Technology and regulation have a dance of ice and fire

Shreyas Prakash headshot

Shreyas Prakash

Let’s take a clear pond flourishing with various aquatic plants — water lilies, duckweeds, water milfoils, you name it. The variety of these plants provide a delicate balance, feeding the pond with nutrients, and this very natural filtration system. The algae still tries to outcompete the aquatic plants, but these plants still have an edge (at least for now). The microbial balance ensures the pond stays in a stable state with manageable algae levels. It’s in stable equilibrium.

But unfortunately, the pond starts having even more higher levels of nutrients feeding in. The runoff is too high, or a group of fishermen have one day decided to start dumping all the fish waste into the same pond.

Now the stability gets disrupted. And it slowly accumulates and grows towards a tipping point. It might just need a small push towards either clear water, or towards an algae-dominated state. In systems theory, you call this tipping point as an unstable equilibrium.

If you think this essay is about the ponds, and the tiny little fish, and the plants, you’re mistaken.

This is a story of the same patterns which we see in ponds, the dance between stable and unstable modes of equilibrium reflected across our economy, especially in the dance between technology and regulation that countries perform — Are they pushing for more regulatory capture? Or are they immersed in risky bleeding-edge technologies?

When Marie Curie won the Nobel Prize for the discovery of Radium in 1903, she also suffered from the consequences of being on the bleeding edge— she was on the receiving end of aplastic anemia and died due to harmful exposure due to radioactive exposure. But then again, the field of radiology was invented giving birth to various other innovations. A gift to humanity that kept on giving.

We see this continue to happen, even now with the advent of biohacking, where we see some of us taking early risks in the middle of a promising technology that’s just taking shape.

I’m recalling a similar such topic where Balaji S talks about the idea of bio-engineered individuals with muscular hypertrophy mutations so huge that they could be likened akin to real-life X-Men scenarios.

This concept may extend to other genetic enhancements in the future, such as artificial immunity to COVID-19 and its variants. Even the case of Lance Armstrong provides an interesting perspective on this issue. While his doping violated Tour de France rules, the chemists who helped him overcome testicular cancer and win the race could be seen as pioneers in performance enhancement. Their work, despite initial risks, could potentially benefit many cancer patients.

This scenario illustrates how early adopters of risky bio-innovations can pave the way for safer, more widespread use.

A similar principle was applied during the COVID-19 pandemic when challenge tests were proposed to accelerate vaccine development. Harvard Professor Matthew Liao suggested using healthy volunteers for controlled SARS-CoV-2 exposure to expedite the process from pandemic outbreak to functional vaccine. These trials offered distinct advantages over traditional field studies which might have taken 10-15 years to develop. When you gain something, you might also lose something (in this example, approximately 1% of early testers reported long-term symptoms).

However, as research progresses, we may see more instances of genetic modifications that push the boundaries of human capabilities, reminiscent of the X-Men but grounded in real scientific advancements.

And for any such ‘frightening technology’ being discussed and implemented, we have the ‘big daddy regulators’ trying to give a pushback on this accelerationism by applying screeching brakes. If the challenge trials represented the tech innovation led approach towards solving the global pandemic. The regulatory approach would be stringent lockdowns implemented by various countries (especially China and UK).

In fact, regulations are one of the major bottlenecks for progress. And they also do exist rightfully so. Talking about our current AI landscape, at present, all the major FAANG companies are building massive GPU cluster farms that are being created needing vast computing resources. In 2024, we were used to having these clusters of 100 MW, two more years, we might have a 1 GW Cluster. If we soon see trillions of dollars invested into these GPU clusters, we might very well see a 100 GW cluster. To accommodate these ambitious targets, we’re seeing a huge push from the libertarian technocrats to now build nuclear capacity to manage the hungry AI farms. Post-Fukushima, it’s been quite controversial to even approach the nuclear topic, and most of these nuclear proposals face an immense regulatory pressure.

