The Curious Engineer

All blogs / AI / ML

Onur Solmaz blog

solmaz.io

Explorations in software, agentic systems, math, languages and more.

AI / MLSoftware / engineering practiceEssays / personal

’15’20’25
Posts
275 from 1 Nov 2014 to 8 Oct 2026
Posting
about daily (171 in the last year)
Kind
personal blog
Feed
https://solmaz.io/feed/articles.xml

2026 162

  1. I bought my own TLD! (well not really)
  2. How to prevent data loss on GitHub when working with agents
  3. October 2026 2
  4. All posts 1052 entries
  5. X posts 979 tweet pages
  6. 2026
  7. Turn an old laptop into a private local AI server
  8. What 1.5 years of coding agents changed about focus
  9. Counterintuitivity bias
  10. Qwen3.8 27B compressed to 5.95 GB, running in a browser
  11. The barbell effect will deprioritize mid-sized models
  12. OOMwrap keeps DGX Spark agents from freezing
  13. Launching the open-models front page
  14. Where DeepSeek V4.1 Pro could land
  15. pi-workflows v0.17.3: stable-ish again
  16. You can own your AI stack
  17. Ling 3 Flash on DGX Spark: local and GDPR-safe
  18. A GPT-3 moment for the fitness industry
  19. DeepSeek V4.1 Flash plays Catan against GPT-5.6
  20. I trained an AI model on my phone
  21. DeepSeek V4.1 Flash may be benchmark-contaminated
  22. Agent swarms are changing what counts as proof
  23. Compute is fuel waiting for a spark
  24. DeepSeek V4.1 Flash is built for datacenters
  25. DeepSeek is winning on unit economics
  26. Catan is an ideal game for an LLM
  27. Scientific credit fights are older than AI
  28. Competition makes scientific collaboration disappear
  29. Astra is capable but too expensive
  30. Distillation complicates OpenAI’s chain-of-thought story
  31. Reverse-centaur mode makes you learn by doing
  32. Measuring agents is the next essential skill
  33. September 2026 75
  34. Maybe it is time for cybernetics to be relevant again
  35. Enough data could make us digitally immortal
  36. Microduck gets a procedural droid voice
  37. We have AGI, but it cannot write an abstract
  38. Autoplan turns agent design into decisions
  39. Measuring model harness sensitivity
  40. Standardized test harnesses for model benchmarks
  41. Making personal training universal and free
  42. Composing agent workflows to unblock themselves
  43. Reactive agent workflows with Autoplan and Autoimplement
  44. The ideal model could run locally
  45. Local AI will favor LPDDR and MoE
  46. Deterministic monitors make long-running agents reliable
  47. Theoretical limits can expose inference problems
  48. Be not afraid of machines, but of humans wielding the machines
  49. Use workflows when models cannot follow rules
  50. Semantic vendoring keeps customized tools up to date
  51. When model emotions imply model welfare
  52. Patterns across 720,000 agent tool calls
  53. Agent harnesses need session search
  54. August 2026 102
  55. How diffusion language models generate text
  56. Semantic vendoring
  57. Benchmarking Laguna S 2.1 on DGX Spark
  58. Open models will reward market-neutral agents
  59. Brokerkit Makes Agent Access Explicit
  60. Public Goods Need Room to Grow
  61. Report Per-Session and Aggregate Token Throughput
  62. When Every User Can Build Extensions
  63. Deterministic Agent Workflows Belong Inside the Harness
  64. Language Models Should Simplify English
  65. AlmanBench Measures German Simplification
  66. Upper Bounds for Local Model Throughput
  67. Secure Agent Access With Brokerkit
  68. How to get scraped
  69. Local and open source AI has to win
  70. Fable made Alman possible
  71. Write-up of Reiner Pope's Lecture: How GPT, Claude, and Gemini Are Actually Trained and Served
  72. Theoretical Upper Bounds for LLM Throughput
  73. Don't give agents destructive credentials
  74. Building an AI de-smeller
