Joe Fernandez
@joenandez.com
🤖 Building AI stuff w/AI & sharing what I learn 👨💻 | 🕶️ AI PM @Meta | 🏋️♂️ Health | 🧠 ADHD hacks | ⛳️ Golf | 💕 Hubs/Dad x2
🎉 Introducing Subspace
A new home for coding agents and their operators.
Closed beta starts this week.
Open beta January 2026.
Onboarding diehard Claude Code users next week.
Sign up below!
A new home for coding agents and their operators.
Closed beta starts this week.
Open beta January 2026.
Onboarding diehard Claude Code users next week.
Sign up below!
December 19, 2025 at 5:43 PM
🎉 Introducing Subspace
A new home for coding agents and their operators.
Closed beta starts this week.
Open beta January 2026.
Onboarding diehard Claude Code users next week.
Sign up below!
A new home for coding agents and their operators.
Closed beta starts this week.
Open beta January 2026.
Onboarding diehard Claude Code users next week.
Sign up below!
My favorite way to start the day is just crushing a bunch of small cosmetic bugs that have been annoying me before i get to the big stuff. Gotta build up the dopamine hits before I try and swallow the frog.
December 18, 2025 at 4:02 PM
My favorite way to start the day is just crushing a bunch of small cosmetic bugs that have been annoying me before i get to the big stuff. Gotta build up the dopamine hits before I try and swallow the frog.
GPT 5.2 is a huge improvement in Codex CLI purely from a conversational/ergonomics standpoint. I always really struggled using Codex, just found it hard to communicate with.
Opus 4.5 still better, but now I don't completely dread switching to Codex when I hit my cap.
Opus 4.5 still better, but now I don't completely dread switching to Codex when I hit my cap.
December 12, 2025 at 5:53 PM
GPT 5.2 is a huge improvement in Codex CLI purely from a conversational/ergonomics standpoint. I always really struggled using Codex, just found it hard to communicate with.
Opus 4.5 still better, but now I don't completely dread switching to Codex when I hit my cap.
Opus 4.5 still better, but now I don't completely dread switching to Codex when I hit my cap.
Once the tools all get figured out, the marketplace for skills, subagents, workflows, etc is going to be so enormous.
Today you can create them and share them, but you can’t really measure or improve them - there’s no way to really evaluate other than vibes and your own direct trial/error.
Today you can create them and share them, but you can’t really measure or improve them - there’s no way to really evaluate other than vibes and your own direct trial/error.
December 7, 2025 at 6:24 PM
Once the tools all get figured out, the marketplace for skills, subagents, workflows, etc is going to be so enormous.
Today you can create them and share them, but you can’t really measure or improve them - there’s no way to really evaluate other than vibes and your own direct trial/error.
Today you can create them and share them, but you can’t really measure or improve them - there’s no way to really evaluate other than vibes and your own direct trial/error.
Thoughts on this terminal/cli agent interaction?
Personally ... I'm obsessed.
Personally ... I'm obsessed.
December 3, 2025 at 9:39 PM
Thoughts on this terminal/cli agent interaction?
Personally ... I'm obsessed.
Personally ... I'm obsessed.
How no one hasn't come up with a better approach for feeding/providing an LLM access to frontend/backend logs is beyond me.
Please tell me someone has a great solution for this?
Please tell me someone has a great solution for this?
December 2, 2025 at 12:45 AM
How no one hasn't come up with a better approach for feeding/providing an LLM access to frontend/backend logs is beyond me.
Please tell me someone has a great solution for this?
Please tell me someone has a great solution for this?
It’s great Anthropic is publishing these articles - the best practices for agentic coding are spread all over the nerdiest corners of the internet.
November 29, 2025 at 8:31 PM
It’s great Anthropic is publishing these articles - the best practices for agentic coding are spread all over the nerdiest corners of the internet.
After several full days working with Opus 4.5, i say this without hyperbole.
We are in a completely different world now.
Product development has changed. I truly believe that.
