If you're reading this on a desktop, we suggest you open the PDF-version, but if on mobile phone - continue reading below.
Except for math, very few things in society are binary. This is a frequent phrase I’ve used when presenting my theses of the future to audiences. I use it to explain that new things take time to proliferate. The reason for this is that I’ve often heard the argument that just because something new is not better than the old, it is a signal for the something new to never replace the old.
When looking back in history and analyzing the emergence of disruptive innovations, it is abundantly clear that difficult things take time to develop and proliferate. Whether the internet, smart phones, e-commerce, autonomous vehicles, or circular business models – adoption rates of disruptive innovations take time and several iterations to reach a state of sound critical mass.
Big disruptive innovations are difficult to develop, and they take time because of it. But once enabling factors align and unleash an improved customer value or economic rationale – the classic S-curve goes exponential.
Over the past few years, the term “Artificial General Intelligence” (AGI) has been popularized to the point where the core definition of it has reached somewhat of a fuzzy territory. Fuzzy in respect to the fact that there are differing views on what AGI is and when we’re able to say that we’ve reached AGI. According to the traditional definition, AGI is when a computer can surpass human cognitive capabilities across a wide range of tasks. And in terms of time scale, projections range between a few years and a few decades.
Listen on Spotify:
OPENAI’S FIVE-LEVEL PROGRESSION SCALE OF AI
Again, what we’ve learnt from history, is that almost nothing is binary. Things develop in stages. And this is exactly in line with the statements that OpenAI and Sam Altman gave a couple of months back, where they outlined a five-level progression scale for AI and its’ capabilities. Very similar to the five stages of autonomous mobility (assistance, partial automation, conditional automation, high automation, and full automation).
Level 1 for OpenAI’s five-level progression scale is represented by Chatbots (or conversations, in other words), which we’re currently in, although very close to bridging over to Level 2 – which is Reasoning. This level was, arguably, initiated with OpenAI’s most recent model, the o1. This level can be defined as a system that can solve basic problems at a level of a person with PhD.
Then we have Level 3, and this is the entire topic and message of this research letter. This level is moving beyond task-specific functions and into generalized decision-making and execution. Level 3 is about Agents, that are able to take action on a user’s behalf. And this is a very big deal.
Because when AI agents can process information, assess the information, and act upon a task – it is possible for a range of potential impacts that are exponentially more impactful than previous levels. According to Sam Altman, this is a big leap forward and could alter the way software applications interact with each other.
But before we go deeper into AI agents, let’s mention level 4 and 5 as well. Level 4 is about Innovators that are able to generate new innovations and ideas, especially in areas like science and engineering. And then we have Level 5, which are called Organizers and represent the endgame (but who knows at this point) of AGI, where AI can manage entire organizations with complex process. Superintelligence in all aspects.
So, in distillation, we have five levels that include Chatbots, Reasoners, Agents, Innovators, and Organizers. Each representing a significant leap forward in making computers work humans.
OpenAI's five levels of AI. Source: OpenAI, Sircular.
2025: THE YEAR OF AI AGENTS
In the coming months, we’ll most likely see a flood of events where AI agents take product form and become a new way of creating a customer value, for basically any type of industry. Think agents helping doctors to analyze millions of data points and take actions for patients that need healthcare, or autonomous vehicles that finds the best possible routes possible for any trip and for anyone on the planet.
Or agents that can understand what type of website or application non-coder people want to create, simply by understanding everything about the idea. Lovable is a Swedish company boasting such vision with their recent release of their GPT Engineer, able to generate production-ready code in real-time, and then debug it and maintain its’ integrity. This company just yesterday (Oct. 7th) announced a USD 7.5 million pre-seed funding round, after getting significant user interest for their beta-product. They aim to create “the last piece of software” and enable anyone to become a digital creator.
Now, if we pose the scenario that AI agents will autonomously be able to, let’s say by spring next year, code any type of website or application. What does that mean for the software industry? I for one believe that (though not a rocket science scenario to come up with) these agents will become our own personal employees, that we never really knew we wished wanted all along. Digital employees that never sleep, take breaks, learn exponentially faster than humans, and process oceans of data in real-time.
This would mean that the barrier of entrance into the software developing space (over time, of course) is diminished and opens the pandora box for a future that’s quite difficult to imagine at this point. But what is certain, is that AI agents will learn from us, adapt to our preferences, and execute tasks without requiring our constant oversight.
