Josh Karamuth
@joshkaramuth.com
Writing about web development. Currently building milliontimer.com - a time tracking and invoicing tool for freelancers.
Read the full post here: joshkaramuth.com/blog/program...
The Programmer's New Role: Guiding AI Code Generation
Learn how programmers are evolving from coders to AI guides—see a Python example proving why human expertise still matters.
joshkaramuth.com
April 7, 2025 at 9:45 AM
Read the full post here: joshkaramuth.com/blog/program...
The best devs in 5 years won’t be the fastest coders.
They’ll be the best at:
- Explaining problems to AI
- Anticipating hidden requirements
- Designing for change
Are you thinking this way yet?
They’ll be the best at:
- Explaining problems to AI
- Anticipating hidden requirements
- Designing for change
Are you thinking this way yet?
April 7, 2025 at 9:45 AM
The best devs in 5 years won’t be the fastest coders.
They’ll be the best at:
- Explaining problems to AI
- Anticipating hidden requirements
- Designing for change
Are you thinking this way yet?
They’ll be the best at:
- Explaining problems to AI
- Anticipating hidden requirements
- Designing for change
Are you thinking this way yet?
The new job of programmers:
✅ Not writing code
✅ Not even reviewing code
✅ Debugging the AI’s understanding of the problem
✅ Not writing code
✅ Not even reviewing code
✅ Debugging the AI’s understanding of the problem
April 7, 2025 at 9:45 AM
The new job of programmers:
✅ Not writing code
✅ Not even reviewing code
✅ Debugging the AI’s understanding of the problem
✅ Not writing code
✅ Not even reviewing code
✅ Debugging the AI’s understanding of the problem
3️⃣ So you teach the AI:
- Add tests for edge cases
- Make logic configurable
- Document assumptions
Now the code is robust, not just working.
- Add tests for edge cases
- Make logic configurable
- Document assumptions
Now the code is robust, not just working.
April 7, 2025 at 9:45 AM
3️⃣ So you teach the AI:
- Add tests for edge cases
- Make logic configurable
- Document assumptions
Now the code is robust, not just working.
- Add tests for edge cases
- Make logic configurable
- Document assumptions
Now the code is robust, not just working.
2️⃣ A junior dev might ship the AI’s first draft.
A senior dev immediately spots gaps:
- Edge cases? (Leap years, time zones)
- Biz rules? (Does week start Sun or Mon?)
- Future changes? (Hardcoded = tech debt)
A senior dev immediately spots gaps:
- Edge cases? (Leap years, time zones)
- Biz rules? (Does week start Sun or Mon?)
- Future changes? (Hardcoded = tech debt)
April 7, 2025 at 9:45 AM
2️⃣ A junior dev might ship the AI’s first draft.
A senior dev immediately spots gaps:
- Edge cases? (Leap years, time zones)
- Biz rules? (Does week start Sun or Mon?)
- Future changes? (Hardcoded = tech debt)
A senior dev immediately spots gaps:
- Edge cases? (Leap years, time zones)
- Biz rules? (Does week start Sun or Mon?)
- Future changes? (Hardcoded = tech debt)
1️⃣ You ask an AI: "Write a function to get the first/last day of the week."
It spits out code that looks right.
But is it actually right?
Spoiler: No. Not yet.
It spits out code that looks right.
But is it actually right?
Spoiler: No. Not yet.
April 7, 2025 at 9:45 AM
1️⃣ You ask an AI: "Write a function to get the first/last day of the week."
It spits out code that looks right.
But is it actually right?
Spoiler: No. Not yet.
It spits out code that looks right.
But is it actually right?
Spoiler: No. Not yet.
Agreed, using MCPs make me feel like a wizard. Nice writeup.
April 4, 2025 at 9:07 AM
Agreed, using MCPs make me feel like a wizard. Nice writeup.
DeepSeek R1 also makes planning features cheap and dead simple.
If you're a programmer, you can't go wrong with DeepSeek until Gemini 2.5 Pro comes out of the experimental stage.
