Listen to this Post
Introduction:
In the rapidly evolving landscape of artificial intelligence, the gap between average AI output and exceptional performance is rarely determined by the tool itself but by the operator wielding it. The current AI discourse is heavily skewed towards the technology’s potential, yet a vast majority of users treat advanced language models like simple note-taking applications, leading to mediocre results and the false conclusion that the technology is overhyped. While you can spend thousands on subscriptions, the true leverage lies in understanding and activating the underutilized features hiding in the sidebar, transforming ChatGPT from a simple chatbot into a powerhouse for research, productivity, and complex problem-solving.
Learning Objectives:
- Master the selection of different AI models (Speed, Reasoning, Deep Research) for specific tasks to optimize efficiency and output quality.
- Learn the five-part prompt formula to move from vague requests to precise, structured, and actionable instructions.
- Understand how to utilize advanced features like Custom Instructions, Memory, Projects, and Agentic Workflows to create a personalized and automated AI ecosystem.
- Implement critique prompting as a high-leverage technique to automatically refine and improve AI-generated content.
- Develop an AI-first workflow that mirrors the strategies of successful founders to achieve a 10x productivity boost.
You Should Know:
- Model Selection: The Right Tool for the Right Job
The most common mistake is using a single model for every task. ChatGPT offers distinct models for a reason. “Instant” (often the smaller, older models like GPT-3.5) is built for speed and low-cost, ideal for simple queries, brainstorming, or tasks requiring minimal latency. “Thinking” (often associated with the O1 series) is designed for complex reasoning, step-by-step problem-solving, and tasks like coding, data analysis, and mathematical logic. “Pro” (often GPT-4 with DALL-E or Deep Research capabilities) is for high-stakes, deep-dive research tasks requiring a large context window and multi-step synthesis.
Step-by-step guide explaining how to use it:
- Assess the Task: Before typing anything, ask yourself: “Is this a simple idea (Speed), a complex logic problem (Thinking), or a research-heavy deep dive (Pro)?”
- Switch Models: In the ChatGPT interface, manually select the model corresponding to your task before entering your prompt. Do not default to the top option.
- Match Model to Output Needs: For a quick email, use “Speed.” For a detailed article, use “Thinking.” For analyzing a 100-page PDF, use “Pro.”
- When Multiple Models Are Needed: Start with “Thinking” to outline a strategy, use “Pro” to gather data, and “Speed” to format the final report.
Practical Application: Task: Write a Python script to automate data entry in a CSV file and provide a summary of the duplicates. - Model: Use the "Thinking" model. - Command in "You are a senior Python developer. Write a Python script that uses the pandas library to read a CSV, detect duplicate rows based on the 'Email' column, remove them, and output a new CSV file. Also, write a summary of the number of duplicates found." - Command to Refine: If the code doesn't work, you might switch to "Speed" to ask, "Generate a command-line instruction to run this script using 'python script.py'."
2. The Five-Part Prompt Formula
A vague prompt yields a vague answer. The structure of your prompt is the single most critical factor in output quality. The formula `Role, Task, Context, Constraints, Output Format` is the skeleton for all high-quality AI interactions. Most users only define the “Task,” leading to average outputs that require extensive rewriting. By incorporating all five elements, you preemptively direct the AI’s focus, style, and reasoning process.
Step-by-step guide explaining what this does and how to use it:
1. Role: Define who the AI is. “You are a strategic business analyst…”
2. Task: State exactly what you need. “…tasked with creating a market entry strategy for a SaaS product…”
3. Context: Provide the background. “…for a startup in the EdTech sector with limited marketing budget but strong organic growth…”
4. Constraints: Set boundaries. “…do not recommend paid advertising. Assume the product is B2C. Keep the strategy under 500 words.”
5. Output Format: Tell the AI how to present it. “…Format the output as a bulleted list in a table with sections on ‘Risks’, ‘Opportunities’, and ‘Action Items’.”