Take for instance the construction of Hinkley Point C in UK which took almost 10 years. By contrast, France and Finland built similar reactors in just three or four years (Hinkley Point C had to go through a 44,260 page environmental impact assessment (EIA) and 2,229 written questions at examination stage of Sizewell C, the next reactor after Hinkley Point C, facing enormous expenses before the spade was even in the ground, it is still not). In fact, Sizewell C’s EIA was more than 30 times longer than the complete works of Shakespeare.

We’re a quarter of century into the 21st century, and we’re still dealing with regulatory bureaucratic paper work. Recently, Elon Musk also highlighted an example from the German bureaucracy during the establishment of Tesla’s first European plant where he revealed that the permit application process required up to 25,000 pages of documentation, which not only had to be printed on paper but also needed multiple copies made. The sheer volume of paperwork was so extensive that it literally required a truck to transport all the documents. And this is 2025.

We see dangers of regulatory capture all around with FDA and European Medicines Agency (EMA), but there are various merits of a healthy dose of regulation. Despite the backlash against European regulation, we see its merits when it comes to food regulation, where Europeans have better life expectancy and have lesser preventable deaths compared to Americans. This is majorly because of Europe’s EFSA where food regulatory policies focusing on additives to be proven safe before approval, whereas US FDA allowing food additives unless proven harmful.

Coming back to the story of the delicate equilibrium of the biosphere in ponds, the dance between regulation and innovation quite mirrors this delicatedness: if you push it too far, there might be unrestrained technological acceleration and we risk drowning in toxic algal blooms of unforeseen consequences, and yet too much regulatory stagnation leaves us with a lifeless pond unable to sustain new growth.

Subscribe to get future posts via email (or grab the RSS feed). 2-3 ideas every month across design and tech