If you're a programmer, you can't go wrong with DeepSeek until Gemini 2.5 Pro comes out of the experimental stage.
April 3, 2025 at 7:43 AM
DeepSeek R1 also makes planning features cheap and dead simple.
If you're a programmer, you can't go wrong with DeepSeek until Gemini 2.5 Pro comes out of the experimental stage.
If you're a programmer, you can't go wrong with DeepSeek until Gemini 2.5 Pro comes out of the experimental stage.
Overall, I'm not impressed by the quality that Claude provides vs DeepSeek when factoring the quality to cost ratio. Since I'm a programmer, I use the latter with my own knowledge to guide it properly.
For context, coding an entire CRUD app cost about $5 with Claude, but only $0.10 with DeepSeek.
For context, coding an entire CRUD app cost about $5 with Claude, but only $0.10 with DeepSeek.
April 3, 2025 at 7:43 AM
Overall, I'm not impressed by the quality that Claude provides vs DeepSeek when factoring the quality to cost ratio. Since I'm a programmer, I use the latter with my own knowledge to guide it properly.
For context, coding an entire CRUD app cost about $5 with Claude, but only $0.10 with DeepSeek.
For context, coding an entire CRUD app cost about $5 with Claude, but only $0.10 with DeepSeek.
If you're not a programmer, you should stick with Claude because it's going to provide good quality code most of the time.
If you're a programmer, or if you have a programmer who can help you out, use DeepSeek because you can use your own knowledge to correct it.
If you're a programmer, or if you have a programmer who can help you out, use DeepSeek because you can use your own knowledge to correct it.
April 3, 2025 at 7:43 AM
If you're not a programmer, you should stick with Claude because it's going to provide good quality code most of the time.
If you're a programmer, or if you have a programmer who can help you out, use DeepSeek because you can use your own knowledge to correct it.
If you're a programmer, or if you have a programmer who can help you out, use DeepSeek because you can use your own knowledge to correct it.
Gemini 2.5 is like magic but the rate limits make it unusable for serious coding sessions. We'll have to wait until a proper version is released to switch to it.
Claude has decent quality, but is way too expensive. Which means...
Claude has decent quality, but is way too expensive. Which means...
April 3, 2025 at 7:43 AM
Gemini 2.5 is like magic but the rate limits make it unusable for serious coding sessions. We'll have to wait until a proper version is released to switch to it.
Claude has decent quality, but is way too expensive. Which means...
Claude has decent quality, but is way too expensive. Which means...
When it comes to quality:
Gemini > Claude > DeepSeek
However...
Gemini > Claude > DeepSeek
However...
April 3, 2025 at 7:43 AM
When it comes to quality:
Gemini > Claude > DeepSeek
However...
Gemini > Claude > DeepSeek
However...
Great idea! One of my least favorite parts of Django is having to write custom nodes for basic logic.
March 10, 2025 at 10:43 AM
Great idea! One of my least favorite parts of Django is having to write custom nodes for basic logic.
Same, except I also want to have enough time to read a ton of books, and take more walks in nature.
March 9, 2025 at 12:08 PM
Same, except I also want to have enough time to read a ton of books, and take more walks in nature.
ANTIDOTES HE’D IGNORE:
- Touch grass
- Have a bad conversation
- Draw a stick figure
Create, don’t critique. Log off, read Dostoevsky, then log off again. Fin.
- Touch grass
- Have a bad conversation
- Draw a stick figure
Create, don’t critique. Log off, read Dostoevsky, then log off again. Fin.
March 8, 2025 at 3:30 PM
ANTIDOTES HE’D IGNORE:
- Touch grass
- Have a bad conversation
- Draw a stick figure
Create, don’t critique. Log off, read Dostoevsky, then log off again. Fin.
- Touch grass
- Have a bad conversation
- Draw a stick figure
Create, don’t critique. Log off, read Dostoevsky, then log off again. Fin.
WHY IT MATTERS:
Dostoevsky shows how isolation weaponizes intelligence. The Underground Man’s tragedy?
He’d rather be “right” alone than vulnerable IRL. (Replace “underground” with “Twitter/X drafts.”)
