Survival of the Lightest

Problem-solving in lightweight mode

“Late to class” © 2025 Jesse Kim

Inside my local supermarket is a sushi section where a usual variety of sushi packs and made-to-order platters are prepared. Unfortunately, this is run by a lady who screams “I HATE MY JOB” visibly and persistently through body language and customer service. As a casual patron, I can’t help but foresee the deployment of a culinary variant of the Atlas humanoid robot as her replacement.

This robot will, of course, predict accurately what will be in demand each day, cook with utmost consistency, greet customers cordially, and eliminate waste throughout procurement, production, and sales. That will be a far happier future for Woolworths, its patrons, and most importantly, the lady in question herself. In the meantime, I am left pondering what’s in store for humans who are likely to remain on the workforce, voluntarily or otherwise.

Beyond the dire need for sushi-making robots

Song Gil-Young, a data expert turned author in Korea, coined the term lightweight civilisation in his 2025 K-publication. I have consumed enough promotional YouTube clips of it to grasp the author’s key observations. In the presently unfolding future of work, there are three emerging patterns in the way businesses engage human contributors. Here's my heavily paraphrased translation:

• Forecast #1. “We meet now.” Gone are the days where employers and colleagues would offer the luxury of “In six weeks, you will have become a subject-matter expert in X-Y-Z, won’t you?” Businesses no longer hire humans only to train them up over a period of time and subsequently lose them. At work, ‘learning’ can only be a past tense unless artfully executed just in time (JIT) without slowing other humans down. A learning ‘curve’ is in fact one straight vertical line. Employers rightfully insist on prompt, responsible delivery of real-life problem-solving outcome.
• Forecast #2. “We meet only for a brief moment.” A problem-solving engagement is designed to produce results in weeks, not months or years. Once the job is done and handed over, a self-actualising contributor moves swiftly on to the next value-adding gig or assignment, wherever that may be, internal or external — that is, before some other move is forced upon the said individual. In the end, sense of belonging in any business hierarchy over any length of time evaluates to false. Self is where it starts and ends, more so than ever before.
• Forecast #3. “We will meet again, maybe.” One successfully executed engagement, however brief it was, can eventually (or immediately) lead to an invitation to another involving the same or expanded parties. Irrespective of employment status, a chain of ‘lightweight’ engagements builds a lasting working relationship with any business. A corollary of this, however, is that it does not take much time, patience, or effort at all to weed out human contributors deemed incompatible with the business.

[End of translation]

Perhaps these are prematurely dystopian views coming from the country Elon Musk loves to pick on as the perfect breeding ground for super-intelligence and singularity-age technology. Or maybe there is nothing premature or dystopian about them at all. Either way, the above observations give rise to the notion of lightweight across just about everything in a work context; everything except the expected magnitude of output, that is. I see various connotations of lightweight, including:

• Simpler, smaller, and shorter transactions
• Reduced complexity, moving parts, and dependencies
• High-speed decision, mobilisation, execution, evaluation, and adaptation
• More execution and less slide decks about execution
• Better-articulated risks
• Fewer toll gates along the road from inception to fruition
• Straightforward ownership
• Cost-effective documentation, maintenance, and futureproofing
• Uncomplicated, ubiquitous, and repeatable adoption
• A smaller team with fewer mouths to feed

The greatest single point of failure is a constant

In short, lightweight is a strategy to augment return on investment by slimming down the investment for the same return. Such has seemingly become common practice among businesses equipped with relevant 2020s’ technology. The essence of lightweight that I see, however, is not at all about “sending the heavy lifting in AI’s way,” or some variation of this fatigue-inducing platitude.

Arguably, what the different kinds of artificial intelligence do deliver, and at what cost, varies from industry to industry and from role to role. In my view of the world, the greatest single point of failure, or the weakest link, in any business undertaking remains the same pre- and post-AI: disconnected, ill-structured, or poorly regulated sources of knowledge (or DPRK for short; maybe I will stick to the long form). If required and backed by requisite domain expertise, fit-for-purpose legwork by available forms of AI can drive advancement in conjunction with traditional, mundane-sounding endeavours — automation, discovery, due diligence, and simplification, to name a few.

Across many of the use-cases in my vicinity, however, the hype-free reality of current language-model AI is a proliferation of resource-hungry solutions looking for problems. Bloat, slop, infrastructure bills, and corrective labour together run the risk of pushing complexity and total cost of ownership up, not down. Without proper stocktaking of the problem and its constraints, any nonchalantly executed brute-force will end up burning more scarce resources than there are benefits. For instance, adopting an agentic AI solution to orchestrate automated actions that leverage corporate knowledge, will achieve very little if the requisite sources of knowledge remain unavailable or unfit for intelligent consumption.

In any case, it won’t take long, I hope, to see which large-language model species survives and emerges from this modern-day *Sengoku* (Warring States) phase, to become a scaffolding for problem-solving upon the aforementioned connotations of lightweight.

A concrete dome or multiple layers of it

An analogy came to mind while I was watching a documentary about a certain tragic event unrelated to artificial intelligence. A unidirectionally additive pursuit of efficiency by pouring in more computing power and non-human capital is akin to constructing a concrete dome, or multiple layers of it, over a problem site with no questions asked. In other words, adding bloat that is not understood or explained violates the substance of lightweight. That cutting-edge bloat may, for instance, introduce a paywall that stands between an organisation and its own data. I see lightweight as a product of human endeavours to:

• Study, question, validate, and synthesise the subject of problem-solving.
• Remove noise, overlaps, and unsustainable dependencies.
• Engineer a solution that either overcomes, nullifies (lawfully), or works through each identified constraint — for example: processing time / raw materials / awareness / jurisdiction / software licences / population density / logistical infrastructure / locally produced ingredients / electricity bills.
• Judge the value created and optimise accordingly.

Even in the hypothetical age of perpetually toiling agents and robots in charge of economic output, problem-solving will require appropriately qualified humans placed before and after the grunt work performed by agents and robots. BEFORE: to articulate principles, constraints, deliverables to attain, and value to create. AFTER: to act as a subject-matter firefighter, auditor, and janitor with enormous sign-off accountability.

Devising a sustainably lightweight solution to a problem, in my opinion, hinges on a set of patently human qualities, namely curiosity, peripheral vision, philosophy, articulation (as opposed to approximation), ownership, finesse, critical thinking, and lateral thinking, plus some degree of clairvoyance and premonition backed by deep domain expertise.

A bloated preamble

All this not-so-lightweight description of lightweight was going to be a preamble to my field notes on addressing the conundrum of disconnected, ill-structured, or poorly regulated sources of knowledge, with technology that facilitates sustainably lightweight execution. That will now be a separate piece.