Practical Application: Instead of typing: "Write an email to a client about a project delay." Use: Role: You are a professional project manager known for transparency and proactive solutions. Task: Draft an email to our client, 'Client X', explaining a 2-week delay in the 'Project Alpha' launch. Context: The delay is due to a third-party vendor failing to deliver the API integration. We have already secured an alternative vendor that will have the integration ready by next week. Constraints: The tone must be professional and reassuring, and avoid technical jargon. Output Format: Provide the email in a ready-to-send format with a subject line, a formal greeting, a body that explains the delay and the solution, and a closing with a request for a meeting.
3. Custom Instructions and Memory: Personalizing the AI
For the AI to be truly effective, it must know your goals, style, and context. Custom Instructions are the place to set your permanent context. This includes who you are, what you do, your preferred tone (formal, casual, technical), and what you want the AI to know about you. Memory is the dynamic record of your interactions; when turned on, it allows the AI to “remember” facts you’ve mentioned across sessions, eliminating the need to repeat yourself.
Step-by-step guide explaining what this does and how to use it:
1. Set Up Custom Instructions: Go to Settings > Personalization > Custom Instructions. Fill out the two sections: “What would you like ChatGPT to know about you to provide better responses?” and “How would you like ChatGPT to respond?” (e.g., “I am a marketing director for a B2B software company. My preferred tone is concise, data-driven, and direct. I dislike fluff.”).
2. Enable Memory: In the same Personalization menu, ensure Memory is toggled “On.”
3. Manage Memory: The AI will automatically remember things like “User prefers Python over R” or “User is working on a project called ‘Project X’.” You can view and delete specific memories in the Settings menu or by asking the AI, “What do you remember about me?”
4. Use Memory in Prompts: You no longer need to restate your role. The AI will automatically apply your Custom Instructions and Memory to every response. Simply say, “Based on our previous conversation, draft a report for Project X.”
4. Projects and Agentic Workflows: Organizing and Automating
Projects are containers where you keep files, instructions, and chat threads all in one place. Instead of searching through 40 disjointed threads, you can organize your work by project (e.g., “Marketing Campaign 2026”). Agentic Workflows represent the next level of automation, allowing the model to plan and execute tasks autonomously. This means the AI can analyze data, research topics, write content, and even critique its own work without needing constant user input.
Step-by-step guide explaining what this does and how to use it:
1. Creating a Project: In the ChatGPT sidebar, click on “Projects.” Click “Create.” Name it and set project instructions (the overarching goal of this project).
2. Adding Files: Upload relevant files (CSVs, PDFs, Word documents) to the project. The AI will have them as context for all chats within the project.
3. Integrating Workflows: Start a chat inside the project. Use a prompt to trigger an agentic workflow: “I need to write a 2,000-word whitepaper on the future of AI in healthcare. Use the documents I have uploaded. Research current trends, find recent statistics (with sources), and create an outline. Then, draft the paper. Do not proceed to drafting until I approve the outline.”
4. Autonomous Execution: The AI will now plan the steps, perform research, synthesize the data, and draft the outline. You approve it, and it will proceed to write the full paper, utilizing the files and instructions stored in the project.
Practical Application (For Market Research):
Steps for an Agentic Market Research Workflow
1. Create Project: “Competitor Analysis”
- Upload Files: Upload 5 PDFs of competitor annual reports.
- Set Instruction: “Act as a senior strategy analyst. Your goal is to identify market opportunities.”
- Prompt (Agentic): “Plan a strategy to identify the top 3 threats and top 3 opportunities for us based on these reports. Do the research, synthesize the data, and present a SWOT analysis as a final output. Start the research now.”
The AI will then organize its plan, execute the steps, and return a comprehensive final report without you providing additional input.
5. Deep Research and Critique Prompting
Deep Research is the feature that enables multi-step research with synthesized outputs. It is designed for tasks where you need to go beyond a simple search and require the AI to analyze patterns, compare data, and synthesize complex reports. Critique Prompting is a high-leverage technique where you ask the model to review its own output before final delivery. By integrating this into your workflow, you create a self-correcting loop.
Step-by-step guide explaining what this does and how to use it:
1. Activating Deep Research: In the chat, ensure you have selected the “Pro” or equivalent model that has Deep Research capabilities. Then, ask a complex research question. Example: “Analyze the 2026 trends in cybersecurity for mid-sized financial institutions in the EU.”
