There is a new breakthrough every day in the generative AI landscape now, perhaps every hour. New frontier AI model updates launching every other day, AI solving math problems that were unsolved for decades and frontier open source models being released every week. It has become difficult to catch up, and that’s understandable. Human context is limited and not as flexible as the growing context windows of frontier models. We have never been exposed to a world where tech moved this fast, although tech has always been defined by change.

A few months ago when I started this blog, I wrote about asking ourselves the end goal with AI . I also wrote that we don’t know when we will achieve Artificial General Intelligence (AGI, the state where AI is broadly as capable as humans), and probably we may not see that in our lifetimes. I think that needs a correction - I believe we are slowly getting there.

Is AGI really here?

It’s very tempting to think of AGI as a state where we have an AI that works like a real human (intelligence). The AI models that we have seen so far aren’t magical, they still run on text prediction and mathematical representations of data in multiple dimensions. Inherently, our brains and AI models work differently. That said, AI models have become really smart now.

The best way to think about the progress is to compare the AI models of today, to those of few years ago. Models have come to hallucinate lesser, the harnesses built around the frontier models (like Claude Code and Codex) have become smarter and offer broader capabilities and an AI layer is getting embedded across every application out there. There is no denying that AI is helping in our daily lives.

Of course, that doesn’t mean everything is perfect now. Even today, you cannot fully trust a model to do critical tasks on your behalf without your oversight. For instance, we are yet to see AI agents managing bank accounts for the masses. Personally in my work, I have still not felt comfortable letting AI agents run autonomously without my approval for important tasks, and at least for me, I hope it will remain the same for some time. Why? Because generative AI in general does not work in a deterministic way. There is a bit of randomness baked into the process, or there is some context missing to the AI, which is hard to evaluate all the time. That said, we have come to trust the models more than we used to, and as the models get more fine-tuned, and harnesses get more defined with greater context, we may be able to trust them more. It’s still hard to define when we’ll hit that ceiling, but I feel we will get there in the near future, at least in an imperfect state.

How should AI progress further?

It’s hard to predict how AI will progress from here, although progress is obvious. It is easy to guess there will be momentum, but difficult to say in what directions.

A few thoughts:

  • While many frontier model companies are requesting slowing down of AI development, practically it’s not easily possible. There are already open weight, open source models that are highly capable, and that can act as a catalyst to further AI developments, without any regulations.

  • Guardrails. More guardrails.

    As more developments happen, we should deliberately focus on making existing and upcoming models safer. Models that are aware of their boundaries and trained/tuned to not cross them would help.

  • Safety is more important than the ability to achieve goals

    An AI model or harness should only help you achieve a goal that is within the boundary of safety. It shouldn’t break into a network/environment that it is not authorized to access. It should be trained to respect safety than just be incentivized to achieve a goal for you.

  • Focus on outcomes over capabilities.

    Before delivering any new capability, frontier model owners should ask and validate what outcomes would the new model’s capabilities bring. Many capabilities while being constructive on the surface could be equally destructive. A model that is inherently aware of them, and fine tuned to act against any destructive capabilities would work better.

A note on staying grounded

Over the last couple of months, I have come to believe that the two most important AI skills of today are the ability to:

  • stay calm in a fast-moving AI landscape
  • cut the noise and stay focused on execution

Honestly, I don’t have both these skills at this point, and I am working on them.

Why are these important? Because it’s impossible to catch up with the pace at which things are happening today. If you want to learn everything and know everything, you are going to get burned out. But if you can focus on what matters to you and stay grounded, you have the ability to execute without fretting.

Let’s take a closer look at these two skills.

The ability to stay calm in a fast-paced AI landscape

Every new model update is a trigger for anxiety to anyone working in tech today. Not because of the latent destructive risks a new AI model could possess which we are starting to see now, but because it can make you question your existence in your profession.

These are some anxious thoughts that can cross your mind:

  • What if this new model/update replaces my job?
  • Will this new feature make my role insignificant?
  • Will there be any value in what I do today?
  • If today’s update can do X, how long until it will do Y and Z, eventually eliminating the relevance of my skills?

What can you do to stay calm?

I have been closely watching how I feel about new AI advancements since the beginning of 2026. There’s one common denominator throughout all these months – most AI advancements won’t change your job role overnight, it takes time. Of course, it is going to affect the way we work as AI augments our work, but it’s not going to replace everything you do overnight. Why? Because right now we do not have systems that can provide the full context of your role to an AI. This is going to change but our work landscape is not crafted for easily shareable context. Ask this to yourself: “If I leave my role today, is there enough context available out there for someone else (human/AI agent) to seamlessly pick up from where I left?”

“If I leave my role today, is there enough context available out there for someone else (human/AI agent) to seamlessly pick up from where I left?”

If you said yes to the above question, your role might be susceptible to AI automation. The only way to stand out in an AI dominated work is by adding reasonable value as a human, that the AI cannot fully influence. If you don’t know what that is, counterintuitively, ask AI what could that be. And if you can’t find that, start upskilling. A volatile world is no problem for someone with multidisciplinary skills. It’s so fascinating that this age old advice of broadening your skillset (especially in tech) is aging like fine wine today.

Secondly, roles that need human connection or presence will thrive. Sales, consulting, customer success (I might be a little biased here as I work in this space) are some examples. It’s not that automating some of these roles is difficult. Human presence helps build trust, which is essential for these roles. Unless a massive cultural shift happens, and we don’t know when/if it would, it is going to stay this way.

Cut the noise and stay focused on execution

The world has never accumulated information at this pace before, thanks to generative AI and people obsessed with the opportunity to create content. There are real implications for the quality of content that’s available, but that’s a topic for another day.

I have been heavily using Twitter/X for the past few months. My feed is all or mostly about what is happening in the world of AI - new model updates, people vibe coding creative apps and a lot of commentary on how the world is going to end because of AI. While some of them are useful, and others polarizing, I have come to realize that most of this information is not helping me achieve my goals. I have also come to realize that the best approach is to not cut it down fully, but to limit exposure. In today’s era of instant gratification, that’s easier said than done.

The bottom line is to preserve your attention span, focus on what moves the needle (your interests, topics that help you learn and work better) and deliberately cut down on the time you spend on these apps or activities.

Summing it up

It’s clear that AI is helping us in our work and lives. We are still in the transition phase where we are getting used to AI as an embedded layer in our work, and at the same time AI is advancing at a rapid pace. We are yet to witness a state where AI is superintelligent to fully overtake what we do. There is a gap in the context that is available for AI and how best it can safely act on our behalf. And more importantly, how greatly we desire AI to act on our behalf.

The right balance is to stay grounded, cut down on how much content we consume and be focused on what matters to us in terms of our interests, work and life.