Chidi
@justchidi.bsky.social
Mobile (KMP) and Data Engr. Interested in decentralized tech, open source ai, wasm, and proteins.
Kotlin and R enjoyoor
Kotlin and R enjoyoor
I got the “wait for 4 days” codex thing because gpt5-codex-high struggled to fix some Kotlin Multiplatform issues I had and used up over 75% weekly limit in the process ($20 plan).
Since then, I’ve been using GLM 4.6 from CC, it exceeds my low expectations although with more support than Codex
Since then, I’ve been using GLM 4.6 from CC, it exceeds my low expectations although with more support than Codex
October 23, 2025 at 10:37 AM
I got the “wait for 4 days” codex thing because gpt5-codex-high struggled to fix some Kotlin Multiplatform issues I had and used up over 75% weekly limit in the process ($20 plan).
Since then, I’ve been using GLM 4.6 from CC, it exceeds my low expectations although with more support than Codex
Since then, I’ve been using GLM 4.6 from CC, it exceeds my low expectations although with more support than Codex
I think curiosity can simulate, if not act as a substitute for passion
October 4, 2025 at 6:09 AM
I think curiosity can simulate, if not act as a substitute for passion
One big issue with Blueksy users obsessed about moderators and platform banning anyone they call out is laziness to use the tools provided by the platform. Blueksy team needs to do a lot more to smoothen and surface the features
October 3, 2025 at 3:37 PM
One big issue with Blueksy users obsessed about moderators and platform banning anyone they call out is laziness to use the tools provided by the platform. Blueksy team needs to do a lot more to smoothen and surface the features
Reposted by Chidi
Harassing the mods into banning someone has never worked. And harassing people in general has never changed their mind.
October 3, 2025 at 1:47 PM
Harassing the mods into banning someone has never worked. And harassing people in general has never changed their mind.
This attack against Jay might just help us get toxic angry users out of Bluesky.
Bluesky isn’t your “cancel anyone we hate” platform.
Bluesky isn’t your “cancel anyone we hate” platform.
October 3, 2025 at 3:30 PM
This attack against Jay might just help us get toxic angry users out of Bluesky.
Bluesky isn’t your “cancel anyone we hate” platform.
Bluesky isn’t your “cancel anyone we hate” platform.
Who’s building the Product Engineer’s product management board?
Collaboratively building products can’t live inside IDEs, Editors and CLI context folders, unless you’re solo building, context is too scattered, agents can do small well defined and clear tasks really well on their own
Collaboratively building products can’t live inside IDEs, Editors and CLI context folders, unless you’re solo building, context is too scattered, agents can do small well defined and clear tasks really well on their own
October 2, 2025 at 2:51 PM
Who’s building the Product Engineer’s product management board?
Collaboratively building products can’t live inside IDEs, Editors and CLI context folders, unless you’re solo building, context is too scattered, agents can do small well defined and clear tasks really well on their own
Collaboratively building products can’t live inside IDEs, Editors and CLI context folders, unless you’re solo building, context is too scattered, agents can do small well defined and clear tasks really well on their own
I need an agent that goes through user manuals and guides of SaaS products doing the same thing, and uses learnings from studying these to create detailed EPICs with a well designed domain modeling and database schema proposal based on different architectural patterns and a recommended
October 1, 2025 at 8:29 PM
I need an agent that goes through user manuals and guides of SaaS products doing the same thing, and uses learnings from studying these to create detailed EPICs with a well designed domain modeling and database schema proposal based on different architectural patterns and a recommended
I think the hack for building product level backends fast with AI is the domain modeling and well thought out schema design foundation, in addition to having as much repeated patterns easily pronounced in docs and code implementations.
These pattern machines would love you for this
These pattern machines would love you for this
October 1, 2025 at 4:39 PM
I think the hack for building product level backends fast with AI is the domain modeling and well thought out schema design foundation, in addition to having as much repeated patterns easily pronounced in docs and code implementations.
These pattern machines would love you for this
These pattern machines would love you for this
After Sonnet 3.5, OpenAI’s coding models in the same generation as its Claude equivalent was always better in planning and execution, gpt5 is better than sonnet or opus in my experience.
September 26, 2025 at 1:01 PM
After Sonnet 3.5, OpenAI’s coding models in the same generation as its Claude equivalent was always better in planning and execution, gpt5 is better than sonnet or opus in my experience.
Each day I heavily use AI for coding, I’m even more convinced that you can’t vibe code a scalable and maintainable app with it as it stands today.
At the same time it’s such an incredibly useful tool for software developers
At the same time it’s such an incredibly useful tool for software developers
September 25, 2025 at 2:23 PM
Each day I heavily use AI for coding, I’m even more convinced that you can’t vibe code a scalable and maintainable app with it as it stands today.
