How I AI: Hiten Shah's Advanced ChatGPT Workflows for a Personal Coach, Sales Playbooks, and Managing Up

A surprisingly effective use of ChatGPT is rehearsing conversations with your boss before they happen. In this episode of How I AI, serial founder Hiten Shah demonstrates three context-heavy ChatGPT projects: a rehearsal partner modeled on his boss’s writing, a Personal OS for self-coaching, and a sales guide built from a formal methodology plus product research.

Hiten's core idea is simple: separate durable context from disposable questions. He creates dedicated ChatGPT projects for domains where context compounds over time, giving each its own files, instructions, examples, and memory instead of mixing everything into one giant history.

The demos cover three kinds of context. One project is grounded in another person’s operating principles and writing. Another combines Hiten’s personality frameworks, communication preferences, and voice materials. The third uses a structured sales methodology plus product research. Across all three, the pattern is the same: the model becomes more useful when the source material is specific, opinionated, and persistent.

Rehearse difficult conversations with a "What Would Morgan Do" project

Hiten built a ChatGPT project to practice conversations with his boss, Morgan. The goal was not to create a fake AI clone. It was to generate advice and feedback shaped by Morgan’s operating manual, communication style, and the articles he shares with the team.

That distinction matters throughout the demo. Hiten treats the project as a rehearsal environment, not an oracle. If the model suggests a framing or objection that feels aligned with Morgan’s real behavior, it is useful. If not, he adjusts the instructions or ignores the answer. The value comes from calibration and preparation before high-stakes conversations.

Create a dedicated project instead of a loose chat

Projects give ChatGPT persistent access to uploaded files and custom instructions across conversations. Hiten keeps the Morgan material isolated in its own project so the context stays focused. Earlier in the episode he explains that he uses temporary chats for random prompts to avoid polluting memory with irrelevant information.

Load primary-source material

Hiten uploaded two categories of material from Morgan:

• Morgan’s operating manual: His explanation of how he works, what he expects, and how he communicates.
• Articles Morgan shared: Writing that offered more evidence about his priorities and point of view, including "Job is Communication."

The emphasis on primary-source material is one of the strongest parts of Hiten’s approach. He is not asking ChatGPT to invent a management style from scratch. He is grounding the project in documents Morgan actually wrote, distributed, or endorsed. Even then, Hiten is careful not to overstate what the model can do. A convincing answer is still an approximation.

Use ChatGPT to draft the instructions

Instead of manually writing a long system prompt, Hiten asks ChatGPT to generate the project instructions from the uploaded files. He treats prompting as an iterative process he calls AI shaping. If the model returns something too generic, he nudges it toward the format he wants instead of starting over.

I'm creating a project with these files and I want to be able to converse with it... and simulate the type of feedback or advice that my boss would give me. Can you create the instructions for that project?

One tactic here is worth copying. Hiten tries to codify good outputs whenever he sees them. If ChatGPT produces a useful instruction set, he saves it into the project settings so future chats inherit the same behavior. He even maintains separate projects whose only job is generating better prompts and project instructions.

Test the project with a real pitch

To test the setup, Hiten asked how he should pitch a deliberately ambitious product idea to Morgan. The point was not the product itself. He wanted to see whether the response reflected Morgan’s preferences around evidence, framing, and decision-making.

I wanna pitch Morgan the craziest products idea I can think of. What is the best way to pitch it to him so we can go after it.

The response emphasized concrete vision, evidence, and preparation for analytical follow-up questions. The boss-simulation workflow shows how Hiten grounds that rehearsal in an operating manual and shared writing without pretending the model is the person.

Build a Personal OS for self-coaching

Hiten’s second project turns the same context-heavy setup inward. His Personal OS gives ChatGPT background information about his personality, working style, and communication habits so it can act as a reflection and coaching tool.

The project is less about productivity hacks and more about interpretation. Hiten uses it to think through reactions, prepare for interpersonal situations, and reframe emotional responses. He stresses that the model is suggesting perspectives, not replacing judgment.

