.

Looking at your own work, what's one complex process you can break down into its atomic AI tasks this week?

---

Want to go deeper? Explore the AI Masterclass Series for hands-on training in AI orchestration, or learn how the KDA Decision Engine provides the strategic framework for AI-driven decision-making.