Archive
93 articles published in 2026.
WWDC 2026 made Apple's AI stack a cheap, private backend for small developers. This article maps the models, PCC quotas, data-access walls, and why context architecture is the new moat.
A conceptual tour of the modern AI stack organized into nine layers, from compute hardware and base models through alignment, inference, retrieval, orchestration, tool protocols, guardrails, and end-user applications.
On algorithmic platforms, distribution is earned by predicted engagement, not by human-judged quality. This article explains why the reward function favors loud content and what quality creators can do about it.
Examines how follower inequality and algorithmic seeding create a Matthew effect in creator markets, making reach itself the scarce resource and leaving quality content without distribution.
Uses the history of the penny press, radio, television, and early internet to show that attention markets follow a recurring pattern, and that the current feed economy is the latest version of an old bargain.
Shows how fragmented attention and shallow work early in a career create compounding costs, while small habits of depth build skill and optionality that shape later career stages and family life.
Most of India's digital advertising spending flows to a pair of foreign platforms, while the creator economy is sharply stratified. This article traces the money using regulatory filings and industry research, and asks what a more balanced attention economy would require.
Compares India's current startup and innovation portfolio with the deep-tech and AI infrastructure it could prioritize.
Generative AI lowers the cost of writing, coding, translating, designing, tutoring, and many knowledge tasks. This article maps what is becoming cheap, what India is building, and why the payoff depends on whether reclaimed attention flows into learning or back into extraction.
Explains why users rarely leave harmful platforms even when alternatives exist, and what interoperability and federation could change.
Examines the design incentives and evidence behind divisive-content amplification in India, from algorithmic feeds to misinformation and hate-speech trends, and asks whether platforms can be designed to reward cohesion instead of fracture.
Surveys public-health precedents to argue that attention-extraction platforms can be treated as design problems, not only user-responsibility problems.
Practical strategies for working adults to reclaim commute and downtime for small acts of learning and creation.
Offers practical guidance for students using AI: active learning, public note-taking, peer teaching, and avoiding the answer-generation trap.
A diagnosis of student screen use in India, drawing on ASER 2024 and NCERT 2022. The same device could support learning but is often captured by entertainment and social feeds, with measurable costs to attention, sleep, and mental well-being.
Examines the design mechanics and usage scale of short-form video in India, separating documented harms from hype and tracing the opportunity cost.
Uses public surveys and academic sources to show that the attention economy imposes different but connected costs at each life stage, from adolescent screen ratios to elderly digital exclusion.
Traces how Jio's 2016 launch collapsed data prices, expanded access, and reshaped Indian internet behavior toward entertainment and short-form video.
India's internet is now overwhelmingly Indic-language, rural, and mobile-first. This article examines how language expansion multiplied the audience, what vernacular feeds look like at scale, and why engagement-maximizing designs operate across languages with fewer guardrails.
Uses the Green Revolution as an analogy for solving one urgent problem while creating slower, structural costs that only become visible later.
Offers household-level strategies for managing screens: device-free meals, modeling, shared creative time, and age-appropriate boundaries.
Examines Gallup engagement data, workplace distraction research, and Indian graduate-employability figures to show how attention extraction affects productivity and the country's demographic dividend.
Catalogues the interaction-design patterns that keep users in feeds longer than intended, and why each one is a choice, not an inevitable feature of digital life.
Connects India's youth demographics to employment and skill data, arguing that the demographic dividend depends on how young people invest their attention.
India's creator economy is reaching massive scale, yet few monetize well and income concentrates at the top. This article examines the gap between aspiration and livelihood, and asks what the AI moment means for young Indians betting on viral attention.
Reframes India's AI opportunity as a compounding function of attention. Uses Gallup, NASSCOM, and public AI-readiness data to show how small shifts in daily learning, work, and creation habits could aggregate into large differences in national capacity over the next decade.
Shows how ordinary citizens can use AI to produce local knowledge and help neighbors, moving from online consumption to offline civic contribution.
Synthesizes the Designing for Substance arc into a practical north star: platform success should be measured by whether users leave with more clarity, skill, or civic capacity than they had when they arrived.