Dostoevsky shows how isolation weaponizes intelligence. The Underground Man’s tragedy?
He’d rather be “right” alone than vulnerable IRL. (Replace “underground” with “Twitter/X drafts.”)
March 8, 2025 at 3:30 PM
WHY IT MATTERS:
Dostoevsky shows how isolation weaponizes intelligence. The Underground Man’s tragedy?
He’d rather be “right” alone than vulnerable IRL. (Replace “underground” with “Twitter/X drafts.”)
Dostoevsky shows how isolation weaponizes intelligence. The Underground Man’s tragedy?
He’d rather be “right” alone than vulnerable IRL. (Replace “underground” with “Twitter/X drafts.”)
COMPLAINING AS A LIFESTYLE
The Underground Man hates math, society, himself.
Today? 10-paragraph vents about minor annoyances (“DAE hate loud chewers?!”) to farm validation.
Same cry, different century.
The Underground Man hates math, society, himself.
Today? 10-paragraph vents about minor annoyances (“DAE hate loud chewers?!”) to farm validation.
Same cry, different century.
March 8, 2025 at 3:30 PM
COMPLAINING AS A LIFESTYLE
The Underground Man hates math, society, himself.
Today? 10-paragraph vents about minor annoyances (“DAE hate loud chewers?!”) to farm validation.
Same cry, different century.
The Underground Man hates math, society, himself.
Today? 10-paragraph vents about minor annoyances (“DAE hate loud chewers?!”) to farm validation.
Same cry, different century.
OVERFICTION PROBLEMS
He invents grievances to feed his self-pity. Modern version: arguing about hypothetical ethics in niche subreddits at 2 AM.
Not to fix anything—to *feel* superior. Classic.
He invents grievances to feed his self-pity. Modern version: arguing about hypothetical ethics in niche subreddits at 2 AM.
Not to fix anything—to *feel* superior. Classic.
March 8, 2025 at 3:30 PM
OVERFICTION PROBLEMS
He invents grievances to feed his self-pity. Modern version: arguing about hypothetical ethics in niche subreddits at 2 AM.
Not to fix anything—to *feel* superior. Classic.
He invents grievances to feed his self-pity. Modern version: arguing about hypothetical ethics in niche subreddits at 2 AM.
Not to fix anything—to *feel* superior. Classic.
ISOLATION = IMAGINED SUPERIORITY
He stews alone, convinced he’s smarter than everyone… while hating himself. Sound like anyone in your replies?
*“I am alone, and they are everyone.”*
Replace “they” with “normies.” Same energy.
He stews alone, convinced he’s smarter than everyone… while hating himself. Sound like anyone in your replies?
*“I am alone, and they are everyone.”*
Replace “they” with “normies.” Same energy.
March 8, 2025 at 3:30 PM
ISOLATION = IMAGINED SUPERIORITY
He stews alone, convinced he’s smarter than everyone… while hating himself. Sound like anyone in your replies?
*“I am alone, and they are everyone.”*
Replace “they” with “normies.” Same energy.
He stews alone, convinced he’s smarter than everyone… while hating himself. Sound like anyone in your replies?
*“I am alone, and they are everyone.”*
Replace “they” with “normies.” Same energy.
Full walkthrough:
joshkaramuth.com/blog/linux-b...
Don't be the "I'll set up backups tomorrow" person. Your future self will high-five you.
joshkaramuth.com/blog/linux-b...
Don't be the "I'll set up backups tomorrow" person. Your future self will high-five you.
Redundant Linux backups with a single command
Set up military-grade, cost-effective backups with Borgmatic—combining local USB and Hetzner cloud storage—in just 15 minutes, ensuring your data is safe even if your laptop takes a coffee bath.
joshkaramuth.com
March 8, 2025 at 9:09 AM
Full walkthrough:
joshkaramuth.com/blog/linux-b...
Don't be the "I'll set up backups tomorrow" person. Your future self will high-five you.
joshkaramuth.com/blog/linux-b...
Don't be the "I'll set up backups tomorrow" person. Your future self will high-five you.