Can you launch a startup without a team? Not really. But you can create a prototype and prepare everything you need to convince people to join your startup. It is much easier to build a team with a website, a financial model, and a prototype rather than just a raw idea. This is how tools like ChatGPT, Midjorney, and website builders like Wix, Tilda or webflow can save you days of work.

To test this hypothesis, a team of venture experts launched an acceleration programme, “My Roboteam”. Anyone with a project idea, even without development skills, could participate. The participants have been working on their projects for a month, from an idea to an MVP. The programme included lectures about each project stage, workshops and homework. As a result, the participants launched seven full-fledged projects.

The project confirmed that launching a startup with only AI tools helps to reduce costs and time for development, hypothesis testing, and project building. And we will reveal the key insights on how to do this. Namely, how to do analytics, build a business model, test hypotheses, create a design and a website with AI, and avoid hallucinations.

Before the beginning

To use AI tools, you need to know the LLM (Large Language Model), its restrictions, and how to use it properly. For example, these tools can give hallucinations and vague results, but this can be corrected with the help of prompts.

The second important thing is to write a step-by-step prompt database dedicated just for startup initial materials creations. From market analysis and ideation to business models and other things. Of course, it’s not enough to have prompt databases. You need to communicate with LLM, adjust the prompts and show some techniques for further communication.

And, of course, even with AI tools you still need to understand at least the basics of methodology. You can ask LLMs to help you with it, but it is always better to have structured information from live humans. It is much easier to communicate and understand. So we had experts and live lectures on various topics with examples of prompt usage.

And lastly, people need space for discussion and communication. They need to receive feedback and discuss ideas, and that's why you need not only chat (messages get easily lost there) but also space with a fixed structure - and that's why we had a sandbox in Trello where everyone had a chance to list idea and discuss it, add useful resources and add their own prompts.

How to do analytics with AI and what can be analysed

With AI help you can search and accumulate summaries of reports, and statistics from databases, and structure information in a given form (for example, in SWOT or Alexander Ostervalders Business Model Canvas). To get data about competitors or trends, use web search plugins like Keymate. Then, the results are combined across different chats to get promising startup ideas and competitive analysis. All this can be of great help at the initial stage when you need to choose a startup idea and formalise it.

How to use AI prompts to make a business model and proto-personas

Before you start building your business model and proto-personas with AI, it's important to flesh out your idea, segment your customers, and clarify what problem you're trying to solve. Both Bard and ChatGPT remember the previous context, so it's best to give them all your data.

Also, you need settings prompts to avoid hallucinations and set a specific role for LLM.  We used that prompt for drawing lean canvas:

“​​Generate a lean canvas for the project above. Use all 9 Lean Canvas blocks: Customer Segment, Problem, Unique Value Proposition, Solution, Channels, Revenue streams, Cost structure, Key Metrics and unfair advantage. Consider the problem. Specify each block. Customer segments should include both demography and behaviour patterns/ jobs. The value proposition should be clear regarding values and show why we are better than the competitors. The solution must include specific features, formats, and technologies. Revenue streams must include average price. Present all the data in tabular form.”

How can AI help with customer development, hypothesis, and testing

First of all, you need to suggest hypotheses to test. If you have, for example, a lean canvas and a problem identified, and you want to formulate a hypothesis for the problem, just ask ChatGPT to list 3-5 hypotheses to test. Same with lean canvas: it is better to provide more specific information in the prompt, like “make 3-5 hypotheses from the lean canvas problem block above.”

To avoid getting vague results, specify the form's expected result and ask it to act as a researcher. Once you have these hypotheses, you must go outside and talk to real people who correspond to your proto-characters.

Next, you need to transcribe your interview. This can be done with tools like otter.ai or descript.com. Then upload the transcription and ask ChatGPT to analyse it to see if your hypotheses are validated. You can also ask ChatGPT to summarise the interview and list new ideas.

How to write code, create UI/ UI design, and a prototype with AI tools

That's the tricky part. No tool can give you a complex solution for a prototype. The product quality that AI can create is acceptable for bloggers or a simple online store, but not for a startup.

But again, you can delegate some of the work to AI tools. For example, ask ChatGPT/Bard to list the main features and functionality that should be included in a prototype or even an MVP. Then ask it to create userflow /site map/ CJM. Then you will most likely have to expand it and adjust it and redraw it, for example in Miro or a similar tool. But the AI will do the initial work for you.

You can ask AI to create a UI style, colours and even descriptions of logos and icons. The modern version of GPT with built-in DALL-E can even draw something (but I still recommend using Midjourney for these tasks).

You can ask the LLM what UI elements you need and then draw them in Midjourney. But then you will need to combine them somewhere. The best tools for this are Wix/Tilda/Webflow. If you need a backend for your prototype, you can ask ChatGPT to help you with tools like Integramat and Airtables. Not the easiest way, but still faster than learning to develop and design, and if you don’t have a designer and developer on your team or the money to hire them.

How to avoid AI restrictions and hallucinations

LLM has some restrictions, sometimes a little awkward, but you can overcome it with the help of prompts. For example, I made "Alien Files" for one of the festivals about aliens living on Earth, with text, pictures, investigative reports, etc. Of course, writing all this funny nonsense would take me many hours, so I used ChatGPT. But when I asked it to give me materials about reptilians, ChatGPT said that it could not produce them because “it is a humiliating conspiracy theory”. So, I had to talk to him and after several hints, looking at this topic from different points of view, I got the required result.

Another example: I recently asked ChatGPT to provide application statistics from my database. ChatGPT could not match the background information with the idea description for one task because it contained “personal information.” Again, you just need to talk to GPT and explain that you uploaded this database and have all permissions.

Hallucinations are usually easier to deal with: you can avoid them with a settings prompt at the beginning of the chat.

So AI tools can be beneficial in launching a startup. Of course, they can’t do all of a team's jobs, but they work like junior employees. They can’t perform complex tasks, write code for a high-load backend system or create a unique design, but they can help you with simple things you need to start with.