Interprets Tele-MANAS call data and adolescent mental-health surveys as a public-health signal, not moral failure, and connects them to attention-extraction design.
Compares India’s IT Rules 2021 and DPDP Act 2023 with the EU DSA, UK OSA, and Australia’s eSafety framework. Regulation can curb harms but cannot alone make substance the easier path; design, business models, and public pressure matter too.
Uses observational and survey evidence to describe how phones and headphones reshape public space in India, and what incidental conversation, learning, and solidarity are displaced.
Maps the ecosystem of non-regulatory accountability: investigative reporting, internal leaks, shareholder pressure, civil-society campaigns, and academic research.
Surveys product and platform alternatives that could shift attention incentives from extraction to substance, evaluates build-new versus change-existing strategies, and proposes five differentiated product concepts with rationale and risks.
Honest uncertainty about the attention economy—causality, culture, regulation, and the choices that will shape India's next decade.
Compares India with Brazil, Indonesia, Nigeria, the US, and China on screen time, platform concentration, and regulatory responses.
Uses historical technology transitions to show that access alone does not determine benefit; the distribution of gains depends on institutions, literacy, and norms.
Uses cross-country historical comparisons to show that missing a technology transition has persistent economic and geopolitical consequences.
Uses government surveys and industry data to examine how women experience India's attention economy differently—through access gaps, unpaid care burdens, online safety risks, and unequal creator-economy opportunities.
Surveys evidence on design interventions that reduce extraction: chronological feeds, friction before sharing, and user control over ranking algorithms.
Argues that early failure and public teaching are features, not bugs, of building substance over time.
Explores how engagement-based ranking and ad-supported business models are design choices, not natural laws. Cites evidence that alternative metrics can reduce divisive amplification and asks what it would take to make substance the default.
Uses government, academic, and industry sources to map India's digital time budget across entertainment, education, work, and communication.
Compares revenue models and their incentive effects, arguing that business-model diversity is a precondition for design diversity.
How Indic-language AI could turn India's language diversity from a barrier into a creative force for substantive content.
A standalone essay on how entertainment-driven screen use is wasting the most precious resource you have — your own attention — and how the same device, used differently, could make you more capable with less effort than you think.
A guide to the Attention, Substance, and the AI Moment series: the diagnosis, the generational stakes, individual practice, systemic design, and curated reading paths.
Surveys proposed alternatives to engagement-based metrics and their tradeoffs: time well spent, learning outcomes, civic value, and sustainability.
Surveys how AI can amplify attention extraction through synthetic media, hyper-personalization, and automated influence, and argues for guardrails before scale.
Practical guidance on treating AI as a collaborator that drafts, translates, tutors, and debugs, while keeping the human learner in the driver's seat.
Argues for age-appropriate design defaults, friction, and verification to protect children, elderly users, and others who are especially vulnerable to extraction.
Helps readers choose where to start in the series based on their question or role, explains the six arcs, and points only to published articles.
This synthesis article maps the actors and interventions that can redirect India's digital infrastructure toward substance—from individual habits and household rules to platform design, investor pressure, regulation, and civic accountability.
Building substance is not about dramatic discipline or quitting the internet. It is about redirecting small windows of attention toward creation, learning, and teaching, using AI to lower the activation energy of starting.
India has the youngest large population on Earth, deep software talent, and cheap AI-assisted tools. The open question is whether the next few years become a compounding bet on building, or a continuation of the attention-economy habit of consuming.
A diagnosis of how India's digital-access success has inverted into an attention-extraction economy, using public government, academic, and industry sources to show the costs across attention, productivity, youth mental health, and social trust.
An exploration of how platform incentives could be redesigned to reward substance over extraction, from regulation and metrics to product alternatives like the fediverse.
A plain-language explanation of how AI tool use extends what a model can do by connecting it to external capabilities, with examples, limits, and an anti-hype check.
A plain-language guide to retrieval-augmented generation: what it is, when it helps, why it sometimes fails, and what older ideas it builds on.
A plain-language explanation of reasoning models: how they use extra computation to work through problems step by step, and where the real limits lie.
A plain-language guide to prompt engineering: how clear instructions, examples, and constraints shape AI outputs, and why it is design rather than magic.