Add personality and communication context

The demonstrated Personal OS included three main inputs:

• Enneagram: Hiten identifies as Type 9.
• Human Design: He has a 1/3 profile.
• Voice and tone guide: He added an older personal-branding document.

Combine your context with someone else’s

Hiten then uploaded Morgan’s operating manual into the same project and asked ChatGPT how someone with his own tendencies should work effectively with Morgan.

This manual is from my boss Morgan. I'm newly reporting to him and would love any and all advice on how best to work with him considering my own personality, etc. By the way, he's an Enneagram type five.

The resulting advice blended multiple layers of context. ChatGPT compared Hiten’s Enneagram 9 profile with Morgan’s Enneagram 5 tendencies, referenced Hiten’s Human Design profile, and suggested tactics like leading with data, giving Morgan room to think independently, and structuring updates clearly. What stood out was not mystical personality typing. It was the practical output: meeting formats, communication patterns, and ways to manage up more effectively.

Use the project to reframe emotional reactions

Hiten also demonstrated how he uses the Personal OS during moments of frustration. He gave the project an emotionally charged complaint about someone trying to take over a project he cared about.

I'm really mad right now about someone trying to get control over one of the projects I'm working on because they are trying to steal the cookie.

The response reframed Hiten's frustration as concern about the integrity of the work rather than simple control. The personal-coach workflow uses durable personality and communication context to generate alternate interpretations that Hiten can accept or reject.

Turn a sales framework into an interactive guide

The third workflow shifts from personal context to operational context. Hiten uses public Winning by Design materials to show how a static sales methodology can become an interactive discovery-call assistant inside ChatGPT.

Upload the framework documents

Hiten gathered public Winning by Design PDFs and uploaded them into a dedicated project. His broader point is that frameworks become more useful when the model can reference the original material directly instead of relying on vague prior knowledge. During the demo he referred to the generated output as a SPICE guide, though Winning by Design’s broader framework is SPICED, which also includes Decision.

Generate a discovery guide for a real product

Hiten then asked the project to create a discovery guide for ChatPRD using the uploaded methodology.

Can you create a SPICE discovery guide for ChatPRD?

The output organized questions around Situation, Pain, Impact, and Critical Event. I immediately spotted useful discovery prompts she was not asking in sales conversations, particularly around where workflows break down. Hiten’s point was that the value came from the framework itself. ChatGPT became useful because it had detailed PDFs teaching it how Winning by Design structures sales discovery.

Improve the output with deeper research

The first pass still lacked detailed knowledge about ChatPRD itself, so Hiten ran a separate deep-research query to gather more product-specific context. Then he pasted that research back into the sales project and asked ChatGPT to revise the deliverable using the new information.

Now here is context on ChatPRD... please improve the deliverable above using this new context.

The second pass combined the sales methodology with richer product context. When the project misunderstood the assignment, I pointed it out, Hiten adjusted the prompt, and we reran it in seconds. The interactive sales-playbook workflow shows how to diagnose the missing context instead of treating the first output as final.

The common pattern behind all three projects

Across every example, Hiten pairs a bounded task with durable context. The Morgan project uses operating manuals and shared writing. The Personal OS combines personality frameworks with communication artifacts. The sales project layers a formal methodology on top of product research. He argues that most people focus too quickly on automation instead of understanding the manual prompting loop first.

That manual loop matters because it teaches you which inputs actually change the quality of the output. Hiten constantly compares responses against examples of what good looks like, adjusts instructions, prunes memory, and reruns prompts. He treats prompting less like issuing commands and more like shaping a system over time.

The most useful projects in this episode are the ones where the user already has enough judgment to evaluate the answers. If you know your boss well, you can tell whether the rehearsal sounds right. If you know your sales process, you can spot weak discovery questions. If you understand your own patterns, you can decide whether a reframe is insightful or off-base. What is worth copying is not the specific frameworks or personality systems. It is the habit of grounding AI in real source material, iterating quickly, and keeping humans responsible for the final interpretation.