It can be easy to think of AI chatbot’s like Anthropic’s Claude and OpenAI’s ChatGPT as merely creators of compelling copy. But, these tools can do so much more. With the right prompts, they can help you identify service gaps where the competition is failing to deliver, or define your brand’s five-year growth projections. They can even power an entire virtual city, which can be utilized for future urban planning in real life. In fact, according to a new study, researchers in China and the US tested whether ChatGPT could create software from start to finish without any prior training. It worked. The chatbot essentially created a virtual software company called ChatDev, staffed entirely by AI bots playing different roles like CEO, programmer, designer, etc.
How They Set Up the Company
After assigning the bots their roles, the researchers prompted them to communicate and collaborate to complete the software development process, from design all the way through release. At each stage, the bots chatted amongst themselves to make decisions, write code, troubleshoot bugs, and finish the software with minimal human input required.
How Long Did It Take and What Was the Investment?
Across 70 different software projects, ChatDev’s AI employees completed development in under 7 minutes at a cost of less than $1 per project, on average. And about 87% of the software they produced worked flawlessly. Yep, flawlessly.
This experiment built on previous research where AI bots powered by ChatGPT’s underlying model ran an entire virtual town themselves. But having the bots develop real software showcases yet another job that can be enhanced by AI automation.
To some, it’s downright scary to see AI performing complex creative and analytical tasks so efficiently. When discussing this with a friend recently, she said, “I feel like a factory worker watching the first assembly line robots roll in, soon to make my job obsolete.”
Personally, I think this overstates a bit. Yes, the AI bots were able to develop the software quickly and efficiently. But, the human inspiration, direction, and oversight is still required.
How They Instructed The Bots
But how exactly did these researchers create an assembly line for software development run solely by bots? First, they broke down software creation into four key stages based on the waterfall model: designing, coding, testing, and documenting.
Next, they assigned different AI agents specific roles aligned to each stage, giving them “vital details” about their designated tasks, communication protocols, and project goals. The head honchos were CEO and CTO bots that handled high-level design decisions. Then coder and designer bots worked on…well, coding and design. You get the gist.
With their roles established, the bots got straight to work chatting through each stage of development as needed for different projects. The CEO would ask the CTO technical questions to guide software design decisions. The CTO would request features from the coder bot, who asked the designer bot for a user interface. Rinse and repeat until every project was finished.
On the simple game project, for example, the CEO kicked things off asking the CTO to recommend a programming language. The CTO suggested Python for its simplicity and readability. The CEO enthusiastically agreed, and off they went with Python as the design foundation.
Through this constant back-and-forth, the AI team iterated from design to release by delegating tasks and collaborating just like human coworkers.
And thanks to capabilities like memory and reflection, the bots could even catch potential code vulnerabilities and debug issues on their own. No IT ticket needed.
Why The Right Prompts Matter
The research shows how far natural language AI has come in not only generating raw content, but performing complex, multi-step workflows. With the right prompts, ChatGPT can apparently coordinate an entire software team now. As has been said many times here, understanding the mind and language of AI can help you solve real business problems and accomplish more than just copy and image generation.
Before we know it, instead of Slacking our coworkers for help, we’ll be Slacking AI bots who complete our work for us.
Why the Bots Won’t Take Your Job
Of course, the researchers acknowledge AI still has major limitations. Errors and biases in the language models can propagate through the development process. So human programmers are still necessary and maintain the upper hand in avoiding messy code.
But as generative AI continues improving, even novice developers could use it for assistance and working alongside such efficient AI can create new opportunities for efficiency and innovation.
So while chatbot project managers aren’t taking over just yet, this research highlights AI’s potential to replicate and enhance complex human-centric workflows in exciting ways.
If you need assistance understanding how to leverage Generative AI in your marketing, advertising, or public relations campaigns, contact us today. In-person and virtual training workshops are available. Or, schedule a session for a comprehensive AI Transformation strategic roadmap to ensure your marketing team utilizes the right GAI tech stack for your needs.Read more: AI Builds Entire Software Company In 7 Minutes for $1
ChatGPT has turned one year old. So, let’s take a look back at the wild ride that has transformed marketing and AI.
AI is transforming the retail space. From hyper-targeted ads to real-time inventory management and more. Here’s how AI is changing retail.
A look at how generative AI is transforming marketing by automating tasks and augmenting skills gaps, so we can free our time for relationship building.