We are in a completely different world now.
Product development has changed. I truly believe that.
November 27, 2025 at 5:52 PM
After several full days working with Opus 4.5, i say this without hyperbole.
We are in a completely different world now.
Product development has changed. I truly believe that.
We are in a completely different world now.
Product development has changed. I truly believe that.
Just to pile on --- Opus 4.5 is insanely good.
What a release from Anthropic.
Crazy we got 5.1, 5.1 Codex Max, Gemini 3 Pro, AND Opus 4.5 in like 1 week.
What a release from Anthropic.
Crazy we got 5.1, 5.1 Codex Max, Gemini 3 Pro, AND Opus 4.5 in like 1 week.
November 25, 2025 at 4:32 AM
Just to pile on --- Opus 4.5 is insanely good.
What a release from Anthropic.
Crazy we got 5.1, 5.1 Codex Max, Gemini 3 Pro, AND Opus 4.5 in like 1 week.
What a release from Anthropic.
Crazy we got 5.1, 5.1 Codex Max, Gemini 3 Pro, AND Opus 4.5 in like 1 week.
Bringing my kids drawings to life with AI is so much fun and the kids absolutely love it.
My daughter’s “land of Sinba” - she’s creating a whole story.
My daughter’s “land of Sinba” - she’s creating a whole story.
November 22, 2025 at 10:24 PM
Bringing my kids drawings to life with AI is so much fun and the kids absolutely love it.
My daughter’s “land of Sinba” - she’s creating a whole story.
My daughter’s “land of Sinba” - she’s creating a whole story.
Fun facts
We named the company Codename because we *love* the thrill of new projects with a great code America
New June’s @newjunehq codename was Hogan.
The iOS app will forever be called Hogan.
Iykyk
We named the company Codename because we *love* the thrill of new projects with a great code America
New June’s @newjunehq codename was Hogan.
The iOS app will forever be called Hogan.
Iykyk
November 18, 2025 at 12:17 AM
Fun facts
We named the company Codename because we *love* the thrill of new projects with a great code America
New June’s @newjunehq codename was Hogan.
The iOS app will forever be called Hogan.
Iykyk
We named the company Codename because we *love* the thrill of new projects with a great code America
New June’s @newjunehq codename was Hogan.
The iOS app will forever be called Hogan.
Iykyk
My obsession this weekend is figuring out how to create an agent agnostic framework/system for skills, subagents, slash commands (human initiated + agent initiated), and I guess plugins.
I love Claude Code. It’s my fave.
I love Claude Code. It’s my fave.
November 16, 2025 at 4:44 PM
My obsession this weekend is figuring out how to create an agent agnostic framework/system for skills, subagents, slash commands (human initiated + agent initiated), and I guess plugins.
I love Claude Code. It’s my fave.
I love Claude Code. It’s my fave.
Spec driven development is about “alignment” not about creating a perfect blueprint.
Llm knows what needs to be built from spec, you align with the LLM on the design to deliver that, tasks keep the LLM on track.
Llm knows what needs to be built from spec, you align with the LLM on the design to deliver that, tasks keep the LLM on track.
November 16, 2025 at 2:49 AM
Spec driven development is about “alignment” not about creating a perfect blueprint.
Llm knows what needs to be built from spec, you align with the LLM on the design to deliver that, tasks keep the LLM on track.
Llm knows what needs to be built from spec, you align with the LLM on the design to deliver that, tasks keep the LLM on track.
Finally getting around to messing around with Claude Skills (i know i know I've been a little busy) and wow.
I've kind of "hacked" this with subagents -- we have a test-lead agent for example with a .md stuffed with golden test examples and best practices.
I've kind of "hacked" this with subagents -- we have a test-lead agent for example with a .md stuffed with golden test examples and best practices.
November 15, 2025 at 5:58 PM
Finally getting around to messing around with Claude Skills (i know i know I've been a little busy) and wow.
I've kind of "hacked" this with subagents -- we have a test-lead agent for example with a .md stuffed with golden test examples and best practices.
