Journal · 8 essays

Essays & build logs

Longer thinking on systems and craft.

Some ideas need evidence and room, so I work them out here at full length — systems, AI, and the craft of building. If you want the short version, Notes lands one argument every Sunday evening.

Long-form

Essays + build logs

Short-form

Notes, every Sunday

Start reading Try the Sunday letter

Essays ↓Get Notes

Journal — open ledger

01GPT-5.6 Sol: Frontier Intelligence at 67% Off and Half the TokensAugust 26, 2026 · 18 min read 02Why Short-Form Video Is Quietly Eroding the Next Generation's CognitionAugust 20, 2026 · 6 min read 03Cursor Origin: Git Forge for AgentsAugust 18, 2026 · 8 min read

EducationAI ArchitectureProductEngineeringDesign

Long-form desk

Searchable archive

8 Essays

In the journal

10 Lanes

Thesis to craft

Long-form

No hot takes

GPT-5.6 Sol: Frontier Intelligence at 67% Off and Half the Tokens

GPT-5.6 Sol now costs $2/$10 via OpenRouter, delivering frontier intelligence with half the tokens and time of rivals.

Gargeya Sharma

August 26, 2026 18 min read

GPT-5.6 Sol: Frontier Intelligence at 67% Off and Half the Tokens

Why Short-Form Video Is Quietly Eroding the Next Generation's Cognition

Heavy short-form video use suppresses cognitive control and erodes students' attention, resilience, and long-term learning capacity.

Gargeya Sharma

August 20, 20266 min read

AI ArchitectureProductEngineering

Cursor Origin: Git Forge for Agents

Origin lets AI agents edit code, open PRs, and review changes without leaving the editor.

Gargeya Sharma

August 18, 20268 min read

EngineeringOpen SourceAI Architecture

GLM-5.3: Pure Post-Training Scaling Takes the Open-Weight Lead

Same 743B base, scaled post-training only — GLM-5.3 now leads open-weight models on coding and cybersecurity.

Gargeya Sharma

August 14, 20264 min read

ProductEngineeringAI Architecture

Gemini 3.7 Flash: Google's Intelligent Workhorse Gets Faster and Cheaper

Gemini 3.7 Flash boosts coding and agentic performance at 340 tok/s with half-price introductory pricing.

Gargeya Sharma

August 14, 20266 min read

ProductAI ArchitectureEngineering

Grok 4.6 Redefines the Frontier Without Raising the Price

Grok 4.6 hits 61 on the Intelligence Index with elite agentic gains at the same $2/$6 pricing as Grok 4.5.

Gargeya Sharma

August 13, 20263 min read

ProductModel ReleaseAI & LLMs

DeepSeek V4-Flash 0731: The Moment the AI Landscape Shifted

DeepSeek's V4-Flash 0731 model, post‑trained on a 13B MoE, beats larger models on agent benchmarks at a fraction of the cost.

Gargeya Sharma

July 31, 20265 min read

ProductModel ReleaseAI & LLMs

Gemini 3.6 Flash Underwhelming

Gemini 3.6 Flash boosts token efficiency and cuts output cost, but stays pricier and weaker than Grok 4.5 and GPT-5.6 Luna on overall intelligence…

Gargeya Sharma

July 21, 20265 min read

Showing 8 of 8 posts

Longer thinking on systems and craft.

GPT-5.6 Sol: Frontier Intelligence at 67% Off and Half the Tokens

Why Short-Form Video Is Quietly Eroding the Next Generation's Cognition

Cursor Origin: Git Forge for Agents

GLM-5.3: Pure Post-Training Scaling Takes the Open-Weight Lead

Gemini 3.7 Flash: Google's Intelligent Workhorse Gets Faster and Cheaper

Grok 4.6 Redefines the Frontier Without Raising the Price

DeepSeek V4-Flash 0731: The Moment the AI Landscape Shifted

Gemini 3.6 Flash Underwhelming

Prefer the short version?