AI Solopreneurship - Part. 1

Experiments in using multiple coordinating AI agents to conceive, design, build, test and market a new business venture

It's been a long time since I really wrote code. While I loved the intellectual process and challenge, I quickly realized it wasn't my calling. However, that experience enabled me to figure out what I do love, which is conceptualizing a vision for a new product or refining an existing one. Collaborating with a team to deliver on that vision was a significant part of my life for ten years at 8ninths, the R&D lab I co-created with William Lai. We had a supremely talented team, masters of product management, design, and development, who built innovative new products for some of the world's biggest companies, such as the Holographic Workstation for Citi.

Five years after our acquisition, I still yearn to re-engage with the entrepreneurial world, and this post outlines how I'm approaching things differently. Not long after ChatGPT launched, buzz emerged around a project named “AutoGPT,“ aimed at evolving GPT-4 into a fully autonomous entity capable of tackling complex tasks with minimal to no human intervention. Similar projects, like Microsoft Research’s AutoGen and ChatDev, started surfacing, focusing on creating cooperative autonomous AI agents. I've been dedicating some time to experimenting with these new tools, and I thought it might be useful to share some of my findings and thoughts.

My Goal

The hype around what AI can and cannot do is abundant. We are in the very early stages of the hype cycle, a journey that I think will likely continue indefinitely.

I sought to answer the question: Can I really come up with an idea and work with AI agents to design, develop, and test a business concept fully on my own?

Inspired by the way many people, including myself, doom-scroll through content on platforms like Reddit, I conceived an idea. Although I don't read many books these days, I voraciously consume podcasts, audiobooks, and videos. Would a “reddit-esque“ short-form format facilitate individuals like me to engage with longer-form content like a book?

That's the premise of “Snackify.“ I planned to use this relatively simple concept to see if I could collaborate with AI to design, build, and test it.

This project is a work in progress. I've encountered some astounding results and surprising pitfalls. I chose to work openly on this venture as I truly believe the evolutions of this endeavor will significantly alter the nature of work in the years to come.

ChatDev

Initially, I delved into ChatDev. Imagine wanting to create software without a team. ChatDev functions like a virtual development company where instead of humans, computer programs called agents perform the tasks. There’s an agent acting as the CEO, a programmer agent who writes the code, a tester agent who checks for mistakes, and so on.

You inform ChatDev about the kind of software you desire, and these agents collaborate, much like how people would in a real company, to build an application. They can design, write code, test it, and rectify any issues, all while working together seamlessly. You can even modify how these agents work to better suit your idea, like instructing the designer agent to focus on bright colors or the programmer to enhance mobile-friendliness.

At the core of ChatDev is a user-specified LLM, in our case “ChatGPT,“ which powers each coordinating agent (you just provide your API key).

You assign a task to the system, it operates in the background, and then you can “replay“ the results through a cute 8-bit style UI, providing context around each agent's actions. The code, documentation and imagery is all created in a directory ready for you to modify or continue to iterate upon.

In my example, I started out with a simple task specified as:

"Create a product that will turn a book into reddit style posts, each paragraph will be a separate post that can be viewed, commented upon, saved, shared and voted upon. Every 10 posts create an AI generated image based upon the posts on the page. The application should be targetted at the web and mobile form factors”

The system ran for about 10 minutes, engaging in design, code development, testing, art generation, and iteration. You can witness a replay of the logs/work in progress in the short video snippet below, where each participating agent is highlighted with a thought bubble as they tackle their section of the problem.

The resulting first iteration of the app, created by my AI companions, is showcased below. Alice in Wonderland, turned into paragraphs. Th Although it's not yet at a stage where I can relax and start collecting ad revenue from the Bahamas, it's a solid start. The architecture, crafted in Python, is structured well enough for continuous evolution and iteration.

So why should I care?

Autonomous AI agents working collaboratively are central to discussions about the future of work. To understand its evolution, consider using tools like Chatdev—a platform for role-playing team compositions. This exercise can help you envision how to integrate autonomous agents with your human team and determine the most effective areas for AI deployment in your organization.

Will professionals like lawyers, doctors, software engineers, product managers, and designers be replaced by relentless, innovative, creative AI agents who can surpass the best in their respective fields?

In my view, this scenario is a “maybe. . .“ depending on the timeframe under discussion. We're definitely not there yet, but a decade from now? We are likely at least to witness a surge in the utilization of specialized agents for tasks, accompanied by a drastically reduced human workforce.

In the near term, I envision a middle ground emerging during these transitional years - a hyper-personalized AI agent that learns from our work ethic, styles itself after us, and represents us or acts as a liaison with other humans and agents toward achieving a goal. This might redefine the role of the information worker, where nurturing, guiding, correcting, and assisting an agent who represents us becomes the norm. We might find ourselves working at multiple companies simultaneously, surprised by some contributions “we“ (whatever ‘we’ is) make to solve a problem, learning about it only during a weekly update from our agent and being rewarded (through the blockchain of course…) for our fractional contributions.

What remains certain is that change is the only constant. I prefer to harbor an optimistic outlook that, akin to every other technological transition humankind has navigated, we will adapt and hopefully thrive.

I hope to continue to share progress on my experiments in AI Entrepreneurship in future posts. Your thoughts on this are highly encouraged, and I would appreciate it if you would share your insights or experiences in the comments below.

Stay ahead and let Sheppard Consulting help navigate your business through the new technology era. Let’s Talk.