The “I” in AI
Our relationship with AI is changing rapidly as it gains more and more capabilities. Currently, it seems there is no end to it with new surprises at each update. Currently agents are all the rage. Next year, robots perhaps?
It has been more than 2 years since I wrote “Writing Code with Github Copilot and ChatGPT Together — I”. Back then I was impressed, today I am totally amazed. It is also scary, but (at least currently) it is the good kind of scary. Also we have more and more “AI”, some of which on top of my head are Manus, Deepseek, Perplexity, Mistral, Cursor etc. This race currently benefits the consumer.
The “I” in the AI represents “intelligence”. But I am starting think that it also increasingly has increasing relevance with “I”. AI now changes how I work, I do things and even how I learn.
Starting/Going Solo
Shaq O’Neal has a fun memory about Kobe Bryant. Briefly told, there were complaints about Kobe not passing the ball and Shaq said “Kobe, there is no I in Team”. Kobe responded “but there is a m-e in it”.
Carta’s Founder Ownership Report 2025 has an interesting finding: “The percentage of all startups incorporated on Carta that are led by a solo founder has more than doubled over the past decade, reaching 35% in 2024. Over this same span, startups with three, four, and five founders have become less prevalent.”
AI has tremendous effects on building something fast and effective. You practically accomplish more in the same duration and (if you know what you are doing) it is practically more robust. Peripheral standard practices are completed without much hinderance. The main speed up is not the actual code completion, but the elimination of friction.
You can spin up MVPs with fewer people and in shorter amount of time. Though, AI cannot help you on the further steps -yet-. You still need a good and performing team to scale and deal with non standard issues. You -still- cannot rule the team out.
Currently I cannot say anything about the job market. As we might not recall but there are many jobs went away with technology but it always created new jobs. Though I have the feeling that “prompt engineer”, however valuable, will not be a real job on its own.
Magnifying Effects on Learning
AI models boost my performance in the areas that I have the knowledge and skill. They also speed me up on other topics. Naturally, it cannot pull miracles -yet- as in the effect of Matrix (“I know Kung Fu” never applied to me, probably because of age).
I am not a big fan of “vibe coding” so my jabs at frontend development stays as a hobby -for now-.
On the other side, today’s AI is a great cheat code in formal education. It might be blackboards and textbooks are partly replaced with Powerpoint presentations and tablets, education process is almost the same in the last several centuries: Lecture -> homework -> project -> exam.
LLMs can do most of your homeworks and projects (even exams when it applies) without breaking a sweat. Honestly they do it better than most of the students can ever do, so they also get better grades. Therefore, many students are tempted to outsource their coursework to LLMs. It is a great learning tool but if you do not want to learn, it is a great job finisher.
Anthropic (Claude AI) recently released an education report. You can also see these tweets for specific insights. In today’s education system, you can opt-out of learning yet still graduate. You can call this cheating but if everyone is using it and people are being rewarded for it; then it is the new rule.
It also gets creative. LLMs help you pass coding interviews by being sneaky. There was a story of a guy who is kicked out of Columbia University for building a tool to automate coding interviews. I have no idea if it works or not but here is his story and his tool.
It is getting harder and harder to assess people based on their graduation studies, skills by automated testing and other remote stuff. Actually it is harder to evaluate people now, without in person contact. All paragraphs above demand their own research (perhaps I shall use LLMs on these).
Age of Agents
When I first heard about agentic AI, it struck me as another thin wrapper on an existing technology. Though, when I think more and talk to people who are actively working on Agentic AI more, I am getting the feeling it is a step between current state of LLMs and autonomous robots. Basically, give AI arms and legs, and you get science fiction.
There are really a lot of potential applications of agents that might have deep impact. There are a lot of pitfalls as well. I hear some use of agents which solves problems that have no practical added value to current products and services (whatever is the equivalent of “this meeting could have been an email”).
For instance, suppose I hate apparel shopping. Occasionally, I have the vague idea of “oh I think I need a new shirt”. I have preferences, but I need to go through all the shops and websites to get what I need. Of course there are recommendation systems trying to optimize the process for me. But most of the time they fail miserably.
What I imagine is I have an agent and I give it a half baked instruction “I need a shirt, my size is M/L, it should either be white or blue, no fancy stuff”. Then my agent will “knock” on the doors of the websites and elicit alternatives. It will give me long list and a shortlist and I will go through that list and fine tune my instructions explicitly or by elimination.
It is still early stages but I see some good developments. Most of the products and services are bound to one retailer/company. Yet there are gems. For instance, Perplexity has a shopping mode working for the US. There is another promising startup Moojo starting to try to solve this problem via chatbots and a shared profile across all websites.
I am not an agent expert but I had some experience with multiobjective optimization and progessive eliciting of preferences. I hope we see the revival of the research field and we are so close.
So…
I don’t know if this “age of AI” is either the beginning or the end,.. but it is definitely a beginning.
It is already transforming or destroying the way we live, learn, eat, work, shop… Practically we either see or will see the effects of AI more and more.
So, for me it is not a question of using it, but how to position it in my life. It already has a magnitude, the difference will be made with direction.
I am deeply thankful that I put all those spinning thoughts to writing and finally made some empty space in my mind. Let’s see if anything changes in the next two years.