At the same time it’s such an incredibly useful tool for software developers
At the same time it’s such an incredibly useful tool for software developers
It’s actually crazy how almost all open source models from China are better than anything that’s ever come out of Meta considering the huge compute capacity and billions of $ investment Meta has done in the LLM space.
I think Zuck should go open an AI office in Shanghai and use Huawei chips 😅
I think Zuck should go open an AI office in Shanghai and use Huawei chips 😅
Meituan, China's largest food delivery company, is releasing more open-weight LLMs.
LongCat-Flash-Thinking: Smarter reasoning and leaner costs!
Performance: SOTA open-source models on Logic / Math / Coding / Agent tasks using 64.5% fewer tokens to hit top-tier accuracy on AIME25
LongCat-Flash-Thinking: Smarter reasoning and leaner costs!
Performance: SOTA open-source models on Logic / Math / Coding / Agent tasks using 64.5% fewer tokens to hit top-tier accuracy on AIME25
September 22, 2025 at 1:47 PM
It’s actually crazy how almost all open source models from China are better than anything that’s ever come out of Meta considering the huge compute capacity and billions of $ investment Meta has done in the LLM space.
I think Zuck should go open an AI office in Shanghai and use Huawei chips 😅
I think Zuck should go open an AI office in Shanghai and use Huawei chips 😅
What happened to Kirk is despicable and should be severely punished.
The aftermath, specially as it concerns public discourse is why Bluesky needs to work. Twitter is a radicalized extreme right echo chamber disguised as a public channel but designed to amplify one side and tone down the other
The aftermath, specially as it concerns public discourse is why Bluesky needs to work. Twitter is a radicalized extreme right echo chamber disguised as a public channel but designed to amplify one side and tone down the other
September 22, 2025 at 7:23 AM
What happened to Kirk is despicable and should be severely punished.
The aftermath, specially as it concerns public discourse is why Bluesky needs to work. Twitter is a radicalized extreme right echo chamber disguised as a public channel but designed to amplify one side and tone down the other
The aftermath, specially as it concerns public discourse is why Bluesky needs to work. Twitter is a radicalized extreme right echo chamber disguised as a public channel but designed to amplify one side and tone down the other
The quality of output between gpt-5-codex high and gpt-5-codex medium is really high
September 22, 2025 at 4:18 AM
The quality of output between gpt-5-codex high and gpt-5-codex medium is really high
Building this stuff alone, I use only codex. I use one model to implement and make sure my ticket wouldn’t cross 50% of context window.
After this I use the highest model to review only the recent git file changes and access if it matches acceptance criterias set out for the ticket.
After this I use the highest model to review only the recent git file changes and access if it matches acceptance criterias set out for the ticket.
September 21, 2025 at 10:39 AM
Building this stuff alone, I use only codex. I use one model to implement and make sure my ticket wouldn’t cross 50% of context window.
After this I use the highest model to review only the recent git file changes and access if it matches acceptance criterias set out for the ticket.
After this I use the highest model to review only the recent git file changes and access if it matches acceptance criterias set out for the ticket.
Reposted by Chidi
You know what didn’t happen?
Pundits going on Gab, Parler, Gettr, or Truth Social to repeatedly say “no one cares about this site, you’re in a bubble, you forfeited your influence, things are awesome on the site you don’t use anymore, so awesome that I come here to tell you you’re dumb unlike me.”
Pundits going on Gab, Parler, Gettr, or Truth Social to repeatedly say “no one cares about this site, you’re in a bubble, you forfeited your influence, things are awesome on the site you don’t use anymore, so awesome that I come here to tell you you’re dumb unlike me.”
September 20, 2025 at 7:29 PM
You know what didn’t happen?
Pundits going on Gab, Parler, Gettr, or Truth Social to repeatedly say “no one cares about this site, you’re in a bubble, you forfeited your influence, things are awesome on the site you don’t use anymore, so awesome that I come here to tell you you’re dumb unlike me.”
Pundits going on Gab, Parler, Gettr, or Truth Social to repeatedly say “no one cares about this site, you’re in a bubble, you forfeited your influence, things are awesome on the site you don’t use anymore, so awesome that I come here to tell you you’re dumb unlike me.”
