ND Coder
@ndcoder.bsky.social
Semi-Retired Engineer/Manager after ~25 years in the tech industry.
Opinions on absolutely everything, many of them incorrect.
Opinions on absolutely everything, many of them incorrect.
1. Do a Google Search
2. It brings you directly to the Gemini AI
3. Ask How to turn off AI search
4. Get a wrong answer
5. Switch the default search to Duck Duck Go
2. It brings you directly to the Gemini AI
3. Ask How to turn off AI search
4. Get a wrong answer
5. Switch the default search to Duck Duck Go
January 15, 2026 at 10:33 PM
1. Do a Google Search
2. It brings you directly to the Gemini AI
3. Ask How to turn off AI search
4. Get a wrong answer
5. Switch the default search to Duck Duck Go
2. It brings you directly to the Gemini AI
3. Ask How to turn off AI search
4. Get a wrong answer
5. Switch the default search to Duck Duck Go
Claude Code . . . . what happened? Bad auto-update on Mac? Instances running on linux still work.
😢
😢
January 7, 2026 at 8:18 PM
Claude Code . . . . what happened? Bad auto-update on Mac? Instances running on linux still work.
😢
😢
"Token anxiety"
When you are worried your current implementation burns too many tokens and could potentially bankrupt you.
When you are worried your current implementation burns too many tokens and could potentially bankrupt you.
January 3, 2026 at 9:56 PM
"Token anxiety"
When you are worried your current implementation burns too many tokens and could potentially bankrupt you.
When you are worried your current implementation burns too many tokens and could potentially bankrupt you.
They be little chonkers!
December 31, 2025 at 8:33 PM
They be little chonkers!
Reached a new milestone! All image assets are procedurally generated.
Still significant fit and finish work to do on the project, but art changes are now much easier.
Still significant fit and finish work to do on the project, but art changes are now much easier.
December 31, 2025 at 2:28 AM
Reached a new milestone! All image assets are procedurally generated.
Still significant fit and finish work to do on the project, but art changes are now much easier.
Still significant fit and finish work to do on the project, but art changes are now much easier.
Back to work!
Blender + Python for game assets.
Blender + Python for game assets.
December 29, 2025 at 11:27 PM
Back to work!
Blender + Python for game assets.
Blender + Python for game assets.
Claude: Here's a detailed plan for what you just requested
Me: Looks good, implement it <steps away for lunch>
Claude: Here's extra detail on what you already approved. What do you want me to to?
(I want you to stop halting your work to ask redundant questions)
🤦♂️
Me: Looks good, implement it <steps away for lunch>
Claude: Here's extra detail on what you already approved. What do you want me to to?
(I want you to stop halting your work to ask redundant questions)
🤦♂️
December 23, 2025 at 11:08 PM
Claude: Here's a detailed plan for what you just requested
Me: Looks good, implement it <steps away for lunch>
Claude: Here's extra detail on what you already approved. What do you want me to to?
(I want you to stop halting your work to ask redundant questions)
🤦♂️
Me: Looks good, implement it <steps away for lunch>
Claude: Here's extra detail on what you already approved. What do you want me to to?
(I want you to stop halting your work to ask redundant questions)
🤦♂️
Telling Claude to "Ultrathink" feels like telling my kids to think about what they've done after they did something wrong.
December 22, 2025 at 10:28 PM
Telling Claude to "Ultrathink" feels like telling my kids to think about what they've done after they did something wrong.
Can confirm, Claude github integration burns through tokens.
I asked it to add details and a plan to address an issue. In the CLI, it would be a small fix. When reading it's own investigation, it decided to walk the codebase again and use a multi-step thinking process with a todo list.
I asked it to add details and a plan to address an issue. In the CLI, it would be a small fix. When reading it's own investigation, it decided to walk the codebase again and use a multi-step thinking process with a todo list.
December 11, 2025 at 6:16 PM
Can confirm, Claude github integration burns through tokens.
I asked it to add details and a plan to address an issue. In the CLI, it would be a small fix. When reading it's own investigation, it decided to walk the codebase again and use a multi-step thinking process with a todo list.