We’ll all create and personalize our very own agents that become our personalized interface with many of the software applications we use. They will cater us the most personalized video service ever, they will oversee entire departments in companies, manage production lines, monitor patient data and recommend treatments, and even become our shopping assistants, life-coaches, and analyze markets and manage investment portfolio.
This will all start happening in the coming months and no doubt materialize during next year to the point where we’ll see a novel wave of startup companies developing AI agents for multifaceted use cases.
AI RESEARCH AGENTS – WHY YOU NEED TO PAY ATTENTION
Now, this is where things get really intriguing, especially for the field of research and investments. For my home turf, in other words.
Up until now, or at least a few months from now, researchers of companies, industries, and new technologies (or anything else), have never been able to create automated scripts that constantly monitor thousands of events every day and distil these events into a summary of prospective impacts for the future.
We’ve never had AI agents that we can instruct a long list of tasks to constantly conduct, and even talk to each other so that they can collaborate in creating the output we want. This is now becoming a reality.
An AI Research Agent. Source: Sircular (made with ChatGPT)
Let’s take James Bond as an analogy to the point I’m trying to convey. Throughout all the movies about the legendary agent 007, he had almost complete autonomy in gathering intelligence in order to execute strategic goals for the British intelligence service (MI6). This includes the license to kill (which is the “double-oh”-part of his codename) if deemed necessary for the sake of the outcome goal. In essence, he is autonomously tasked to analyze and execute.
The research counterpart in this analogy is that we’re all going to be able to employ our own AI 007 agent with the license to autonomously analyze and execute. Not execute in the brute kind of way, but execute in terms of the research output we want.
A tangible example, in my case, is that I have constructed almost 2,000 future scenarios for all sectors of society, that might or might not happen in the coming decades. Everything from autonomous mobility depreciating personal car ownership and general-purpose robots disrupting many traditional product categories, to AI enabling renewable materials and a circular economy.
When each of these scenarios are governed and operated by AI agents that are tasked with understanding their respective scenarios in extreme detail, it is possible to instruct them to create any type of research output I want. They will be employed with assessing every single news item that relates to their scenario and write an analysis on their prospective impacts. In practical terms, this means monitoring competitor moves, keeping an eye on regulatory updates, pulling insights from conversations on social media, and constantly trying to discover new startup companies.
Having been both a researcher and investor throughout my career, not only do I think this opportunity means that AI agents are licensed to research autonomously and tirelessly – I believe AI agents comes with the license to win. And by win, I mean being aware of prospective signals for the future much earlier than other investors, as well as being able to cross-pollinate events and future scenarios that simultaneously happening all around the world.
Our vision for Sircular is for our customers to subscribe to both pre-defined and self-created future scenarios, governed and operated by AI agents, making them much more effective in their investment strategies. Essentially, investors will have their own digital workforce that constantly streams insights on-demand.
CONNECTING DATA WITH AGENTS
Now, AI agents will not only take on multiple tasks in their respective fields (whether cars, healthcare, finance, or research) they will also collaborate with each other across fields. They will be able to be connected to each other, by their human owners. Agent-to-Agent (A2A) delivery and collaboration could emerge as a next big thing for both businesses and consumers.
We’ll even be able to open proprietary databases for AI agents to explore and create output from. Kind of like an API today, but with the ability to assess the best possible route to an output, without pre-defined APIs. This is why we’re developing a product called “Second Brain”, which will enable investors to place their data in this Second Brain (like pitch decks, industry reports, etc.) and own their data without having the risk of it ending up in the hands of someone else. And by the way, our Second Brain is comprised of the latest generation of Knowledge Graphs, enabling fully truthful output without hallucination.
The connection between Second Brain and AI agents comprises our vision for the future of research and investing. Or in practical terms; connect your own proprietary data with AI agents that constantly work for you.
This is why we believe that AI agents hold the license to win in the game of investing.
All the best,
Christopher Lyrhem
Chief Future Officer at Sircular
This research letter dives into what we believe could be the most disruptive development in the AI space over the next few months. If you're interested in understanding how AI agents will reshape the future, stay tuned for more insights and sign up for "Welcome to the Future"—Sircular’s flagship research letter providing foresight into long-term future scenarios across all sectors.
תגובות