- Dropbox soared thanks to its convenient interface — despite the fact that there were already plenty of cloud storage programs by then. AI platforms are also becoming plentiful 😉 Is it time to create your own “Dropbox” in the world of AI?
- After all, chat is far from the most convenient interface for interacting with AI platforms. And a “magic button” that, when pressed, AI does what it thinks is necessary — is not the best way to get the result you need.
- AI is already being used in many places. So, you can create many different “Dropboxes” 😉 While using the principle of creating user-friendly interfaces for AI platforms like this:
Project Essence
Zylon has created a platform that helps employees of companies utilize AI in their daily activities.
“We focus on the target audience that AI product developers usually forget,” says one of the startup’s co-founders. “These are people who lack technological skills, which constitute the vast majority of employees in small and medium-sized businesses.”
The first important property of the Zylon platform for companies is data confidentiality.
The AI engine, which sits under the hood of the platform, can be installed locally on the company’s servers. This prevents the transmission of data to AI model developers (LLM, Large Language Model) for training, which could lead to their exposure to other users of the same LLM.
Technologically, this became possible because the Zylon platform uses the open-source PrivateGPT package. The main developers of PrivateGPT are also the founders of today’s Zylon.
The Zylon platform was initially designed as multi-user. This allows employees of the same company to work together with each other and the AI engine on task execution, as well as to see how other employees use AI — to learn from their experience.
Zylon believes that their platform is “for everyone.” In the sense that it is so simple that anyone can use it — including “ordinary” people without special technological knowledge.
The startup was founded in 2023, and their platform is currently in beta mode — but they have already raised $3.2 million in investments.
What’s interesting
The most interesting aspect is the reasons why the founders of Zylon believe that their platform can be used by everyone. They outlined their thoughts on this matter in their blog.
Their main idea is that “using AI shouldn’t look like magic” because magic is unpredictable.
The user presses the “magic button” to summon AI and hopes that somehow AI will do exactly what they need. If it doesn’t work the first time, they press the button again and again or try to perform equally magical dances with prompts, with the same unpredictable result.
AI platforms with “magic buttons” try to guess what the user wants to achieve. With the right AI platform, you can explicitly and immediately specify what you need from it — without lengthy explanations or finding the right prompts.
When the user can request anything from the AI platform in any way, the result can turn out to be anything 😉 This can include something far from what the user actually needs or simply incorrect, as AI engines can hallucinate.
Moreover, the presence of a “magic button” implies that the desired and correct result can somehow be achieved in one step. Although nothing prevents the AI engine from performing the process of obtaining the result step by step, at each step following the user’s explicit instructions — to clearly approach the desired result.
If some initial data is missing, or the AI engine cannot make a reliable conclusion, it can simply inform the user about it. In many cases, this will be much better than if it assumes or invents something — providing a result based on who knows what, but which the user takes as the truth.
Chat is a common interface for user interaction with AI engines. But it’s not the best option because users may not know or understand how to formulate their requests to get the desired answers.
Oddly enough, a traditional interface with input fields, switches, and buttons can be a simpler way to clearly convey to the AI engine what is required and obtain a predictable result.
Moreover, a single input window for anything to solve any task is like a blank sheet of paper that can leave everyone stumped every time they need to write something on it. After all, the concept of the “blank sheet syndrome” didn’t come out of nowhere.
Therefore, the right interface should follow the “Job To Be Done” principle. The user should clearly see where they need to go and what to click on when they need to perform a specific task.
The Zylon interface is designed to minimize the magic in the AI usage process. At the top level, it consists of separate tiles, each designed to perform a specific task.
Inside these tiles, there is no chat, only controls. Therefore, the user can instruct the AI engine on what is required without becoming a “prompt engineer” — and get from it a predictable result that corresponds to the task they are currently performing.
Where to go
The considerations of the Zylon founders reminded me of a recent presentation by Amjad Masad, the founder of the well-known startup Replit — in which he considered the emergence of “Just-in-time UI” (JIT UI, adaptive user interface) inevitable. Here are a few slides from it.
While personal computers were initially used only by techies, they were satisfied with a command-line interface. As the user base expanded to include ordinary people, graphical interfaces like Mac OS and Windows emerged.
However, as the functionality of programs expanded, these graphical user interfaces began to become overly complex.
Just a few months before the appearance of ChatGPT, the idea was expressed that the “most universal user interface is text.”
However, the initial experiments using text as a universal interface in ChatGPT and its counterparts showed that such interfaces also have problems. Writing the right prompts turned out to be a separate skill, which even started to turn into a profession.
Therefore, some AI platforms now feature “adaptive interfaces,” through which platforms ask users additional questions or show additional controls depending on the context of the task the user is trying to solve.
This is what “just-in-time” interface is — an interface that is automatically constructed by the platform at every moment in time based on the task the user is currently solving.
These are still early experiments from which conclusions need to be drawn before moving forward. Do users like such interfaces? Can they get used to them? And most importantly — what opportunities do they hold for startups!
The right interface is crucial. Immediately, the example of Dropbox comes to mind, which suddenly became popular solely because of its convenient user interface, despite the fact that there were already plenty of file storage programs in the cloud by then. And the revolutionary aspect of the first iPhone was not really its large screen — but its convenient interface that could be operated with one finger.
Today’s Zylon is trying to apply the same approach by creating its platform to help small and medium-sized businesses use AI. But the potential here is much greater.
Therefore, a possible direction to take is to create conditional “Dropboxes” for different cases of specific AI usage. With a simpler and more convenient interface that even technically inexperienced people could use, quickly obtaining predictable results without dealing with prompt dances.
All the technology for this already exists. It’s just a matter of choosing the application area and wrapping these technologies in a simple and convenient interface. Which field will you choose for this? What technologies will you use? In what simple and convenient wrapper will you package them?
About the company
Zylon
Website: zylon.ai
Last round: $3.2M, 13.02.2024
Total investments: $3.2M, rounds: 1