Turn ChatGPT into a Sushi Conveyor Belt: 5 Prompts to Automate Workflows that (Seriously) Save Hours

You've heard me talk about the prompt as the product, now let's dive deep on workflows—how can you save hours a week with a five minute prompt? This post has 5 examples and teaches you how!

I keep realizing that we are under-estimating what these models can do, and that includes me!

As just one little example, I ran across a tweet from Ethan Mollick yesterday that called out how o3 can one-shot an entire business idea from scratch:

That’s just one small example. One of the most powerful things about these models is their ability to let you do a lot of decent quality work very very quickly.

Before, I had to stare at a blank piece of paper to think of marketing copy. Now I can throw a dirty draft or vibes into ChatGPT and get back a bunch of ideas in no time at all.

And that’s been true for awhile. So what’s changed now?

Ethan’s tweet is a good example of how models are fundamentally moving from chat » reasoning » workflow.

A little example will help illustrate how far we’ve come here.

Beforetimes—raw human: Nate stares at wall, goes for walk, spends all afternoon banging head on desk and staring at hundreds of customer anecdotes to product a Product Requirements Document (PRD) draft that is kind of ok. Total time = 6ish hours.

ChatGPT 2023—just chat: Nate gets 15 bad ideas for his PRD but they’re kind of ok and they get him off the blank page and he moves about an hour or so faster. Feels magical. Total time = 5ish hours.

ChatGPT December 2024—chat + reasoning: Nate has an actual thoughtful conversation with a reasoning model about his PRD. Develops a sharper angle on his product thesis. The document is tangibly better and faster. Total time = 4ish hours.

ChatGPT April 2025—chat+reasoning+workflow: Nate throws hundreds of anecdotes into o3, one-shots an entire quantitative and qualitative analysis, takes 10 minutes to develop a grounded thesis, fights about an outline for 10 minutes, and then rips through a deeply customer-grounded draft in half an hour. Then takes a final 5 minutes to schedule an update on customer anecdotes weekly. Total time = ~1 hour.

TLDR; I’ve compressed work on this task 6x in 3 years, and the biggest single jump was from December 2024 to April 2025. Yes, you read that right.

And that’s the breakthrough I want to lay out for you today. I wrote an article earlier this week on the idea of prompt-as-product, where we have to update our assumptions about prompting for these new models. This article builds on that concept and goes much further, focusing specifically on what I think is the most powerful aspect of o3 in particular: agentic workflows.