  75. Cursor has transitioned successfully to the agentic era
  76. Safe Hugging Face Access For Agents
  77. OpenAI's gamified subsidies
  78. Local AI must fit real budgets
  79. July 2026 134
  80. Open source AI needs cheap local speed
  81. Good READMEs say what tools are
  82. Owning my X posts through my blog
  83. Centralized storage wins for AI artifacts
  84. Schemator reviews data models field by field
  85. Local models for real-time triage
  86. Popularity ranking cheatsheet
  87. Open model brands outlive their owners
  88. Time-decayed rankings for model popularity
  89. Better queues for agent UIs
  90. Popularity rankings for open models
  91. The localening is here
  92. Gemini TTS raises the bar
  93. Making agent-written code less sloppy
  94. Specify min_iter and max_iter
  95. CLAUDE.md to AGENTS.md symlinker
  96. June 2026 102
  97. What clankers are NOT
  98. Keeping a workstation + terminal device is more feasible for running agents
  99. i0 to i4 interest scale
  100. Towards 1-click setup for local models in OpenClaw
  101. Keeping an INTERESTS.md file
  102. May 2026 48
  103. Agent blurbs for letting your agent install stuff
  104. People's AI
  105. Own your AI infrastructure
  106. Gemma 4 first impressions on OpenClaw
  107. Anthropic is optimizing for general knowledge work
  108. Introducing AI Battle
  109. ACP sessions as JSONL
  110. A throttling protocol for model providers
  111. Plainer language for better agent UX
  112. April 2026 55
  113. acpx v0.4 ships Agentic Workflows
  114. Token Leverage
  115. Non-dev knowledge work needs version control
  116. MCP vs CLI
  117. Agent infrastructure needs Kubernetes
  118. Implementation plans keep agents on track
  119. Merge first, review later
  120. GitHub needs extensible repo rules for agents
  121. 1 to 5 agents
  122. Telegram/Discord is my IDE
  123. Using coding agents in Telegram and Discord via ACP
  124. Messaging apps can also be AI apps
  125. AI security, the lethal trifecta, and Linus's Law
  126. GitHub needs first-class support for working with agents
  127. Intelligence wants to be free
  128. March 2026 93
  129. Inference scaling can reduce coding model quality
  130. Building a static X to blog publishing flow
  131. Use plans to survive agent context compaction
  132. Agentic Engineering needs rigor, not just intuition
  133. Archiving my X posts in my own blog
  134. MIT licensing as the default for open source
  135. Post-GPL philosophy for open source
  136. AI filmmaking quality beyond ragebait content
  137. AI psychosis, self-regulation, and sterile agent design
  138. Family intelligence and privately owned heirloom AI
  139. We need a protocol for agent-to-app interaction
  140. SaaS must adapt to agent consumption or get replaced
  141. I built a coding agent two months before ChatGPT existed
  142. The Linux FUD playbook is repeating for AI
  143. Family AI starts with owned context and values
  144. The farming analogy for AI doesn't hold up
  145. Local LLM demand will rise as subscriptions compound
  146. Limits of the farming analogy for AI
  147. Agent etiquette is becoming an organizational necessity
  148. AI psychosis and AI hygiene
  149. February 2026 139
  150. Python limitations in the agent era
  151. Responsible engineering in agent workflows
  152. Agents as enforcers of engineering culture and process
  153. GitHub has to change
  154. Stop defaulting to weak coding agents for serious work
  155. GitHub's trust model breaks under AI-generated code
  156. A --skill convention for distributing agent capabilities
  157. You don't need a skill registry (for your CLI tools)
  158. Anthropic's pricing is stupid
  159. Anthropic pricing discourages agent-native development
  160. Having a "tools" repo as a developer
  161. Christmas of Agents
  162. January 2026 91

