We're here to help

AI agents solve problems by taking an input (prompt, request or query) and taking action on them. These actions can be in parallel to you, which is a copilot, or independently and autonomously for you, which is an autopilot. For example, if the input prompt is "help me write a cover letter for this job application", a copilot agent will generate the first copy that you can then refine and send. From there an autopilot agent would write and send the cover letter on your behalf. There is a considerable amount of work happening behind the scenes for AI agents to work. AI agents are able to tackle much more complex problems and tasks vs. their assistant predecessors because of Large Language Models. Training these AI agents requires much more compute power than we've previously had until now.

There are generally two types of tasks an AI agent can perform – information tasks and action tasks. Information tasks could be research-related (eg. find all the books that relate to rainforests, what's the difference between fusion and fission, and what features do the most popular national parks have in common), where action tasks relate to completing a specific piece of admin related to an underlying skills such a travel, outreach or scheduling (eg. book my flight to NYC, contact this customer to see how they like our product, or move this meeting for me).

As with anything on the internet, user safety comes down to two things: 1) the company providing the service being incredibly diligent, and 2) the user’s diligence in ensuring it is a reputable company. With Ninja, we’re creating the safest possible environment – from security, to data privacy to ethical use of our AI agents. Our agents are trained on thousands of interactions and tested rigorously for ways to break them; they are trained to not engage in unethical content with or at the request of a user. Without user trust, AI agents won’t be able to achieve their potential as their efficacy is based on how often they are used. It is critical that NinjaTech AI and all AI-focused companies take user trust and safety very seriously.

AI is making agents more intelligent, capable, personalized, context-aware, and proactive. This is making AI-led assistance more useful and valuable to users, by taking on a wider range of complex problems thereby freeing up time for busy professionals to be more productive. With AI agents, we're pursuing artificial "nice" intelligence. This means that we believe the future of assistance is with a plethora of AI agents that are highly trained and specialized on certain tasks. This multi-agent future will see AI agents for an increasing number of problems and use cases, accessible through a personal AI like Ninja.

Potentially. AI agents cannot understand every language - yet - but they can understand many languages. AI agents are trained on massive datasets and code. However, these datasets haven’t been expanded to every language yet – but progress is happening!

AI agents learn over time by interacting with users and being exposed to new data. They use this feedback to improve their skills and perform better in the future. AI agents can also learn across tasks; for example there are commonalities between booking a meeting and booking a flight. As AI agents get exposed to more problems and learn which solutions are successful at solving certain problems. They can store these learnings and draw on them for future problems. It is also important to point out that these AI agents can actually learn from each other -- in effect AI agents collaborate with each other as they solve problems by drawing on their specialties.