ESSAYS · 35 IN TOTAL
Essays
Long-form analysis on AI capability, market dynamics, and the state of the discourse. Measured voice; named sources; explicit uncertainty.
Hype vs Reality
GPT-5 release: capability deltas vs the narrative
Measured comparison of what shipped against the pre-release framing — and why the "phase transition" rhetoric mostly didn't survive contact with the benchmarks.
Model Comparisons
Reasoning models — o1 → o3 → DeepSeek R1 → Claude Opus 4.x thinking
What's actually new in the reasoning-model wave, where the capability ceilings sit, and which benchmarks are starting to get gamed.
Hype vs Reality
Agentic coding: Cursor, Devin, Claude Code, Replit Agent — adoption data vs marketing decks
Where the published adoption metrics actually land for each agentic coding product, and what gets quietly conflated when vendors talk "AI software engineer."
Industry & Investment
The DeepSeek pressure: have inference prices actually collapsed?
Three months after the price-war narrative crystallized, what's happened to enterprise inference economics — and what the frontier labs' price-card revisions actually reveal.
Technical Deep Dives
SWE-bench is broken: how coding evals get gamed and what replaces them
How the canonical agentic-coding benchmark is being optimized against, the Anthropic eval-paper findings, and what credible coding-eval looks like from 2026 onward.
Hype vs Reality
AI productivity papers: Goldman, MIT, BCG — what they actually show and don't
The three most-cited 2024-2026 papers on AI productivity contribution, the methodological caveats their summaries skip, and what would constitute durable productivity evidence.
Industry & Investment
The capex-revenue gap: $200B AI spend vs ~$40B AI revenue (2025)
Sequoia, Stripe, and the FT have all run the math on the 2025 AI capex-revenue divergence. The numbers are not seriously disputed — what they imply is.
Model Comparisons
Open-weight momentum: Llama 4, Qwen 3, DeepSeek V3 — share-eating?
Open-weight model adoption metrics from HuggingFace, Together, and Fireworks: where the closed-vs-open share is genuinely moving and where the narrative outruns the data.
Hype vs Reality
AI agents in the enterprise: ROI signals at the 18-month mark
The 2024-deployment cohort of enterprise AI agents is now hitting 18 months in production. What the Gartner / IDC / a16z surveys actually show — and where they're self-selected.
Hype vs Reality
The AI bubble question, 2026 edition
A measured walk through the 2026 state of the bubble debate — capex, revenue, valuations, capability deltas, alternative-cycle comparisons — without taking a side.
Company Profiles
OpenAI’s trajectory: funding rounds, product velocity, and the competitive chessboard (2024–2026)
How capital structure, enterprise adoption, and frontier model releases shaped OpenAI’s path—and what rivals, regulators, and customers should watch next.
Company Profiles
Anthropic, Constitutional AI, and the enterprise bet on steerability
How Anthropic frames alignment as a product feature, why enterprises care about refusals and long-context workflows, and where Claude fits in the competitive stack.
Company Profiles
Google DeepMind and Gemini: integration promise, product friction, and enterprise reality
Why combining frontier research with Google-scale distribution creates unique coordination challenges—and what buyers should validate beyond benchmarks.
Technical Deep Dives
RAG patterns for enterprise AI: retrieval architecture, failure modes, and production-grade guardrails
How retrieval-augmented generation actually ships inside companies—from chunking and embeddings to hybrid search, access control, and the prompt-injection battleground.
Hype vs Reality
Job displacement versus augmentation: how to read the labor-market debate in the LLM era
Economists, founders, and workers disagree on whether AI will mostly replace jobs or amplify them. We map the evidence, the mechanisms, and what employers should plan for between 2024 and 2030.
Model Comparisons
Open weights versus closed APIs: the real tradeoffs behind the AI deployment debate
A sober look at transparency, safety liability, operational burden, and enterprise procurement when choosing between downloadable models and hosted frontier APIs.
Industry & Investment
The AI chip market in motion: NVIDIA’s lead, AMD’s challenge, and the rise of custom silicon (2024–2026)
How accelerator economics, software ecosystems, and hyperscaler-designed ASICs are reshaping who captures value in AI training and inference—and what buyers should expect next.
Industry & Investment
China’s AI strategy and capabilities: an assessment of talent, data, chips, and deployment constraints
A balanced look at how China’s national AI agenda, industrial base, and market scale interact with semiconductor limits, export controls, and internal regulatory priorities.