I've kind of "hacked" this with subagents -- we have a test-lead agent for example with a .md stuffed with golden test examples and best practices.
As someone actively working on AI features that even foundation models struggle to get consistently right, features that are critical to our “free” product @newjunehq… Kimi K2 Thinking is so wild.
Overnight our cost of “free” cuts in half. Less depending on how you host/deliver the model/tokens.
Overnight our cost of “free” cuts in half. Less depending on how you host/deliver the model/tokens.
November 11, 2025 at 4:32 PM
As someone actively working on AI features that even foundation models struggle to get consistently right, features that are critical to our “free” product @newjunehq… Kimi K2 Thinking is so wild.
Overnight our cost of “free” cuts in half. Less depending on how you host/deliver the model/tokens.
Overnight our cost of “free” cuts in half. Less depending on how you host/deliver the model/tokens.
How to tell someone lives all day in Claude Code - thinking in Context Windows as a measure of complexity
November 11, 2025 at 3:12 PM
How to tell someone lives all day in Claude Code - thinking in Context Windows as a measure of complexity
Been playing around with a few memory tools and plugins for Claude Code.
I've never been more convinced that the UI for Agentic Coding needs to change dramatically.
- is memory enabled?
- what does it remember?
- can i control when it remembers?
I've never been more convinced that the UI for Agentic Coding needs to change dramatically.
- is memory enabled?
- what does it remember?
- can i control when it remembers?
November 6, 2025 at 8:33 PM
Been playing around with a few memory tools and plugins for Claude Code.
I've never been more convinced that the UI for Agentic Coding needs to change dramatically.
- is memory enabled?
- what does it remember?
- can i control when it remembers?
I've never been more convinced that the UI for Agentic Coding needs to change dramatically.
- is memory enabled?
- what does it remember?
- can i control when it remembers?
Its been said before, but more people need to realize that LLMs are mirrors.
They take what you say and expand upon it. Very easy to silo yourself in a direction that you yourself are biased towards.
You need to explicitly ask for counter arguments, options, different perspectives.
They take what you say and expand upon it. Very easy to silo yourself in a direction that you yourself are biased towards.
You need to explicitly ask for counter arguments, options, different perspectives.
November 6, 2025 at 8:07 PM
Its been said before, but more people need to realize that LLMs are mirrors.
They take what you say and expand upon it. Very easy to silo yourself in a direction that you yourself are biased towards.
You need to explicitly ask for counter arguments, options, different perspectives.
They take what you say and expand upon it. Very easy to silo yourself in a direction that you yourself are biased towards.
You need to explicitly ask for counter arguments, options, different perspectives.
Building @newjunehq is such a roller coaster sometimes.
Some days we're like "f*** yeah this works perfectly" then the next day we're like "f*** there is no way we can make this work" and then the next day we're like "f*** yeah this works 10x better than I thought it would" repeat.
Some days we're like "f*** yeah this works perfectly" then the next day we're like "f*** there is no way we can make this work" and then the next day we're like "f*** yeah this works 10x better than I thought it would" repeat.
November 4, 2025 at 8:52 PM
Building @newjunehq is such a roller coaster sometimes.
Some days we're like "f*** yeah this works perfectly" then the next day we're like "f*** there is no way we can make this work" and then the next day we're like "f*** yeah this works 10x better than I thought it would" repeat.
Some days we're like "f*** yeah this works perfectly" then the next day we're like "f*** there is no way we can make this work" and then the next day we're like "f*** yeah this works 10x better than I thought it would" repeat.
I've been working out of coffee shops more frequently.
Great people watching.
A lot of people bring computer monitors! I was very surprised by this. But on days I spend more time here - I see like 3 or 4 people that do this!
Great people watching.
A lot of people bring computer monitors! I was very surprised by this. But on days I spend more time here - I see like 3 or 4 people that do this!
November 4, 2025 at 7:49 PM
I've been working out of coffee shops more frequently.
