Knowledge Base
Deep dives, tutorials, and research notes on the things I build with — AI, the web platform, accessibility, and system design.
Before asking whether AI will replace humans, ask how much human intelligence we've been failing to see all along. Hidden Figures is a story about invisible intelligence — and as AI automates the visible layer of work, human value moves upstream to framing questions, judging answers, and giving numbers meaning.
Engineers are trained to solve problems inside constraints. Science fiction trains the ability to imagine what happens when the constraints themselves change. As AI makes ideas cheap, richer mental models and stranger inputs become the differentiator — science fiction isn't escapism, it's mental infrastructure.
We grew up in a culture where effort wasn't just a means to an outcome — it was evidence of character. AI makes effort invisible, and that collides with how we've learned to measure merit. But productivity is not a suffering contest: stop treating effort as the product, and start treating it as an investment.
India has no shortage of AI ambition — the harder question is whether it reaches deep enough. An Indian app built on someone else's model, cloud, chips, and tooling is still a foreign dependency. Real ecosystem independence means building down the stack as well as up: compute, models, tooling, and orchestration.
The biggest weakness in an AI system may not be failing a test — it may be meeting a situation nobody thought to test. Testing is becoming the rehearsal, not the performance. Production AI needs runtime evaluation, dynamic adaptation, and bounded self-healing: the ability to notice when it should stop trusting itself.
A voice AI can be intelligent, natural, and accurate — and still feel like a machine. Humans are extraordinarily sensitive to conversational timing, and a chain of streaming systems has to behave like one organism to keep silence from turning into doubt. Optimize perceived latency, not just measured latency.
The hardest AI problems are no longer about writing the perfect prompt. After context engineering comes harness engineering — treating the model as one component inside a runtime built for evaluation, observability, orchestration, and continuous adaptation. Reliability is a property of the system, not the model.
As AI drives down the cost of producing code, coding is moving down the value chain. The scarce resource is shifting to knowing what should be built, why it matters, and how the business actually works — and people with deep domain knowledge just gained unusual leverage.
“Why is engineering taking so long?” is often the wrong question. Behind most slow teams is reliability debt, compounding complexity, and MVP shortcuts that borrowed from the future — and good systems engineering often looks like slowness right before it creates speed.
The hardest part of AI adoption is rarely the AI. Organizations train individuals on prompts and copilots, then send them back into old approval chains, incentives, and identities. Real AI upskilling is a change-management, workflow, and human acceptance problem — the technical learning is the smallest piece.
Machines aren't just getting smarter — they're getting narrower, deeper, and extraordinarily good at specific things. Meanwhile the most valuable humans are moving the other way: broader, more curious, and better at connecting ideas from fields that were never supposed to meet.
The most expensive mistake in enterprise AI may not be building the wrong agent — it may be building too many agents without designing the system that tells them how to work together. Building an agent is getting easier. Orchestrating a network of them is becoming the real discipline.
AI agents don't need motivation, praise, or promotions — they need clarity. As managers start orchestrating swarms of specialized agents, the core leadership skill shifts from managing capacity to designing intelligence: turning an ambiguous goal into a work packet another intelligence can execute exceptionally well.
AI is compressing execution from days into hours, and when everyone can produce more, production becomes a weaker signal of value. The bottleneck has moved — to the very human work of finding meaning in complexity, developing taste, and telling the story that makes people care enough to act.
When the easiest way to repair a creation is to throw it away and regenerate it, our relationship with the things we make starts to change. Ownership is shifting from “I made this” to “I can produce this” — and the scarce resource may no longer be the ability to create, but the ability to care.
Give a machine a personality, a memory, a goal, and enough autonomy, and you stop using software and start negotiating with something that feels like a mind. AI agents have no human psychology — but their behavioral patterns are real enough to profile, predict, and manipulate.
For decades, an idea and a shipped product were separated by a painful canyon of skill, capital, time, and permission. AI is quietly collapsing that gap — and the new bottleneck is no longer technical capability. It's decisiveness.
Engineers who once celebrated automation are suddenly suspicious of AI coding — “did you really build it?” We asked the same about Python, about C, about compilers. What we call craftsmanship is often just familiarity wrapped in nostalgia.
Three years ago engineers asked which JavaScript framework to bet on. Today it's which AI stack will still matter in six months. Different nouns, same anxiety — and the exhaustion is no longer technical. It's existential.
AI mutates in public — a tool you mastered in three weeks gets replaced over a weekend. But the real problem isn't intelligence, it's cognitive bandwidth. You don't need to know everything about AI to benefit enormously from it. Trying to is the fastest way to learn nothing deeply.
Smart people are starting to confuse speed with wisdom. You can generate a strategy deck in twenty minutes and ship code before lunch — but synthesis is not understanding. You can outsource thinking; you cannot outsource the part where meaning gets made.
The most valuable person in your life is rarely the smartest one in the room — it's the one who gets you before you finish the sentence. Intelligence has become abundant. Understanding has not. So the better question isn't “How do I compete with machines?” but “Am I becoming someone who truly understands people?”
Showing 22 of 22 entries.