A plain-language guide to prompt caching: what it reuses, why providers offer it, where the savings are real, and what builders should check before relying on it.
A plain-language explanation of planning and reflection in AI agents, showing how systems break work into steps, check their own output, and revise before continuing.
A plain-language explanation of multi-agent systems: how multiple AI workers are assigned different roles, how they coordinate, and where the design tradeoffs really matter.
A plain-language explanation of the difference between context and memory in AI systems, with everyday analogies, practical examples, and clear limits.
A plain-language explanation of why AI agents need both loops and goals, with everyday analogies, practical examples, and clear limits.
A plain-language guide to what makes AI sessions stay coherent across long tasks, where they drift, and how to keep them on track.
A plain-language guide to fine-tuning: what it changes, how it differs from prompting and retrieval, where it helps, and where the hype overpromises.
A plain-language guide to AI evaluations: what they measure, how to design them, and why a good score does not always mean a useful system.
A plain-language guide to context management: how language models choose what goes into their working window, why it matters, and where the limits lie.
The guide article for the AI, De-Mystified series, introducing the series promise, article order, and how to read the articles.
A plain-language guide to what AI agents are, how they combine goals, tools, loops, and memory, and where the current hype overstates their autonomy.
Shows how delegation design changes when AI output may affect rights, money, legal duties, education, public records, or institutional accountability.
Refocuses agent networks around replaceable capability contracts rather than human job titles or org-chart theater.
Defines checkpoints, self-remediation, interruption quality, budgets, rollback, and stop conditions for long-running AI delegations.
Replaces human-org mimicry with explicit control loci for routing uncertainty in agent-native systems.
Argues that operators need control routing across delegations, not only traces, summaries, dashboards, or activity feeds.
Defines the delegation record and shows why it is more operational than a transcript, summary, ticket, or pull request alone.
Explains why conversation should remain the interface while delegation becomes the durable work primitive for consequential AI workflows.
A guide to the seven-part AI Delegation Orchestration series, covering durable agent work from conversation thresholds to high-stakes commitment boundaries.
This article proposal explores whether voice-first, two-way audio agents can become a legitimate assistive medium for knowledge workers facing screen fatigue.
A short, practical onboarding recipe that lets non-technical adults and teens start using AI agents by copying one prompt into any capable model, with worked examples for explaining a utility bill and turning meeting notes into summary, email, and action items.
A practical follow-up for non-technical readers who have tried an AI agent once or twice. It covers privacy limits, recovering from wrong answers, trust, better prompts, small automations, choosing a model, and five safe practice conversations.
This article argues that agent orchestration is evolving from hand-written workflows into a governed control-plane layer that routes across models, tools, memory, evaluators, policies, and execution environments.
A short, source-backed overview of The Long Human Road to AI Season 1, showing how computers and AI emerged from older human patterns and what readers will learn across seven articles and this overview.
A general-reader history of the learning turn in AI, from Samuel's checkers and Rosenblatt's perceptron to ImageNet and AlexNet, with caveats about generalization and understanding.
AI systems are social arrangements, not just technical artifacts. This article explores labor, governance, education, access, and trust as the human systems that shape what AI becomes and who benefits.
Foundation models revived the ambition of general-purpose AI. This article traces the transformer, pretraining, scaling, post-training, multimodality, and tool use—and why broad capability is not reliable understanding.
A readable walk from Boole and Frege through computability, switching circuits, information theory, and cybernetics, showing how formal ideas made later computing and AI legible.
How the 1956 Dartmouth workshop named and organized artificial intelligence, what early symbolic systems actually demonstrated, and why the era's optimism both helped and overpromised.
A narrative history of calculation before electronics: human computers, abaci, Napier's rods, mechanical calculators, automata, Jacquard cards, and Babbage's engines.
A history of AI winters and expert systems shows that intelligence claims survive only when they meet grounded tests, maintenance plans, and institution-aware deployment criteria.
A possibility thesis on how AI shopping agents could weaken passive brand loyalty, increase product exploration, and create demand for privacy-preserving, adversarially tested product assurance infrastructure.
A first-principles argument for publishing essays as human-readable narratives backed by portable, agent-auditable research bundles.