LLMs makes me TDD. Good and responsible LLM usage in coding encourages best practices and makes it easier
September 21, 2025 at 6:17 AM
LLMs makes me TDD. Good and responsible LLM usage in coding encourages best practices and makes it easier
This proposed Twitter/X feature is why they called Bluesky an echo chamber, lol
September 20, 2025 at 8:52 AM
This proposed Twitter/X feature is why they called Bluesky an echo chamber, lol
As a primarily Android dev who’ve written Flutter and RN for some years, it’s fully how native Android developers love saying Compose Multiplatform is better than Flutter or React Native, because it’s true 😆😎
September 18, 2025 at 12:33 PM
As a primarily Android dev who’ve written Flutter and RN for some years, it’s fully how native Android developers love saying Compose Multiplatform is better than Flutter or React Native, because it’s true 😆😎
Part of being of a senior engineer is developing opinions about structure, algorithms and data structures of a program.
AI for coding has made it even more important to hone this skill, not less, because now we’ll need to be validating or reinforcing public opinion on code and be accountable for it
AI for coding has made it even more important to hone this skill, not less, because now we’ll need to be validating or reinforcing public opinion on code and be accountable for it
September 18, 2025 at 6:50 AM
Part of being of a senior engineer is developing opinions about structure, algorithms and data structures of a program.
AI for coding has made it even more important to hone this skill, not less, because now we’ll need to be validating or reinforcing public opinion on code and be accountable for it
AI for coding has made it even more important to hone this skill, not less, because now we’ll need to be validating or reinforcing public opinion on code and be accountable for it
LLM providers are baiting us and trying to make us over dependent on the models, their goal is to make it impossible to not have a job without the models or be able to do anything.
My goal is to be able to use all the AI stuff without falling into this trap.
Only the Chinese labs can save us.
My goal is to be able to use all the AI stuff without falling into this trap.
Only the Chinese labs can save us.
September 17, 2025 at 9:44 AM
LLM providers are baiting us and trying to make us over dependent on the models, their goal is to make it impossible to not have a job without the models or be able to do anything.
My goal is to be able to use all the AI stuff without falling into this trap.
Only the Chinese labs can save us.
My goal is to be able to use all the AI stuff without falling into this trap.
Only the Chinese labs can save us.
TikTok US is an example of American hypocrisy, but then, better we spy on ourselves than an adversary
The people who’ll suffer this are Dems and their supporters because this clique that now owns it now knows they owe it to Trump to pay him back in kind.
The people who’ll suffer this are Dems and their supporters because this clique that now owns it now knows they owe it to Trump to pay him back in kind.
September 17, 2025 at 7:49 AM
TikTok US is an example of American hypocrisy, but then, better we spy on ourselves than an adversary
The people who’ll suffer this are Dems and their supporters because this clique that now owns it now knows they owe it to Trump to pay him back in kind.
The people who’ll suffer this are Dems and their supporters because this clique that now owns it now knows they owe it to Trump to pay him back in kind.
I’m not updating my version of codex yet until I’m done with my current EPIC as the existing version which allows me to pick a model is very important feature for me, I don’t trust the model company to do this automatically in good faith. 😅
September 16, 2025 at 1:20 PM
I’m not updating my version of codex yet until I’m done with my current EPIC as the existing version which allows me to pick a model is very important feature for me, I don’t trust the model company to do this automatically in good faith. 😅
For some reason, building with AI has made me read more technical articles and books than I’ve done in the past.
I’m building this product for the entire stack (mobile, backend, infra, devops, web) with AI, and this has somehow led me to completing some books I’ve hoped to complete in the past.
I’m building this product for the entire stack (mobile, backend, infra, devops, web) with AI, and this has somehow led me to completing some books I’ve hoped to complete in the past.
September 15, 2025 at 11:51 AM
For some reason, building with AI has made me read more technical articles and books than I’ve done in the past.
I’m building this product for the entire stack (mobile, backend, infra, devops, web) with AI, and this has somehow led me to completing some books I’ve hoped to complete in the past.
I’m building this product for the entire stack (mobile, backend, infra, devops, web) with AI, and this has somehow led me to completing some books I’ve hoped to complete in the past.
Don’t just let an AI agent to go figure out context by itself. Provide it with just enough context it needs for the task, like selecting the right files, concise prompt query, instruction/rule files, etc.
Only time to let it figure out context is during planning, you’re not doing this everytime
Only time to let it figure out context is during planning, you’re not doing this everytime
September 15, 2025 at 11:45 AM
Don’t just let an AI agent to go figure out context by itself. Provide it with just enough context it needs for the task, like selecting the right files, concise prompt query, instruction/rule files, etc.
Only time to let it figure out context is during planning, you’re not doing this everytime
Only time to let it figure out context is during planning, you’re not doing this everytime
My new metric for measuring AI coding agents is how long it can think productively.
September 14, 2025 at 10:32 AM
My new metric for measuring AI coding agents is how long it can think productively.