How I use ChatGPT as a Solopreneur
Plus a Wanderly rebranding update
I started Wanderly in February when I turned a hobby project into a software business using my personal savings and a lot of elbow grease.
To be honest, I've gotten so much farther than I ever thought possible. But I wouldn't have been able to get where I am without ChatGPT. So today I thought it might be interesting to share how I’m using ChatGPT to enable solopreneurship.
Note: I only ever use ChatGPT with the GPT-4 model, which requires a ChatGPT Plus subscription of $20/mo. For me, it's totally worth it.
There are two major things ChatGPT helps me with: coding and marketing. With coding, it’s magical: I ask ChatGPT to write snippets of code, write scripts, find bugs, and navigate productionization. With marketing, ChatGPT can generate a lot of ideas, but I have to rely on my own judgment to make something good happen.
ChatGPT can help with marketing but has limitations
I’ve used ChatGPT as part of my workflow for all parts of Wanderly marketing: website copy, ad copy, creating a tagline, and rebranding Wanderly. A couple of things to note:
It has access to all culture, which is helpful when asking for inspiration
It has its own copy style; by default, it uses lots of adverbs and a large vocabulary, and most naming exercises usually sound like bad startup names.
By default, it usually assumes that you’re a large company with lots of resources.
It never runs out of new ideas and is always willing to engage with all of your ideas.
A typical brainstorming session usually goes like this (sample prompts in footnotes1):
I describe my vision for Wanderly, in as much detail as I can.
I frame the marketing task I’d like to do.
Say “Before you start brainstorming, ask me some questions so I can provide more context”.
Answer the questions, and then tell ChatGPT to start brainstorming.
The thing about using ChatGPT to help with creative tasks is that it always defaults to the average advice… but marketing isn’t about finding things that are average. Marketing is about finding a unique spark and cultivating it. But the average advice is a lot of good advice. So I’ve used ChatGPT to remind me of table stakes and to get inspiration, e.g., think of SEO, referrals, and synonyms. Ultimately, I have to be the one shining the light on valuable things. My final marketing deliverable is usually something inspired by my conversation with ChatGPT (never verbatim), but I got there much faster due to its help.
ChatGPT is like having an experienced, patient pair programmer on my engineering team
For the readers who aren’t familiar with software engineering, there’s a practice called “pair programming” where two engineers work together on the same code at the same time. It’s often used to help engineers come up to speed or when working on a particularly gnarly problem.
With ChatGPT, I have an experienced pair programmer constantly at my side: it writes code instantly, knows many industry tools and libraries, proactively suggests improvements, and it’s good… most of the time. Here’s my basic workflow (sample prompts in the footnotes2):
I describe my tech stack, to give the AI relevant context, e.g., I’m writing a React app.
I give it a large chunk of my code (I’m talking up to hundreds of lines of code and/or an entire stack trace)
I describe the change I want to make or the problem I want to solve
I ask for it to return example code
I copy-paste the code, run it, and then ask for improvements
Always say thank you… just in case ;)
It’s pretty wild. One of the first things I noticed is that coding with ChatGPT almost completely erased my imposter syndrome. One of the reasons I became a PM instead of a software engineer in my early career is that when I found a bug, I often felt like it was *me* (instead of thinking that it was the difficulty of the problem I was trying to solve). Now I can power through solving problems; my only barrier is willpower and patience.
It’s not all 🌈s and 🦄s yet. I have an incredibly tough time using ChatGPT to help me with development environments and production deployment challenges. It’s still far better than trying to figure it out on my own, but ChatGPT can send me on wild goose chases. When I find things getting too weird or difficult, I start from scratch and initiate a new conversation with ChatGPT; if that fails, I’ll go to help forums like GitHub and StackOverflow. I’m not entirely sure what makes this type of problem so much more difficult than my other coding-related requests, but I hear many professional engineers bemoan this aspect of software development, too. Some things still require a decent amount of human-level intelligence.
As I’m writing this, I have two feelings: 1) I know how incredibly quickly this post will age, and 2) I can imagine the future of coding with AI. My prediction: in 5 years, I’ll be able to build an entire product mostly conversing with AI. I’ll be able to feed my entire codebase into an LLM as context, and then ask it questions and change code. But also: I’ll be able to come up with a product idea, develop a business plan, brainstorm wireframes and high-fidelity user interfaces, and develop code… all by just talking to an LLM the same way I’d talk to a team at work.
