AI and the Dragons that Dance

Technological Self-Reliance is a key goal of development. With the rapid advance of AI coding capabilities, the work of human coders will now incorporate double-checking AI-generated code rather than authoring it. However, AI will not replace human coders. Rather, AI is gradually altering the human skills needed in any coding-related job.

In Beijing, apart from AI dragons that dance like DeepSeek, Manycore Tech, Game Science, Deep Robotics, Unitree Robotics, and BrainCo, the city now has over 4,500 dragons developing and selling AI services. Primary and Secondary Schools in China will introduce AI courses this year. Beijing plans to invest 10tn Yuan over the next fifteen years in AI.

This is in addition to the 60 billion AI investment fund created in January 2025. The fear in the West is that AI is hungry for data, and the more it mines and refines, the smarter it gets. This underpins every concern about DeepSeek, TikTok, and RedNote. Children in China, at the age of seven, are playing with toys made from small coloured programmable bricks. The toys are controlled through code assembled by the child on a smartphone. Whalesbot is making AI toys for three-year-old children. Every package of blocks comes with a booklet of code. The toys are designed to teach the child to code.

Code is a manifestation of the ideas and concepts of human coders. This is why I remain resolute in my view that AI is a human. Coders will develop user stories and choose infrastructure. Human coders spend a lot of time figuring out what human users in the infosphere want from a piece of software and try to create a frictionless solution using the pain points unearthed by use cases.

Good product ideas come from developers and the prototypes they build. However, AI-driven changes to critical roles mean that in a few years, countries will see developers gradually moving toward higher-level tasks as they evolve from being primarily code writers to mostly code assessors.

This entire process is fluid, experimental, exploratory, and iterative. The software that is first imagined – or even first coded – rarely sees the light of day. Opacities, false assumptions, and orientation changes make coding human. Coding a particular algorithm or an optimal function is not formidable. What is difficult is building a complete end-to-end system that solves a human problem.

Human coders will always focus on thinking up interesting ideas, configuring the best user interaction design, delegating, and then figuring out how to review the outputs at scale. Ultimately, the final stages will require some static analysis and some AI-driven analytic tools.

Today, there are huge legacy opportunities for AI-generated code in bureaucracies. At the local government level, many projects do not attract the talents of coders. In these cases, AI is a perfect catalyst. These projects include intricate legacy migrations and modernisations. These “death march” projects in municipalities can be accelerated using AI.

Human developers will continue to handle the more abstract work that AI models cannot manage and provide system oversight. Lit by the slow fuse of the imagination, developers will revisit abandoned ways, dead ends, shelved projects, and blockades using serendipitous breakthroughs from parallel projects. Humans will still work on how the products align with human needs, how to unpack the insights that the use cases uncover, and how to design a cohesive product strategy.

The nexus between how coders work and how creative people do what they do sheds light on how to introduce coding as a transversal competency in schools. Examinations of the early diaries and notebooks of the Nobel Laureate Poet, Derek Walcott, show how he used glue and small bits of paper to meticulously paste over parts of poems with rewrites. All of these edits would be dated in the margin. This technique allowed him to create spirals of metaphors inside a poem that induce the reader to mentally circle back in ever-widening spirals of images.

Elsewhere, Walcott would Mokojumbie his metaphors so that the vehicle of one generates the tenor for another. This is especially true of his poetry in “The Star-Apple Kingdom”. In Walcott’s poetry, metaphor becomes a flowing metamorphosis, not a set of stable or punctuated analogies. In “Dream on Monkey Mountain”, the main metaphor is an unappealing old man with a round white full moon above his shoulder.

The journey of that moon which drew the unattractive old man through the cycle of one night, layers into a stream of persistent problems that colour the Caribbean. And like the evening sun that resembles a brass gong vibrating over the cane fields of Couva, the visible resonances of figure, sound, and image become concentric and subject to all kinds of true and perhaps incongruous interpretations.

This way of working is not alien to the way code is now written. Just as the poetic experience of setting up a dominant metaphor creates auxiliary metaphors in the work stream, ideas now flourish across coding work streams in concentric circles. For Walcott, a rich metaphor is generative, while for the coder, an unconventional idea becomes a gong that sends waves of designs outward in concentric circles across squads and product owners.