Great people watching.
A lot of people bring computer monitors! I was very surprised by this. But on days I spend more time here - I see like 3 or 4 people that do this!
Great people watching.
A lot of people bring computer monitors! I was very surprised by this. But on days I spend more time here - I see like 3 or 4 people that do this!
Having an "only with AI experience" and its really cool I have to share.
We're working on our @newjunehq shot detection/prediction algorithm.
The concept of working with a highly capable engineer, that also deeply understands the nuances of golf and can reason about...
We're working on our @newjunehq shot detection/prediction algorithm.
The concept of working with a highly capable engineer, that also deeply understands the nuances of golf and can reason about...
October 31, 2025 at 4:27 PM
Having an "only with AI experience" and its really cool I have to share.
We're working on our @newjunehq shot detection/prediction algorithm.
The concept of working with a highly capable engineer, that also deeply understands the nuances of golf and can reason about...
We're working on our @newjunehq shot detection/prediction algorithm.
The concept of working with a highly capable engineer, that also deeply understands the nuances of golf and can reason about...
OMG i've never been so happy. Claude Code startup is basically instant now!
Thank god... i was dealing with literal 15-20s startup times.
I've been submitting bugs and seems like they are really paying attention to these.
Incredible work from the Claude Code team.
Thank god... i was dealing with literal 15-20s startup times.
I've been submitting bugs and seems like they are really paying attention to these.
Incredible work from the Claude Code team.
October 31, 2025 at 3:38 PM
OMG i've never been so happy. Claude Code startup is basically instant now!
Thank god... i was dealing with literal 15-20s startup times.
I've been submitting bugs and seems like they are really paying attention to these.
Incredible work from the Claude Code team.
Thank god... i was dealing with literal 15-20s startup times.
I've been submitting bugs and seems like they are really paying attention to these.
Incredible work from the Claude Code team.
Super smart of Cursor to be focused on speed and innovating on the editor UX.
I think speed is really a top painpoint. Current CC/Codex models straddle between sync work and async work.
I'd rather be either FULLY sync or FULLY async.
cursor.com/blog/2-0
I think speed is really a top painpoint. Current CC/Codex models straddle between sync work and async work.
I'd rather be either FULLY sync or FULLY async.
cursor.com/blog/2-0
Introducing Cursor 2.0 and Composer · Cursor
Built to make you extraordinarily productive, Cursor is the best way to code with AI.
cursor.com
October 29, 2025 at 5:19 PM
Super smart of Cursor to be focused on speed and innovating on the editor UX.
I think speed is really a top painpoint. Current CC/Codex models straddle between sync work and async work.
I'd rather be either FULLY sync or FULLY async.
cursor.com/blog/2-0
I think speed is really a top painpoint. Current CC/Codex models straddle between sync work and async work.
I'd rather be either FULLY sync or FULLY async.
cursor.com/blog/2-0
Have been chatting with some early stage investors -- been super helpful forcing function to work on the @newjunehq pitch.
I had it all in my head but having to explain the whole value prop AND strategy AND risks to other humans live - different beast entirely.
I had it all in my head but having to explain the whole value prop AND strategy AND risks to other humans live - different beast entirely.
October 28, 2025 at 11:25 PM
Have been chatting with some early stage investors -- been super helpful forcing function to work on the @newjunehq pitch.
I had it all in my head but having to explain the whole value prop AND strategy AND risks to other humans live - different beast entirely.
I had it all in my head but having to explain the whole value prop AND strategy AND risks to other humans live - different beast entirely.
A current fave prompt
“You are a Sr Principal Engineer. The owner of this plan took suspiciously long to author it.
“You are a Sr Principal Engineer. The owner of this plan took suspiciously long to author it.
October 26, 2025 at 8:16 PM
A current fave prompt
“You are a Sr Principal Engineer. The owner of this plan took suspiciously long to author it.
“You are a Sr Principal Engineer. The owner of this plan took suspiciously long to author it.