In order to make the best products that surprise and delight, I still think there will need to be a human with a vision piloting the output of these AI tools, but maybe that’s just wishful thinking. There have been whispered rumors of OpenAI having achieved Artificial General Intelligence (AGI), and maybe an AGI in five years will be able to build products far better than I can. But until then, I’m going to keep trying to build and manifest my vision for what I believe should exist in the world, and hopefully, I can use these tools to accelerate it.
I still miss my team 🥹
So, does all of this talk of ChatGPT and solopreneurship make me feel like I don’t need a team: Absolutely not. I really miss having a team. Thankfully, my husband is a great sounding board and support. But it’s not quite the same. A team can shine light in blindspots and provide the experience and taste AI can’t offer yet. And nothing beats the camaraderie of a group of people all working 100% towards the same thing.
I hope I’ll be in a position to hire a team soon, but until then, ChatGPT is allowing me to de-risk Wanderly while keeping my burn rate extra low.
So how is Wanderly doing?
About 6 weeks ago, I rebranded Wanderly. It was hard and I was really nervous, but it’s gone so much better than expected. My CAC dropped 25% on social ads, I’ve gotten a lot more questions about the product, and the comments are much nicer. Wanderly just crossed over 1K stories told, with an average star rating of 4.5. I launched a new story type: Help an Animal.
I’m learning quickly, but there’s still lots of work to do. I’m experimenting with coupons, some social media promotions, and marketing site improvements, and I will also be working on improving the first-story experience.
Info Diet
Some hopium for the soloprenuer: Being patient and disciplined can pay off.
OpenAI is on 🔥:
They announced DALLE-3 in ChatGPT (demo video + Midjourney v6 vs. DALLE-3 comparison)
TBH, these announcements are 50% intimidating and 50% exciting because I’ll be able to use these APIs. I keep telling myself that it’s one thing to do a demo video of a one-off children’s story, but it’s another thing to productionize and refine AI-powered children’s stories that people want to return to regularly, and that these demos are functionally an advertisement for Wanderly’s use case… but I only convince myself half the time. 🙂
~20% of Americans use ChatGPT, and for many, it’s not part of their core workflow
Amazon to invest up to $4B in Anthropic, tying these two companies together similarly to how Microsoft and OpenAI are tied. The open question on everyone’s mind is: What happened to Google’s $300M investment in Anthropic?
Another good Hard Fork podcast about Bard Extensions and facial recognition.
Feeling like humans weren’t made for the endless grind of capitalism? Watch this 30-minute video on the origins of the 9-5, how medieval workers used to work, and how the start of the workday was shaped by a bargain between a businessman and a church. 👀
Marketing brainstorming prompts:
General prompt: You are a world-class brand marketer who specializes in children’s products. I’m building a [LONG_DESCRIPTION]. I’m looking to write [COPY_TYPE]. Please ask me 10 or more questions before you start suggesting [COPY_TYPE].
Follow-up prompt 1: [ANSWER_QUESTIONS]
Follow-up prompt 2: I really liked where you were going with [something it suggested]. Can you come up with 5 more iterations on that?
To brainstorm naming or taglines:
Similar to the general prompt, and I’ll also say what feelings I want the name to evoke. Then I’ll just wander (pun intended) through related words and phrases, prompting for synonyms, related quotes, famous characters that demonstrate certain behaviors, etc. until something jumps out at me.
To brainstorm ad copy:
Similar to the general prompt, but I’ll iterate through different marketing angles. E.g. Now give me 5 versions of the ad copy that evoke [awe | guilt | educational value | anger | delight | etc.].
Coding prompts:
To debug code: Here is my code: [CODE_SNIPPET]. The code isn’t doing what I want. I’m expecting it to ______, but instead it’s doing _____. I’m getting this error: [STACKTRACE]. I’ve already tried _____. Can you find the bug and/or give me steps to debug to get more information?
To build something new: I’m building a [describe what I’m working on and what language it’s written in]. I’d like to build out some new functionality: _____. Can you describe the best approach to build this, or if there are multiple options, can you describe the tradeoffs to consider for each approach?
Follow up-prompt: I really liked approach #X. Can you tell me step-by-step how to implement this, including code samples if needed?
To create regular expressions: I’d like to create a regular expression in my [Python | Javascript, etc] function that parses out specific sections of a string. For instance, if the string was [STRING], then I’d want the final output to be [DESIRED_OUTPUT]. Here’s another example: [ANOTHER_EXAMPLE]. Can you write the code and regular expression for me?