I asked it to add details and a plan to address an issue. In the CLI, it would be a small fix. When reading it's own investigation, it decided to walk the codebase again and use a multi-step thinking process with a todo list.
First time hitting the Claude 5x usage limit. Very productive morning with major changes.
Time for a lunch break.
Time for a lunch break.
December 10, 2025 at 8:12 PM
First time hitting the Claude 5x usage limit. Very productive morning with major changes.
Time for a lunch break.
Time for a lunch break.
. . . . and now we're isometric.
December 9, 2025 at 11:39 PM
. . . . and now we're isometric.
It's getting there. Too bad I forgot how to math and can't calculate angles properly.
December 9, 2025 at 10:04 PM
It's getting there. Too bad I forgot how to math and can't calculate angles properly.
(Opens app on dual monitor setup)
Me: Wow, this is taking so long to load, maybe I need a faster machine.
App: Hey, I'm over here on the other monitor, idiot.
Me: Wow, this is taking so long to load, maybe I need a faster machine.
App: Hey, I'm over here on the other monitor, idiot.
December 7, 2025 at 3:50 AM
(Opens app on dual monitor setup)
Me: Wow, this is taking so long to load, maybe I need a faster machine.
App: Hey, I'm over here on the other monitor, idiot.
Me: Wow, this is taking so long to load, maybe I need a faster machine.
App: Hey, I'm over here on the other monitor, idiot.
Refactoring with Coding Agents is like eating your vegetables. You should be doing it every day.
Agents bolting on new features have limited context and do not prioritize reusing code. Even with agent definitions being told to do so, managing context windows has side effects.
Agents bolting on new features have limited context and do not prioritize reusing code. Even with agent definitions being told to do so, managing context windows has side effects.
December 5, 2025 at 7:27 PM
Refactoring with Coding Agents is like eating your vegetables. You should be doing it every day.
Agents bolting on new features have limited context and do not prioritize reusing code. Even with agent definitions being told to do so, managing context windows has side effects.
Agents bolting on new features have limited context and do not prioritize reusing code. Even with agent definitions being told to do so, managing context windows has side effects.
(Minor change)
Claude: Do you want to auto-approve the edits?
Me: yes
Claude: <asks permission for every tool used to read a directory related to the change>
Claude: Do you want to auto-approve the edits?
Me: yes
Claude: <asks permission for every tool used to read a directory related to the change>
December 3, 2025 at 8:24 PM
(Minor change)
Claude: Do you want to auto-approve the edits?
Me: yes
Claude: <asks permission for every tool used to read a directory related to the change>
Claude: Do you want to auto-approve the edits?
Me: yes
Claude: <asks permission for every tool used to read a directory related to the change>
Fought Claude Sonnet 4.5 on a bug for about 20 minutes, chewing through ~30% of the 5h quota. Very frustrating.
Upgraded to Max and tried Opus 4.5. It fixed it first try.
Happy for Opus, still angry with Sonnet.
Upgraded to Max and tried Opus 4.5. It fixed it first try.
Happy for Opus, still angry with Sonnet.
December 2, 2025 at 11:42 PM
Fought Claude Sonnet 4.5 on a bug for about 20 minutes, chewing through ~30% of the 5h quota. Very frustrating.
Upgraded to Max and tried Opus 4.5. It fixed it first try.
Happy for Opus, still angry with Sonnet.
Upgraded to Max and tried Opus 4.5. It fixed it first try.
Happy for Opus, still angry with Sonnet.
I'm seeing the value of Claude Code's interactions limits, rather than just tokens. Its back and forth lead to significant token usage, and not fully under user control.
By counting interactions, the user is not penalized. It also doesn't stop working mid-interaction leaving code broken.
By counting interactions, the user is not penalized. It also doesn't stop working mid-interaction leaving code broken.
November 28, 2025 at 6:32 PM
I'm seeing the value of Claude Code's interactions limits, rather than just tokens. Its back and forth lead to significant token usage, and not fully under user control.
By counting interactions, the user is not penalized. It also doesn't stop working mid-interaction leaving code broken.