Read more

  1. My agentic engineering workflow (step by step)agentic-coding
  2. Every darn thing is a kekulean loop if you notice itdesign-thinking
  3. Hammock driven developmentagentic-coding
  4. Peculiar ways number three fits into our funny little brainsmental-models
  5. AI sandwich as a defacto principle for anything agentic engineering relatedagentic-coding
  6. How I write essays in 2026writing
  7. Authority in the guise of evidencecritical-rationalism
  8. Map is not the territoryphilosophy
  9. Self hypnosis as a manifestation ritualmeditation
  10. Hegelian dialectic for structured reasoning with AI agentsphilosophy
  11. How I prepare for tough negotiations nowadaysnegotiation
  12. When should we steelthread somethingproduct-development
  13. Learning and re-learning my mother tongue in Malayalam
  14. Breadboarding, shaping, slicing, and steelthreading solutions with AI agentsproduct-management
  15. Healthy conflict in teams have a tipping pointteam-building
  16. How I deslopify AI writingwriting
  17. How I started building softwares with AI agents being non technicalagentic-coding
  18. Read raw transcriptswriting
  19. Legible and illegible tasks in organisationsproduct
  20. L2 Fat marker sketchesdesign
  21. Writing as moats for humanswriting
  22. Beauty of second degree probesdecision-making
  23. Boundary objects as the new prototypesprototyping
  24. One way door decisionsproduct
  25. Finished softwares should existproduct
  26. How I periodically rank my rough draftsobsidian
  27. Flipping questions on its headinterviewing
  28. Vibe writing maximswriting
  29. How I blog with Obsidian, Cloudflare, AstroJS, Githubwriting
  30. How I build greenfield apps with AI-assisted codingagentic-coding
  31. We have been scammed by the Gaussian distribution clubmathematics
  32. Classify incentive problems into stag hunts, and prisoners dilemmasgame-theory
  33. I was wrong about optimal stoppingmathematics
  34. Thinking like a shipmental-models
  35. Hyperpersonalised N=1 learningeducation
  36. New mediums for humans to complement superintelligenceagentic-coding
  37. Maxims for AI assisted codingagentic-coding
  38. Virtual bookshelvesaesthetics
  39. It's computational everythingtrends
  40. Public gardens, secret routesdigital-garden
  41. Git way of learning to codeagentic-coding
  42. Style Transfer in AI writingagentic-coding
  43. Understanding codebases without using codeagentic-coding
  44. Vibe coding with Cursoragentic-coding
  45. Virtuoso Guide for Personal Memory Systemsmemory
  46. Writing in Future Pastwriting
  47. Publish Originally, Syndicate Elsewhereblogging
  48. Poetic License of Designdesign
  49. Idea in the shower, testing before breakfastsoftware
  50. Technology and regulation have a dance of ice and firetechnology
  51. How I ship "stuff"software
  52. Writing is thinkingwriting
  53. Song of Shapes, Words and Pathscreativity
  54. How do we absorb ideas better?knowledge
  55. Read writers who operatewriting
  56. Brew your ideas lazilyideas
  57. Trees, Branches, Twigs and Leaves — Mental Models for Writingwriting
  58. Compound Interest of Private Noteswriting
  59. Conceptual Compression for LLMsagentic-coding
  60. Meta-analysis for contradictory research findingsdigital-health
  61. Proof of workproduct
  62. Gauging previous work of new joinees to the teamleadership
  63. Task management for product managersproduct
  64. Beauty of Zettelswriting
  65. Stitching React and Rails togetheragentic-coding
  66. Exploring "smart connections" for note takingwriting
  67. Deploying Home Cooked Apps with Railssoftware
  68. Repetitive Copypromptingwriting
  69. Questions to ask every decadejournalling
  70. Balancing work, time and focusproductivity
  71. Hyperlinks are like cashew nutswriting
  72. Brand treatments, Design Systems, Vibesdesign
  73. How to spot human writing on the internetwriting
  74. Can a thought be an algorithm?product
  75. Opportunity Harvestingcareers
  76. How does AI affect UI?design
  77. Everything is a prioritisation problemproduct-management
  78. How I do product roastsproduct
  79. The Modern Startup Stacksoftware
  80. In-person vision transmissionproduct
  81. How might we help children invent for social good?social-design
  82. The meeting before the meetingmeetings
  83. Design that's so bad it's actually gooddesign
  84. Lessons learnt interview prepping for product rolesinterviewing
  85. Obsessing over personal websitessoftware
  86. English is the hot new programming languagesoftware
  87. Better way to think about conflictsconflict-management
  88. The role of taste in building productsdesign
  89. Dear enterprises, we're tired of your subscriptionssoftware
  90. Products need not be user centereddesign
  91. World's most ancient public health problemsoftware
  92. Pluginisation of Modern Softwaredesign
  93. Let's make every work 'strategic'consulting
  94. Making Nielsen's heuristics more digestibledesign
  95. Startups are a fertile ground for risk takingentrepreneurship
  96. Insights are not just a salad of factsdesign
  97. Minimum Lovable Productproduct
  98. Methods are lifejackets not straight jacketsmethodology
  99. How to arrive at on-brand colours?design
  100. Minto principle for writing memoswriting
  101. Importance of Whytask-management
  102. Quality Ideas Trump Executionsoftware
  103. Why I prefer indie softwareslifestyle
  104. Use code only if no code failscode
  105. Self Marketing
  106. Personal Observation Techniquesdesign
  107. Design is a confusing worddesign
  108. A Primer to Service Design Blueprintsdesign
  109. Rapid Journey Prototypingdesign
  110. Visualise detailed file structures on CLIcli
  111. Do's and Don'ts of User Researchdesign
  112. Design Manifestodesign
  113. Complex project management for productproducts
  114. How might we enable patients and caregivers to overcome preventable health conditions?digital-health
  115. Pedagogy of the Uncharted — What for, and Where to?education
  116. Future of Ageing with Mehdi Yacoubiinterviewing
  117. Future of Tacit knowledge with Celeste Volpiinterviewing
  118. Future of Rural Innovation with Thabiso Blak Mashabainterviewing
  119. Future of Equity with Ludovick Petersinterviewing
  120. Future of work with Laetitia Vitaudinterviewing
  121. Future of Mental Health with Kavya Raointerviewing
  122. Future of unschooling with Che Vanniinterviewing
  123. How might we prevent acquired infections in hospitals?digital-health
  124. The why to endure any howentrepreneurship
  125. Design education amidst social tribulationsdesign
  126. How might we assist deafblind runners to navigate?social-design