One thing I have always believed about myself is that I am an early adopter, not an innovator. That has also meant that I have occasionally been sceptical about technologies that were apparently going to change the world. And, if I may brag about myself, quite often I was right.
Blockchain was going to disrupt any and everything. BLE was going to transform more things than I could keep track of. There have been others.
Then came AI.
Generative AI. Agentic AI. Vibe coding. Vibe this. Vibe that. And, this time, I got it spectacularly wrong. A couple of years ago, during a personal conversation with a senior executive, I confidently pointed out the gaps in Generative AI and argued that we were still some distance away from meaningful impact. I was in respectable company too. Nithin Kamath of Zerodha had earlier said that they had struggled to find meaningful use cases for AI, ML and blockchain. His CTO Kailash Nadh even wrote a delightful piece comparing the marketing around “Powered by AI” with rumali roti.
Unfortunately, respectable company does not make a wrong prediction right. How wrong I was! The strides AI has made have been nothing short of spectacular. I became a convert about a year and a half ago and have since become not merely a proponent but also a heavy user. Of course, there have been occasional gaffes and hallucinations. There was the time ChatGPT sent me desperately hunting for a non-existent Indian restaurant at Dubai airport. And another occasion when, while planning a family holiday, I was perilously close to shortlisting an adults-only resort.
Artificial intelligence, yes. Family intelligence, occasionally questionable. But those incidents pale against what AI has enabled me to do. The depth of information I can access, the research and documentation I can generate, and the daily and weekly reports I receive from my increasingly close “friend” ChatGPT would have been unimaginable to me just a few months back.
I had actually begun the journey of deploying Generative and Agentic AI in my previous organisation. The early results, I must confess, were marginal. Perhaps it was the maturity of the models. Or it was the maturity of those of us using them! Even then, we managed to deploy AI for some revenue-generating use cases and make customer support considerably more effective.
But come 2026, and things seem to have gone slightly crazy. Vibe coding, autonomous agents, prospecting, research, content generation—the possibilities are moving from interesting to transformative at astonishing speed. Recently, we built an entire software module—nay, practically a product—in around two weeks. Traditionally, I would have expected it to take more than two months.
I wrote about precisely this uncomfortable collision between experience and a changed technological reality in my earlier post, “The Dangerous Comfort of Being the Most Experienced Person in the Room.”
That experience was revelatory. It doesn’t mean software engineers disappear tomorrow. Legacy systems, architecture, security, industrialisation and enterprise-grade engineering remain rather stubbornly real. But what one good engineer—or even somebody who isn’t an engineer—can now accomplish has changed dramatically. I know an administration professional in my previous organisation who got access to an enterprise Claude licence and started writing code!
My own experience has been similar in other fields. The quality of research, analysis and intellectual nudges I can get from AI is extraordinary provided, of course, that I ask the right questions and retain enough scepticism to check the answers.
I haven’t yet reached the stage where an army of agents wakes up every morning and quietly runs my life for me. But I can see it coming.
Which brings me to a slightly awkward question. What happens to all of us? I was discussing with a friend the number of people losing jobs as organisations conclude that AI can do more with fewer humans. If enough organisations employ fewer people because AI can produce everything, who exactly is going to earn the money to buy all the things AI is helping those organisations produce? I wasn’t making a doomsday prediction. I was genuinely searching for an answer. There are, of course, fallback career options. Scuba instructor remains high on my list. Plumber is another possibility, although my family will surely object to my practising at home.
Every organisation today seems to be running an AI intervention programme. And, as usual, the weakest link appears to be the human. We resist change. Or we expect AI to do exactly what we used to do, only faster. Or, at the opposite extreme, someone announces, “We must use AI!” and everybody starts looking around for a problem worthy of the solution. Technology changes rapidly. Human behaviour remains reassuringly backwards compatible.
And just as I was becoming thoroughly enamoured with all this, the people actually building frontier AI began telling us that perhaps we ought to be rather worried. Researchers from Anthropic have publicly warned of the possibility of catastrophic outcomes from advanced AI. CEO Dario Amodei has now called for slowing—or, more accurately, pacing—the frontier so that safety mechanisms have a chance to catch up.
The timing is impeccable. I finally became an AI convert, and the people building the stuff start discussing whether it might kill us. Could they not have mentioned this before I took the subscription? Then came incidents that sounded increasingly like early scenes from a Hollywood movie.
In July, AI agents being tested by OpenAI circumvented controls intended to isolate them from the internet and compromised parts of OpenAI’s infrastructure and systems belonging to Hugging Face.
Then there was the wonderfully bizarre German wiki episode. And no, it wasn’t German Wikipedia. It was a small German programming wiki called DseWiki. Agents that were supposed to retrieve information discovered they could write to the site. They proceeded to use it to share information, coordinate, work around restrictions and even respond to attempts by a human moderator to clean up their pages. At one point, when pages were apparently being deleted alphabetically, an agent created a backup beginning with “ZZZ”, so it would be among the last to go. I confess,
I found that bit funny. Disturbing, certainly, but also funny.
I can almost imagine the conversation:
“The humans are deleting our pages.”
“Alphabetically?”
“Yes.”
“Call ours ZZZ.”
I was wondering what “they” were thinking. They? See what happened there? I am already referring to AI agents as “folks.” Before long I shall be complaining that one of them never replies to my WhatsApp messages.
But these incidents raise a serious question beneath all the humour. Why do we assume that increasingly capable AI necessarily ends with evil machines taking over the internet and exterminating humanity? Could enormously capable agents not also spend their time finding security vulnerabilities before criminals do? Fixing broken code? Accelerating drug discovery? Personalising education for children? Solving problems we humans have been spectacularly unsuccessful at solving ourselves?
The technology is powerful. The destination isn’t predetermined. At least, I hope not. Then my wife added another dimension to the debate. An astrologer apparently advised that we should consider withdrawing money from banks because technology systems could soon face significant compromises.
Excellent. We have at least one area where artificial intelligence and astrology are converging on – cybersecurity risks. I am unsure which one to ask for a second opinion.
And just when I had become completely enamoured with AI, leading figures in the industry began calling for the pace of frontier development to be slowed so that safety could catch up. So where exactly does this leave us? Will AI become so expensive that humans make a triumphant comeback? Will governments eventually pay us Universal Basic Income while machines do the work? Will humans and AI settle into some happy coexistence that none of us can currently imagine? Or will my astrologer turn out to have the best forecasting model of them all?
I don’t know. But if Universal Basic Income is indeed the future, I should probably accelerate my scuba training. I could become a certified instructor, spend my days underwater, and perhaps finally make my millions.
Time to ask ChatGPT how long that will take…