Model Comparisons
GPT-4, Claude 3, Gemini Ultra, and Llama 3: what benchmarks actually measure—and what they miss
A practitioner’s guide to comparing frontier models across reasoning, coding, multimodal tasks, and safety—without mistaking leaderboard scores for product fit.
Company Profiles
xAI and Tesla under Elon Musk: ambitious AI claims, execution pressure, and the delivery gap
An editorial analysis of how xAI’s Grok roadmap and Tesla’s autonomy and robotics narratives intersect—what has shipped, what remains contested, and how investors and buyers should read the hype cycle.
Policy & Regulation
U.S. executive orders and agency guidance on AI: a map of federal signals for builders and buyers
How White House directives and U.S. regulator guidance shaped AI governance, procurement, safety expectations, and sector-specific compliance from 2023 through 2026.
Policy & Regulation
AI safety institutes: research agendas, oversight mechanisms, and the tension with commercial pace (2024–2026)
How national AI safety bodies are shaping evaluations, standards, and information-sharing—and what enterprises should expect as policy intersects with frontier model deployment.
Hype vs Reality
‘AI will solve healthcare, law, and coding’: a sector-by-sector reality check
Headlines promise end-to-end automation of medicine, legal practice, and software engineering. Here is what actually changes first—workflow, liability, incentives—and what stubbornly remains human, professionally and ethically.
Technical Deep Dives
RLHF and modern alignment techniques: reward modeling, preference optimization, and what ‘helpful’ really costs
From classical reinforcement learning from human feedback to DPO, constitutional training, and critique-based pipelines—how alignment layers shape model behavior and where the field is heading.
Industry & Investment
Startup valuations meet revenue: a reality check on AI company multiples, margins, and sustainability
Why AI startups trade on different fundamentals than classic SaaS, how inference costs distort unit economics, and what investors and founders should scrutinize before believing the sticker price.
Hype vs Reality
AGI timelines: expert predictions, survey evidence, and how to read them without losing your mind
From Metaculus forecasts to lab roadmaps, we unpack what people mean by AGI, why timeline estimates diverge by decades, and how to translate prediction markets into planning—not prophecy.
Technical Deep Dives
Transformer architecture evolution (2017–2024): from 'Attention Is All You Need' to trillion-parameter stacks
A technical tour of how the original Transformer blueprint became the substrate for GPT-class models, efficiency innovations, and the engineering tradeoffs that define modern LLM stacks.
Policy & Regulation
Export controls on AI chips: geopolitics, compliance, and second-order effects for the global market (2024–2026)
How semiconductor restrictions reshape cloud geography, startup strategy, and enterprise procurement—and why compliance is only the entry fee to a much larger strategic puzzle.
Policy & Regulation
EU AI Act: a practical requirements and compliance guide for teams building and deploying AI systems
What the European Union’s Artificial Intelligence Act means for providers, deployers, and downstream users—risk tiers, documentation, conformity, and operational steps through 2026.
Industry & Investment
The Big Tech AI arms race: a structured comparison of strategies across Microsoft, Google, Amazon, Meta, and Apple
How hyperscalers and platform giants are betting on foundation models, cloud distribution, open weights, and on-device intelligence—and where their incentives align or collide.
Technical Deep Dives
Model quantization and deployment optimization: INT8, GPTQ, and the inference economics of production LLMs
How post-training quantization, hardware-aware kernels, and serving strategies shrink latency and cost—without pretending precision loss is free.
Hype vs Reality
Is another AI winter coming? Definitions, funding cycles, and what would actually freeze progress
After years of headline breakthroughs, skeptics ask whether hype outran fundamentals. We dissect the ‘AI winter’ concept, compare past busts to today’s compute-and-data regime, and outline plausible slowdown scenarios through 2026.
Policy & Regulation
Copyright, training data, and generative AI: an analysis of lawsuits, doctrines, and what builders should expect next
How courts and regulators approached copying, fair use, licensing, and opt-out regimes for web-scale training—plus practical implications for model developers and enterprises through 2026.
Hype vs Reality
Enterprise AI adoption: an ROI reality check beyond pilots and press releases (2024–2026)
Why productivity gains from generative AI are uneven, how hidden costs erode returns, and what disciplined measurement looks like for leaders who want durable outcomes—not slide-deck optimism.
Industry & Investment
VC funding trends in AI (2020–2024): waves, valuations, and what changed after generative AI went mainstream
A data-grounded tour of venture capital flows into AI from the pre-LLM era through the generative boom—what drove rounds, how valuations behaved, and which patterns look durable versus cyclical.