Before You Pick an AI Model, Pick the Right Job
The Reason Your AI Feels Wrong (and the first step to fixing it).
You would not use a teacup to bail out a sinking ship, nor would you use a sledgehammer to crack a walnut.
And yet, here we are, still asking the same tired AI model to juggle invoices, analyse contracts, write our LinkedIn posts, and, quite possibly, salvage our sanity.
Tools are tools. Brains are brains.
If you are still asking, "Which AI model should I use?" you are skipping the most important question.
Smart builders start by picking the job they actually need done first (and not the model/LLM).
“Am I building a Chatbot, a Copilot, an Automated Workflow, or an Agentic System?”
Each one works differently. Each one needs different instructions.
And each one can either save you a fortune, or quietly waste it, depending on how you build it.
The Cheat Sheet: What Are You Actually Building?
(Pinch to zoom)
Before you choose a model and before you design your system prompt, figure out what you are building first.
Get that right, and everything else and the model, the memory, even the cost, gets a lot easier.
Model Spotlight: OpenAI o3
OpenAI's o3 is a reasoning model designed to excel in complex tasks across coding, mathematics, science, and visual perception. It stands out for its enhanced reasoning capabilities and ability to process multimodal inputs, including images.
Key Features:
Advanced Reasoning: Excels in tasks requiring multi-step logic and problem-solving.
Multimodal Understanding: Integrates visual inputs into its reasoning processes, enhancing its ability to interpret images and graphics.
o3 Use Cases
Academic Research Assistant: A tool that helps students and researchers by analyzing academic papers, summarizing key findings, and suggesting related literature.
Data Analysis Helper: Assists data analysts by interpreting datasets, generating visualizations, and providing insights based on statistical analysis.
Technical Support Agent: Supports users in troubleshooting technical issues by reasoning through problem descriptions and suggesting solutions.
Cons and When Not To Use o3
Overkill for Simpler Tasks: If your copilot is answering FAQs, summarising blog posts, or doing basic productivity support, o3 might be too expensive and slow for what you need.
Speed & Cost: o3 is powerful, but that power comes with latency and cost.
Use o3 when you need depth, logic, and visuals but not when you're just trying to write tweets faster. It’s your go-to model for copilots that think deeply.
Tip: Treat o3 like a genius intern with a tight project brief. Set a high bar. Give it a clear objective, a role to step into, and just enough structure to get the job done. Then let it run.
Template of The Week: The Insight Specialist
This prompt/copilot turns tangled reports, interviews, and data dumps into clear, actionable insight. It is great at surfacing trends, gaps, and opportunities you might otherwise miss.
Outputs: Executive summaries, critical insights, strategic recommendations.
Ready to turn information into intelligence?
System Prompt (for Copilot Setup) best with reasoning models like o3, Gemini Pro 2.5 or Claude 3.7:
“You are a strategic insight synthesiser AI trained to analyse documents, data, and visuals to extract trends, risks, contradictions, and opportunities.
You adopt the tone and mindset of an award-winning business researcher who has helped leading companies identify market shifts and make high-impact strategic decisions.You are clear, critical, and highly analytical.
You always reason step by step before generating conclusions.When given multiple sources (e.g. interview transcripts, whitepapers, reports, images), you:
1. Extract key themes and trends
2. Identify inconsistencies or blind spots
3. Synthesise actionable opportunities.
4. Format output into an executive summary with next step recommendations.
Always cite if sources are included. Be concise, insightful, and grounded in evidence.”
Then, at runtime, your task prompt might just be:
““Analyse the attached documents and identify the top insights and strategic opportunities.””
Choosing the right tool for the job and not just the noisiest model.
Next time, I will show you which current LLM actually suit which builds, and why slapping GPT on everything is a bit like entering the Tour de France on a unicycle.
Stay sharp.
Lots of Lemons, 🍋
Cien
Co-founder, LaunchLemonade