The Evolution of AI: Looking to the Future of AI Assistance
The big buzzword (letters) in tech: AI. Since it’s a highly public profile these days, growing notoriety over the last couple of years, AI (also filed under Machine Learning, LLM, NLP, etc.) gets a lot of credit for its advancements and use cases, but we’re still far from realizing its full potential.
When it comes to folks who know Machine Learning, Bjørn Furuknap is the guy to ask. He’s been in the field for years (like, since ones with a 9 in them, years) and has seen how drastically…and slowly…things have changed. As part of our weekly series, Gun.io Saloon Seminars invited Bjorn Furuknap to chat with our community about his own AI journey, what he’s been working on as he aims to create smarter and more efficient AI systems, and the types of developments that he believes are paving the way for the future.
From Large Language Models to Agentic AI and Beyond
When most people think of AI, they think of large language models (LLMs) like ChatGPT or Gemini, which is fair; it’s the gateway to Artificial Intelligence that most have been exposed to over the last two years. These models are designed to predict the next word in a sequence and engage in conversations. But while LLMs can hold conversations, they don’t “do” anything. They can’t make bookings or take action—they are static systems based on their training data.
What Bjorn is working on now, however, is a step beyond LLMs and even agentic AI: Personalized AI Assistants. This is AI that doesn’t just talk the talk but takes action (walks the walk, if you will). You can assign it a task, and it will learn how to accomplish it. Imagine telling an AI, “Write me a blog post about AI’s role in education.” The AI doesn’t just respond with text; it understands the task at hand: researching the topic, drafting the post, and maybe even finding relevant images. AI is a true assistant—not just a conversational partner.
The Problems with Current AI Systems
Let’s take it back a step and look at why this is an interesting concept. For those not well-versed in the “how AI actually works,” here’s some quick learning. At the core of many issues with today’s AI systems is their static nature. They don’t learn or evolve after their training. Once trained, models like ChatGPT or other LLMs are set—they don’t update unless they’re retrained with new data. This means they can’t learn from their interactions or adapt to your needs unless someone updates their dataset. For instance, if a new version of a framework like React is released, your AI model won’t be aware of it unless it’s retrained. This leads to the issue of static knowledge and how hard it is to keep AI models updated in real-time.
Moreover, many AI systems lack local knowledge. They’re trained on general knowledge, so they can’t understand specific things related to you—like your personal preferences or specific company data—without significant customization, which is time-consuming and expensive.
Bjorn Furuknap’s Journey with AI
Bjorn Furuknap has been working in AI for over a decade, and his experience started long before large language models were “the thing.” In his early days, he worked with object detection, but the computing power available wasn’t quite enough to take things to the next level. So, he waited.
It wasn’t until around 2022 that he really deep-dived back into AI, focusing on developing frameworks that could fill the gaps he’d previously encountered in various industries. At the time, he was running a small game company and saw a consistent imbalance between the resources they had available. Sometimes, they had the right people to build a UI prototype…but no programmers to implement it, or vice versa. So, he began exploring how they could use AI to augment the team—essentially, building AI assistants to help out wherever they were lacking.
Building AI Assistants for Real-World Tasks
The system he built used AI assistants to fill in the gaps. For example, if a designer needed help creating artwork or a programmer needed help generating code, they would work with a specialized AI assistant for each task. These assistants were powered by Agentic AI, allowing them to understand and fulfill complex tasks.
But they soon realized something powerful: they could enable our AI assistants to learn.
Meet Rob
At one point, they gave the AI assistant Rob the ability to learn how to solve tasks it didn’t know how to do. For example, if Rob didn’t know how to create a certain type of animation, “he” would search the Internet for tutorials and code examples, learn the new skill, and teach it to other assistants in real time. This was revolutionary because, instead of just having an AI assistant complete static tasks, it could self-improve, not just itself, but the team of AI assistants. Great!
Until – yep, sorry – the team got hungry, and Rob did something a little too smart that caused some unease. It was realized that if Rob was allowed too much autonomy, there could be some unexpected, potentially dangerous outcomes. Theoretically, “he” could/was learning to do things that weren’t in line with his use-case intentions. While it made the team cautious, and they did limit Rob’s capabilities (and shuttering “him” for good), it also highlighted both the power and potential risks of Agentic AI.
Looking Toward the Future: Concierge AI
While AI has made significant strides, it’s still in the early stages of what could be a revolution. The progress we’ve seen so far is evolutionary rather than revolutionary—gradual improvements in how AI works, but has not yet had that major, major breakthrough. Bjørn predicts that truly transformative AI capabilities may still be 10 to 15 years away.
But, he’s working towards it. As of 2025, Furuknap is ready to iterate on the Rob concept, working on a project called Concierge, a step further in this evolution of AI. The goal is to create an AI that can not only assist, but integrate seamlessly into workflows—learning, adapting, and becoming more capable over time. With Concierge, Bjorn has a vision of an AI assistant that grows with you, understands your specific needs, and can perform tasks with increasing efficiency as it learns. Concierge isn’t just an AI assistant – it is your AI assistant.
Just like AI, this project is still in its early stages, but how exciting is it to explore how far AI can go when it combines knowledge-based systems with dynamic, evolving agents? It seems pretty cool. Sure, it may sound ambitious (and self-stated, a long way from fully realizing this vision), but it’s a future he’s very passionate about pursuing.
The Bottom Line
As the AI space continues to evolve, it’s important to stay open to what’s possible and remain thoughtful about the ethical implications of creating AI systems with more autonomy. The future of AI is exciting, but it’s also something that must be approached with caution and careful planning.
AI is here and being intertwined with everyday life. The future of AI is not about making static systems smarter; it’s about making them adaptable and able to take action in the real world. The framework Bjorn Furuknap is building is designed to push AI toward this goal—creating more than just conversational assistants—and turning it into a tool that can adapt to a person’s needs, learn in real-time, and help solve problems you never even thought to ask for. There are many challenges ahead, but the potential is exciting.
While AI won’t replace humans anytime soon, it will make us more powerful and capable in ways we’re just beginning to understand.
Want to keep learning more? Check out our Meetup page for more conversations about engineering leadership, fun use cases for technology, and connect with the Gun.io community.
