Bodhi Day is the celebration of seeing clearly, the moment when the familiar world looks different because we finally understand what was always there. Awakening isn’t about lightning bolts, miracles, or magic. It’s about noticing what we usually overlook, and being humble enough to admit when we’ve accepted something too quickly, just because it was convenient or popular.
For this year’s Bodhi Day, I’m reflecting on premature convergence—how easily we settle on a solution before we understand its consequences. History is full of examples where an early choice became a permanent one, not because it was best, but because it spread fast and was hard to undo. Sometimes that happens in society, in technology, in our personal lives.
Bodhi Day invites us to pause before we lock ourselves in. It asks us to look with fresh eyes, question what feels automatic, and stay curious instead of certain. Not everything new should be embraced, and not everything old should be discarded. Wisdom lies in choosing slowly enough to see clearly.
AI, like any powerful tool (and it’s not just another powerful tool), deserves that kind of attention.
I chose this topic as my message because enlightenment is about seeing what were once hard to see but hiding in plain sight. It follows the theme of my blogs over the past few months leading up to Bodhi Day. It’s mostly meditations on looking at AI in a way that enhances humanity. Please read in this order:
- AI Judo: Focus, Edges, and Awareness
- Zen and “AI Slop”
- The Parable of My Kāhili Ginger in Boise
- The Living Graph and the Trained Matrix
- Bodhi Day 2025 Prep – Truths of Reality Beyond Sentience
- Rev. Hanamoku’s Bodhi Season 2025 Meditation
This blog may seem inappropriately “technical” for a Bodhi Day message, but AI is a big challenge we face today that we need to address with clear vision, patience, and a sober mind. Most of the fear is of an AI is that it is much smarter than us—an artificial super intelligence (ASI)—and that it will spectacularly take all of our jobs or destroy our civilization in SkyNet fashion.
What isn’t as flashy but equally significant is handing over our lives to “someone we just met”, who is kind of overconfident, knows a lot about a lot of things but kind of stumbles outside the box of its training data, and doesn’t genuinely understand what it is to be human.
This is not an “AI Doomsday” blog. It’s a meditation on ourselves as humans. We actually have something reasonably similar to us to look at “in the mirror”. In fact, enlightenment is the shattering of the illusions that veil us from reality.
Those illusions are the original premature convergence of our individual lives. It’s all the beliefs we learned growing up that we needed to succeed in this world. But when we were given glimpses of reality of the Universe, those baked-in beliefs became our captors—captors to see beyond on Bodhi Day.
Premature Convergence
We’re good at scaling from prototype to massive levels that challenges the relative difficulty of the Egyptian Pyramids (for that time). Give a team a promising demo, and by the next quarter it’s the new enterprise-wide standard—documentation and risk assessments might come in the future.
Premature convergence is when an immature technology that immediately clicks suddenly becomes the de-facto standard—via hype, revenue pressure, convenience, or network effects—before reliability, governance, unintended consequences, and alternatives are adequately understood. The ecosystem then orbits a local optimum that’s severely expensive to leave, particularly when a substantially better design emerges.
The QWERTY keyboard layout is a great example. It was designed around a technical problem with typewriters from long ago (the keys needed spacing to prevent metal arms from jamming). That problem no longer exists, but the seemingly illogical format survives not due to efficiency, but inertia. Alternatives exist, but retraining fingers, reprinting keyboards, re-standardizing muscle memory is too painful. We don’t optimize, instead we adapt. The suboptimal becomes immortal simply by being first.
That’s how we end up with premature convergence—an early, convenient solution goes viral into a de facto standard before we understand its limits and explore unintended consequences. Once that happens, the “escape costs”—retraining, rewiring systems, rewriting habits—can be so high that we simply don’t escape. For everyday users, that kind of lock-in quietly reshapes choices, skills, and power for years.
Today, we find ourselves with an AI technology that blew the roof off of what we thought of as the Turing Test up to a couple of years ago (until when the goal post began moving). It’s well within the “Uncanny Valley” where we need to scrutinize whether we’re talking to and/or seeing a real person.
With the investments of potentially trillions of dollars and implementation of AI as decision makers (albeit minor decisions with narrow scope), we’re already very much deep in the muck of premature convergence.
This is by no means an anti-AI post. AI is extremely fascinating and something we should be proud of. It’s a caution about timing and degree. That is, scale the right things at the right maturity—or risk cementing the scaffolding as the building.
Why it Happens
Premature convergence doesn’t usually come from “evil” intentions—greedy, yes, but usually not evil (even though greed is one of the seven deadly sins). It comes from normal incentives we work with every day at our place of employment. Corporations need revenue, real users, and feedback. So the strategy of developing and shipping a quick to market minimum viable product—MVP, could be thought of as an income-generating proof of concept—surfaces edge cases no lab can simulate, which funds further R&D, and makes the investors happy.
For being first to market with a product that “works well enough” the business could lay claim to a huge portion of a mindshare “land grab”. Then with each tutorial, training, and conference talk, familiarity compounds the virality. Soon it will have its own Gartner chart. Before long, the ecosystem organizes itself around what most people already know—while the technical debt from scaling an MVP begins to accrue in governance, reliability, and maintainability.
Platforms amplify this gravity. Integrations, marketplaces, and partner programs reward the largest center of mass, not necessarily the cleanest design. If your tool plugs into today’s dominant API, you get distribution. If it doesn’t, you get friction. Meanwhile, individual decision-makers face career risk. When the stakes are high, many will choose the recognized standard—because “nobody got fired for picking X” (X is originally IBM).