By counting interactions, the user is not penalized. It also doesn't stop working mid-interaction leaving code broken.
Claude 5 hour usage limit is frustrating.
1. It does not show how many are left. /cost only shows tokens, not available interactions
2. Most of the interactions were explaining what it did wrong, not on new development.
Non-transparent limits are bad. Don't penalize the user for your bad code.
1. It does not show how many are left. /cost only shows tokens, not available interactions
2. Most of the interactions were explaining what it did wrong, not on new development.
Non-transparent limits are bad. Don't penalize the user for your bad code.
November 27, 2025 at 7:24 PM
Claude 5 hour usage limit is frustrating.
1. It does not show how many are left. /cost only shows tokens, not available interactions
2. Most of the interactions were explaining what it did wrong, not on new development.
Non-transparent limits are bad. Don't penalize the user for your bad code.
1. It does not show how many are left. /cost only shows tokens, not available interactions
2. Most of the interactions were explaining what it did wrong, not on new development.
Non-transparent limits are bad. Don't penalize the user for your bad code.
Claude thought it new best and initiated a git submodule in the project directory, which is now not tracked from other areas. Time for manual cleanup 🤦♂️
November 27, 2025 at 7:07 PM
Claude thought it new best and initiated a git submodule in the project directory, which is now not tracked from other areas. Time for manual cleanup 🤦♂️
Claude workflow:
1. Basic requirements prompt ask it to do Q&A to fill in details
2. Build a PRD, track open questions and tasks
3. YOLO code mode
It ignored most of what was in the plan.
1. Basic requirements prompt ask it to do Q&A to fill in details
2. Build a PRD, track open questions and tasks
3. YOLO code mode
It ignored most of what was in the plan.
November 27, 2025 at 4:48 PM
Claude workflow:
1. Basic requirements prompt ask it to do Q&A to fill in details
2. Build a PRD, track open questions and tasks
3. YOLO code mode
It ignored most of what was in the plan.
1. Basic requirements prompt ask it to do Q&A to fill in details
2. Build a PRD, track open questions and tasks
3. YOLO code mode
It ignored most of what was in the plan.
First impressions with Claude Code:
- Much better at breaking down tasks than Gemini
- Better at managing context window by offloading to markdown
- Better interface for question and answer based planning workflows
- Easy creation of specialized agents (PM, SWE, QA, etc)
- Much better at breaking down tasks than Gemini
- Better at managing context window by offloading to markdown
- Better interface for question and answer based planning workflows
- Easy creation of specialized agents (PM, SWE, QA, etc)
November 27, 2025 at 1:12 AM
First impressions with Claude Code:
- Much better at breaking down tasks than Gemini
- Better at managing context window by offloading to markdown
- Better interface for question and answer based planning workflows
- Easy creation of specialized agents (PM, SWE, QA, etc)
- Much better at breaking down tasks than Gemini
- Better at managing context window by offloading to markdown
- Better interface for question and answer based planning workflows
- Easy creation of specialized agents (PM, SWE, QA, etc)
Codebase is ~2500 lines, with ~100 lines of README.md and other instructions. Tried to refactor how the game tracks game pieces and renders enemy locations. It was a constant game of whack-a-bug.
November 25, 2025 at 8:28 PM
Codebase is ~2500 lines, with ~100 lines of README.md and other instructions. Tried to refactor how the game tracks game pieces and renders enemy locations. It was a constant game of whack-a-bug.
Bad Gemini CLI behavior: ask it to roll back with a "git reset --hard" then it tries what it was doing again without being asked.
November 25, 2025 at 8:12 PM
Bad Gemini CLI behavior: ask it to roll back with a "git reset --hard" then it tries what it was doing again without being asked.
@hboon.com how good is Claude Code at avoiding spaghetti code?
November 25, 2025 at 7:13 PM
@hboon.com how good is Claude Code at avoiding spaghetti code?
Challenge of the day: articulating instructions to AI without using ambiguous language when concepts are similar.
November 25, 2025 at 7:04 PM
Challenge of the day: articulating instructions to AI without using ambiguous language when concepts are similar.