"Done, all tests pass." It was not done.
Tonight 9:30 PM IST, live: how I work with AI coding agents in Claude Code and OpenCode, and the checks that catch an agent grading its own work.
https://t.co/OgmvYQXy3r
Claude Code or OpenCode? Wrong first question.
Tomorrow, live: AI-first development with both tools, the same task side by side, and what it costs.
Sunday 9:30 PM IST: https://t.co/OgmvYQXy3r
133 bugs filed by my own apps against my own framework and libraries in five weeks. 31 releases. 0 I planned. A framework nobody builds on is a document. New post: Nothing Improves a Framework Like Building on It https://t.co/MaC6GW95o4
On Last Sunday's stream I found that four TfLens pages showed all repositories added together, no way to pick one. Said so on air, fixed it after, live today: a repo picker on each page. Correction note on the post (at bottom) with demo credentials : https://t.co/xOipOUtP9C
ΰ€Άΰ₯ΰ€ ΰ€§ΰ€¨ΰ€€ΰ₯ΰ€°ΰ€Έ β¨
May this Dhanteras bring health, wealth and new beginnings to you and your family.
Happy Dhanteras from my family to yours.
#Dhanteras2026#DhanterasWishes
@MladenMacanovic Well Astra is over-rated and so is sol, I have been personally using Claude models since over a year now and I shifted after using GPT models for a year and I still see Anthropic models are far better in coding than OpenAIs but Qwen really surprised me.
@cheaf25master Depends which language/framework you are coding in :) If it's .NET then no buy a windows laptop but if you have to build anything related to apple ecosystem then you defiantly need a Mac.
@MladenMacanovic Codex is not a model, it's a coding harness like claude Code or OpenCode. The model it support is GPT Astra, Sol, Lunna. The best coding model currently are from Anthropic Fable, Opus and Sonnet. But open source comes in second c my experiment results https://t.co/npHUSUR7Co
@MladenMacanovic Reason , open source models have come at par with closed source for example Qwen 3.8 27B on my local macbook did wrote better code than online DeepSeek and Kimi k2.7 code. The only issue it's a bit slow on my Mac and I can't do anything else while it running. 2/2
@MladenMacanovic If you have the money you should buy it, I frankly regret my decision of not buying one when I thought of buying it. Now it's waaaaay over my budget (not that earlier it was in my budget :) but it was in my stretch budget :) ) 1/2
TfLens : My AI observability dashboard. It reads the five metrics files TechieFlow writes and shows what AI-first development costs/project, phase, model and tool. Free, every number next to its source file. Will explain that Live Tonight 9:30 PM IST : https://t.co/hhJhkfGSoy
The full Stream 2 repository is public: https://t.co/npHUSUR7Co
Nine AI models, one brief, phase 1 only (requirements, architecture, screen mockups), zero corrections from me. Every model's documents and mockups are in there unedited, which model should be 'MyDiary' builder?
Tonight 9:30 PM IST, live: one brief, nine AI models, same framework, zero corrections. Opus, Sonnet, Haiku, Kimi, DeepSeek, GLM, GPT 6 Luna, Qwen Max, and a local Qwen on my Mac. 8 minutes to 15 hours for the same plan. You vote on who builds it https://t.co/YRxnL9RjXa
New lesson learned, even if you have a powerful Mac don't run local model and do live streaming simultaneously π€¦ . Will keep that in mind next time. Please watch the stream recording after 18 minutes if you are planning to watch this video.
AI puts you on a pedestal.
Then you look down and see the pedestal has no base.
A year of building with AI: 21 repositories, 1 shipped product, and the numbers to go with it. https://t.co/6nd6IrruKi