2. Specify Depth: Ask for a synthesized output: “Produce a comprehensive report that synthesizes findings from at least 15 different sources, identifies patterns, and predicts future threats.”
3. Critique Prompting: After the AI provides the output, don’t accept it immediately. Instead, add a new prompt: “Critique this report. Identify potential weaknesses, missing data points, logical fallacies, or areas where the argument could be stronger. Then, produce a revised version.”
4. Iterate: If the first critique is minor, ask for a second pass: “Provide a more aggressive critique. What assumptions are we making that could be challenged?”
5. Final Delivery: Once the AI has addressed its own critique, accept the final, now-highly-refined output.
Practical Application (Linux/Windows Commands for AI Research):
To use ChatGPT for manual research, you can use command-line tools to simulate an agentic workflow locally.
Linux Command: Using wget to download a research paper.
wget https://www.example.com/research_paper.pdf
Windows Command: Using curl to fetch API data for research.
curl -X GET "https://api.example.com/data/latest" -H "Authorization: Bearer YOUR_TOKEN"
Python Script: A simple script to summarize text using AI (requires OpenAI API key).
import openai
openai.api_key = "YOUR_API_KEY"
response = openai.ChatCompletion.create(
model="gpt-4",
messages=[{"role": "user", "content": "Summarize this financial data: [insert data here]"}]
)
print(response.choices[bash].message.content)
What Adam Biddlecombe Say:
- Key Takeaway 1: The tool isn’t overhyped; the operator is underperforming. The “average” output is a result of average input, not a limitation of the AI model itself.
- Key Takeaway 2: There is a massive gap in leveraging advanced features. Most users completely ignore “Projects,” “Agentic Workflows,” and “Memory,” paying for features they never use.
Analysis:
Biddlecombe’s argument is a classic case of “skill gap” versus “technology gap.” The entire ecosystem of AI has shifted from “Can the AI do this?” to “Can the user extract this from the AI?” He highlights that the difference between a founder wasting $500 on AI and a founder getting 10x ROI is simply workflow optimization. By advocating for structured prompting, model selection, and critique loops, he reframes AI as an extension of user intent rather than a replacement for it. The lack of adoption of “Memory” and “Projects” suggests a fundamental misunderstanding of AI as a collaborative tool versus a simple Q&A database. His point about “Critique Prompting” is particularly salient; it mimics the process of a manager reviewing a subordinate’s work, effectively making the user a “manager of AI” rather than a “user of AI.” This shift in mindset is what separates the “10% users” from the power users.
Prediction:
+1 As enterprise adoption of Generative AI continues to grow, tools will increasingly be judged not by their raw intelligence but by their interface and workflow integration capabilities. The market will reward platforms that make feature discovery and utilization seamless.
+1 Custom Instructions and Memory will evolve into a core “AI persona” that becomes a unique asset, similar to a professional reputation, transferable across different AI platforms. Users will treat their AI “configuration” as a valuable intellectual property.
-1 The proliferation of advanced features without proper education will lead to a widening productivity gap, where knowledge workers with AI literacy will significantly outpace those using AI on a surface level, exacerbating inequality in the tech sector.
-1 As agents and deep research become more autonomous, the risk of “hallucination chains” — where the AI builds a flawed premise upon another flawed premise without human oversight — will increase, requiring new auditing techniques.
+1 The ability to build and sell AI-powered businesses, like Mindstream, will become a viable startup path not just for AI engineers but for prompt engineers and domain experts who can build effective workflows.
+N The democratization of high-quality output is a double-edged sword; while it enables SMEs to compete with large corporations, it also floods the market with AI-generated content, making true, original human insight more valuable than ever.
▶️ Related Video (88% Match):
🎯Let’s Practice For Free:
🎓 Live Courses & Certifications:
Join Undercode Academy for Verified Certifications
🚀 Request a Custom Project:
Secure, high-velocity infrastructure and disruptive technological engineering. Contact our engineering team for high-tier development and proprietary systems:
[email protected]
💎 Smart Architecture | 🛡️ Secure by Design | ⭐ Trusted by Thousands
IT/Security Reporter URL:
Reported By: Adam Biddlecombe – Hackers Feeds
Extra Hub: Undercode MoN
Basic Verification: Pass ✅