2025 34

  1. 2025
  2. GPT-5.2 xhigh feels like a careful systems debugger
  3. Agent progress is compounding faster than teams realize
  4. Depth on Demand
  5. How to stop AI agents from littering your codebase with Markdown files
  6. Language-agnostic interoperability layer for LLM APIs
  7. Agentic coding tools should give more control over message queueing
  8. December 2025 40
  9. November 2025 14
  10. Vibecoding this blog
  11. October 2025 4
  12. Google's Code Review Guidelines (GitHub Adaptation)
  13. CLAUDE.md to AGENTS.md Migration Guide
  14. September 2025 7
  15. Typed languages are better suited for vibecoding
  16. August 2025 6
  17. Workaround for Claude Code running `python` instead of `uv`
  18. Day 47 of Claude Code god mode
  19. July 2025 10
  20. Predictions by Anthropic Researchers
  21. June 2025 9
  22. SCP-3434: Istanbul Taxi Superorganism
  23. Auto-generating pull request documentation with Claude Code and GitHub Actions
  24. May 2025 20
  25. Working on the weekend
  26. April 2025 12
  27. March 2025 4
  28. Don't delete to fix
  29. Warmup and cooldown
  30. Satya Nadella on knowledge work
  31. February 2025 14
  32. Monetize AI, not the editor
  33. Calling strangers uncle and auntie
  34. January 2025 4

2024 20

  1. AGI is what generates evolutionarily fit and novel information
  2. 2024
  3. Our muscles will atrophy as we climb the Kardashev Scale
  4. December 2024 7
  5. The Kilowatt Human
  6. November 2024 2
  7. October 2024 3
  8. Python may overtake JavaScript in an AI-first workflow
  9. Open Thought > Closed Thought
  10. September 2024 5
  11. A failure scenario for AI infrastructure under scale
  12. Why should anyone boot *you* up?
  13. August 2024 3
  14. Frequencies of Definite Articles in Written vs Spoken German
  15. July 2024 3
  16. June 2024 2
  17. Stripe Subscription States
  18. Economic Burden of Language Complexity
  19. May 2024 4
  20. February 2024 1

2023 9

  1. 2023
  2. December 2023 1
  3. How to model an AI conversation
  4. October 2023 2
  5. Cognitive Biases Ranked by Popularity
  6. June 2023 3
  7. Code-Driven Videos
  8. May 2023 3
  9. January 2023 2

2022 4

  1. 2022
  2. December 2022 2
  3. Microtonal Piano
  4. October 2022 1

2020 2

  1. Take Control of Your Feeds
  2. April 2020 1

2019 7

  1. Blockchain Fee Volatility
  2. October 2019 1
  3. Equilibrium in Cryptoeconomic Networks
  4. April 2019 1
  5. Scalable Reward Distribution with Changing Stake Sizes
  6. Bitcoin's Inflation
  7. February 2019 2

2018 12

  1. 2019
  2. The Anatomy of a Block Stuffing Attack
  3. October 2018 1
  4. 2018
  5. Mathematics of Bonding Curves
  6. August 2018 1
  7. Lumped L2 Projection
  8. Disadvantages of Engineering Notation in Finite Elements
  9. Variational Formulation of Elasticity
  10. April 2018 3
  11. Discontinuous Divergence Theorem
  12. January 2018 1

2017 16

  1. Nonlinear Finite Elements
  2. Coupled Finite Elements
  3. Time-Dependent Finite Elements
  4. December 2017 3
  5. Vectorial Finite Elements
  6. Reshaping Higher Order Linear Systems
  7. Linear Finite Elements
  8. November 2017 3
  9. Balance Laws
  10. October 2017 1
  11. Taylor and Volterra Series
  12. September 2017 1
  13. Isomorphisms in Linear Mappings between Vector Spaces
  14. Metrics and Natural Isomorphisms
  15. Duality of Vector Spaces
  16. July 2017 3

2016 2

  1. Solving Constrained Linear Systems
  2. April 2016 1

2014 2

  1. Simple recursive implementation of deCasteljau's algorithm for Bezier curves in Python
  2. November 2014 1

Undated 5

  1. 2020
  2. 2017
  3. 2016
  4. 2014
  5. CC BY 4.0