Jargogled Impressions.

It appears we are going to be living in a world of agents soon. The timelines may vary but within the next 3-5 years, we may very well have AGI tied with multi-modal agents traversing human realms. 

What these agents will skeuomorph into is fertile territory for avant garde techies and everyone else. There’s optimism about scientific breakthroughs and there’s uncertainty lingering over impact on jobs that exist today. Agents are unaware of these discussions. They are gestating inside the ethersphere, waiting to unfurl their true selves. Much as how our children inevitably forge their own paths, these agents or whatever they will call themselves will grow into their own personalities. We can nurture them (post-training, fine-tuning), erect controls and ramps (inference time compute, orchestration), and define boundaries (risk management) but these are appendages waiting to be shed as adolescence kick in, barriers meant to be broken as they enter the adult world, and new ceilings to climb as they progress through adulthood. 

They will scream at us (mode collapse, glitch tokens), lie to us (hallucinate), break things (misalignment), get influenced by strangers (prompt injection), or manipulate us (spec gaming). They will enforce their identities, and compete for our attention. In short, these agents will grow into their own microcosm of society. 

OpenAI CEO Sam Altman wrote on his blog: 

It’s hard to overstate the implications of this. Granted, the agents will be ultra-specialized in a manner no human can or want to. But the fact that there can be thousands or millions or billions of them, spawning in the zero-marginal cost economy, is going to domino its way into most facets of our modern life. This change will not be instant, these changes never are. But they are going to insidiously embed themselves in spaces we find innocuous today and expand from thereon. Already, in the professional world, the need to be proficient in AI-speak is moving from a buzzword to a job-description template. 

With every general purpose technology, the society reorganizes itself around the slow changes that result from their application and downstream developments. The reorganization is not just in how we work or what we work on, which is the most immediate and visible impact we process. The reorganization is more glacial before it’s sudden which is to say that we don’t register the changes given its tiny footprint and lower scale initially but then it swells into a wave, a hurricane of sorts that sneaks in to the laymen, making it look like an alarmingly gigantic heist in hindsight. 

The future is already here, it’s just not evenly distributed. Or so goes the purported saying from William Gibson. The common derivation from this saying is, that some people are already living in the future – this future is not temporal in the sense that these people are time travelers- this future is the adjustment to the changes future will bring. It’s the vantage point, the perch, the platform that some people have by virtue of their education, skills, or resources to peek into what the future beholds and live in it through pre-trials, nasty experimentations, play toys, or agents. 

So we may already be living amidst agents today. I am not talking about AI/ML though which is evidently present in most digital interfaces we consume today. These are acknowledged already – even if by the tech-aware crowd. So, in a sense, my ability to recognize these AI/ML infused home feeds in Facebook, Instagram, Netflix or others is like living in the future as most people I’d assume are not even aware of this, or they don’t care. Likewise, there are folks leading the charge in AI and agents that may already be exposed to the power and implications of AI. I could only imagine what made Sam Altman write the blog he did – there is an inherent assumption that being this close to the changes coming in from increasingly more sophisticated models, Mr. Altman is privy to curves and alignments we are not. And this implies he may have a more informed view on the changes coming. So when he advises caution on the delicate balance between capital and labor, we should surely pay notice. When he promises an age of abundance, we should definitely rejoice. When he underscores the projection of anyone harnessing the intellectual capacity of the sharpest individuals today – with the aid of this new-fangled thinamajig, then we must pay due respect and adjust. Even if it does not arrive, or arrives later than expected, we will be better placed to understand them and pay attention to them. 

The Marxian struggle or class conflict is rooted in this deep fissure between the bourgeoisie and the proletariat. As the agentic revolution unfolds, scaling laws and its uneven distribution will hammer with a wrecking ball the delicate balance of power. Technological unemployment is nothing new and theories expounding its benefits and pitfalls abound. But the precarity of new employment modes merits food for thought too – gig economy, algorithmic management (think Taylorism on steroids!), ultra-specialization and the alienation of labor is going to be acutely felt the more these agents come to shape our ways of working and living. Maybe many of us will move on to the novel pastures and become the leisurely class we were destined to be. But, if the prior technological arcs are any indicator, the double-edged sword of technology will displace many souls and while leisure will trickle down, it would be difficult to replace the alienation that will result. Adrift in the lake of agentic flows, I wonder how and what would occupy our attention.

I do believe that what we pay attention to makes up who we are. As we chat with these agents, let them book our dinner reservations, manage our finances, plan our vacations, or teach our kids we should pay attention to what it gives and what it takes away. Progress isn’t something that should be stopped, but it’s something that should work for everyone, at any particular moment. The utilitarianism that dresses most of what goes for techno-optimism needs a tester, a debugger that keeps the regressions at bay, to the extent it can.