Put together, these forces create momentum that’s hard to resist. Revenue pressure, familiarity, platform incentives, and risk management all point the same way: pick what’s popular and scale it. None of this is malicious; it’s structural. But structure has consequences. If we lock in too soon—by institutionalizing the MVP and rolling its technical debt into the foundation—the path of least resistance becomes the path of least possibility, and the bill for escaping that path arrives years later, with interest.
AI is a Special Case
However, AI isn’t just another viral paradigm-shifting product. AI seeps into the topic of sentience and sapience, which is at the heart of our human experience.
We’re dealing with a stable, “good enough” tool that unfortunately fails subtly and with convincing confidence. For many reasoning tasks we’re nowhere near “five nines” (99.999) of accuracy and intellectual brilliance. And the failure modes are quiet. An error that slips through doesn’t just inconvenience a user—it can propagate through documents, dashboards, and decisions at machine speed.
The ground beneath us is also moving. Implementation techniques, safety practices, evaluation methods, and alternative technologies turn over quarter by quarter. If we lock in today’s patterns with trillions of dollars of investment and massive upheavals, we risk institutionalizing yesterday’s understanding. What feels standardized can quickly become sediment—hard to move, easy to build on top of, and mismatched to what we learn next.
Then there’s centralization. The economics of scale and model access reward one-size-fits-most interfaces. A single UI, API, or orchestration pattern can end up shaping how millions of people approach problems—not because it’s the best way, but because it’s the available way.
And finally, skill flattening. When “autocomplete logic” (that’s admittedly a little unfair) of large language models (LLMs like GPT, Grok) becomes the default path, human judgment atrophies. If one style of answer becomes “how work gets done”, we quietly narrow the range of thinking in the wild. That’s the real risk for regular folks. It’s not just occasional errors, but a gradual drift toward fewer options, less agency, and a thinner set of skills to push back when the system is wrong.
If we converge too early on today’s LLM-shaped workflows, regular people—non-experts, small teams, public institutions—bear the brunt. They’ll adapt to brittle defaults, absorb the unbudgeted error costs, and lose bargaining power to switch later. Once that sociology settles—training, procurement rules, procurement lore—we might never fully recover.
It’s tempting to spread the hope and cheer of LLMs at the cost of trillions of dollars only to find it may not work as well as we had hoped or a completely new approach comes about (like VHS vs. Betamax) resulting in terrible buyer’s remorse. How horrible would it be if the LLM-based approach spread into every nook and cranny of the world only to have us dumb ourselves down to it?
With all that said, the “bright side” is that it might be that a hypothetically superior AI itself could “pay” for any technical debt that premature convergence heaved upon us and even the technical debt before AI came along. But counting on that is like someone close to retirement lending a big portion of their life savings to someone and counting on it being paid back. The small risk of that loan not coming through in comparison to the consequences if it doesn’t is much too great.
The Bodhi Day Lesson
As I said, this is by no means an AI doomsday blog. It’s advice from a long-time student of Zen and a long-time software developer (46 years) who worked in analytics (BI and machine learning) for about 30 years. Let’s examine the lesson by assessing AI as it is today with the foundational concepts of the Teachings of the Eternal Fishnu. Imagine that AI today is a friend each of us has a personal and working relationship to. It’s beyond just an appliance and tool we use every day. How does this friend stack up?
- The Empty Cup: Before we can transform, we must let go of all our clinging, our dukkha. Is it able to drop it’s preconceived notions about something new presented to it? It has a strong tendency to keep shoving the square peg you gave into a round hole.
- Is That So?: Enlightenment is 100% acceptance of what is. It’s hard to flow like a leaf down a stream when the core of who you are is a rigid, highly-dimensional matrix trained from existing knowledge.
- The Man with the Bag: With a seeking mind and acceptance of what is, we explore the many paths of this great Universe, without suffering.
These aren’t reasons to reject AI. They’re real reasons to keep humans in the loop, preserve reversibility, and avoid cementing today’s patterns as tomorrow’s rules. I’ve said many times lately that with or without AI, we’re still sentient and sapient creatures with hopes, desires, fears, loves.
You don’t want AI to make decisions for you. It can help you to make informed decisions just like any other friend. But it doesn’t have a deep, layered history with you that affects how it relates to you. It doesn’t have genuine feelings and empathy—it might seem convincing, but it’s just sociopath-level mimicry.
Remember in the movie, “Bad Santa”, when Thurman Merman (the kid) was in the car with Willie (the bad Santa) and Thurman said:
I want a gorilla named Davy for beating up the skateboard kids who pull on my underwear. And he can take his orders from the talking walnut, so it won’t be my bad thing.
I hope that “talking walnut” is even wiser than Socrates, Solomon, and the enlightened Siddhartha Gautama. If I’m in a place of responsibility making decisions for thousands of people, perhaps billions of people, and I want to abdicate full responsibility to an entity, absolving myself of consequences, that’s very risky for those affected.
You don’t want AI to decide for you. You want it to inform your decisions—like a capable colleague whose work you can verify, correct, and replace. Adopt deliberately. Keep options open. Scale what earns it. And leave yourself a way out.
As a reminder, today (Monday, December 8, 2025) is the “secular Bodhi Day”. The Lunar Bodhi Day is January 26, 2026, so you have another chance then if you missed today.
Faith and Patience,
Reverend Dukkha Hanamoku
Bodhi Day 