@teemuso.bsky.social
Customer support AI just gave me refund for ChatGPT
@OpenAI
Either it works extremely well, or it hallucinated on me.
Not sure how to feel about this.
@OpenAI
Either it works extremely well, or it hallucinated on me.
Not sure how to feel about this.
December 23, 2025 at 2:42 PM
Customer support AI just gave me refund for ChatGPT
@OpenAI
Either it works extremely well, or it hallucinated on me.
Not sure how to feel about this.
@OpenAI
Either it works extremely well, or it hallucinated on me.
Not sure how to feel about this.
Make it work before you make it pretty.
Once the functionality is solid, then design.
The interface always changes as the product takes shape.
Once the functionality is solid, then design.
The interface always changes as the product takes shape.
November 12, 2025 at 7:32 PM
Make it work before you make it pretty.
Once the functionality is solid, then design.
The interface always changes as the product takes shape.
Once the functionality is solid, then design.
The interface always changes as the product takes shape.
Did you build app with AI?
Does it ACTUALLY work for users?
We know AI builds fast but breaks things.
So I built prooven - AI tests your app like a real user.
Share link to your web app and I'll test it, or try it out yourself 👇
Does it ACTUALLY work for users?
We know AI builds fast but breaks things.
So I built prooven - AI tests your app like a real user.
Share link to your web app and I'll test it, or try it out yourself 👇
November 11, 2025 at 9:06 PM
Did you build app with AI?
Does it ACTUALLY work for users?
We know AI builds fast but breaks things.
So I built prooven - AI tests your app like a real user.
Share link to your web app and I'll test it, or try it out yourself 👇
Does it ACTUALLY work for users?
We know AI builds fast but breaks things.
So I built prooven - AI tests your app like a real user.
Share link to your web app and I'll test it, or try it out yourself 👇
I just realized my app:
"Prooven keeps you in control when building with AI."
works for arbitrary tasks on the internet.
Prooven tests the user flows in your app, but the user flows can be anything!
Small launch tomorrow, free usage incoming!
"Prooven keeps you in control when building with AI."
works for arbitrary tasks on the internet.
Prooven tests the user flows in your app, but the user flows can be anything!
Small launch tomorrow, free usage incoming!
November 10, 2025 at 11:48 PM
I just realized my app:
"Prooven keeps you in control when building with AI."
works for arbitrary tasks on the internet.
Prooven tests the user flows in your app, but the user flows can be anything!
Small launch tomorrow, free usage incoming!
"Prooven keeps you in control when building with AI."
works for arbitrary tasks on the internet.
Prooven tests the user flows in your app, but the user flows can be anything!
Small launch tomorrow, free usage incoming!
It's all fun and games before its sudo rm -rf /
November 8, 2025 at 7:05 PM
It's all fun and games before its sudo rm -rf /
Create high quality PRD, and guard its quality with your life.
Let AI read the PRD, propose updates, but always review them by hand.
Sounds good?
The PRD-generator v2 is on the way!
Testing the MCP server by dogfooding 👇
Let AI read the PRD, propose updates, but always review them by hand.
Sounds good?
The PRD-generator v2 is on the way!
Testing the MCP server by dogfooding 👇
November 6, 2025 at 7:03 PM
Create high quality PRD, and guard its quality with your life.
Let AI read the PRD, propose updates, but always review them by hand.
Sounds good?
The PRD-generator v2 is on the way!
Testing the MCP server by dogfooding 👇
Let AI read the PRD, propose updates, but always review them by hand.
Sounds good?
The PRD-generator v2 is on the way!
Testing the MCP server by dogfooding 👇
I'll show you how to go from idea-to-MVP, quick.
Day 0: research APIs and verify documentation.
I already have a Lovable design for the MVP, so now I simply write everything down for the AI.
Day 0: research APIs and verify documentation.
I already have a Lovable design for the MVP, so now I simply write everything down for the AI.
November 6, 2025 at 6:00 PM
I'll show you how to go from idea-to-MVP, quick.
Day 0: research APIs and verify documentation.
I already have a Lovable design for the MVP, so now I simply write everything down for the AI.
Day 0: research APIs and verify documentation.
I already have a Lovable design for the MVP, so now I simply write everything down for the AI.
Creating PRD-generator v2 docs with prd-generator.
The PRD generator has been extremely helpful for building MVPs, but it's time to step it up with MCP and some agents.
Meanwhile, you can still use the old PRD generator for free below.
The PRD generator has been extremely helpful for building MVPs, but it's time to step it up with MCP and some agents.
Meanwhile, you can still use the old PRD generator for free below.
November 6, 2025 at 10:14 AM
Creating PRD-generator v2 docs with prd-generator.
The PRD generator has been extremely helpful for building MVPs, but it's time to step it up with MCP and some agents.
Meanwhile, you can still use the old PRD generator for free below.
The PRD generator has been extremely helpful for building MVPs, but it's time to step it up with MCP and some agents.
Meanwhile, you can still use the old PRD generator for free below.
Writing docs for AI?
Describe results, not systems.
For UI projects --> document screens & user flows.
For AI agents --> document input shapes and expected outcomes.
Minimal effort, maximal control.
Don't write boilerplate documentation that AI will skip over.
Describe results, not systems.
For UI projects --> document screens & user flows.
For AI agents --> document input shapes and expected outcomes.
Minimal effort, maximal control.
Don't write boilerplate documentation that AI will skip over.
November 5, 2025 at 10:00 PM
Writing docs for AI?
Describe results, not systems.
For UI projects --> document screens & user flows.
For AI agents --> document input shapes and expected outcomes.
Minimal effort, maximal control.
Don't write boilerplate documentation that AI will skip over.
Describe results, not systems.
For UI projects --> document screens & user flows.
For AI agents --> document input shapes and expected outcomes.
Minimal effort, maximal control.
Don't write boilerplate documentation that AI will skip over.
Always guide the AI with manually reviewed docs.
Let AI maintain internal docs (it's more verbose anyway).
Keep human-readable "source of truth" that you review yourself. Give up control for the AI internal docs, and ask AI to sync their changes to your docs.
Let AI maintain internal docs (it's more verbose anyway).
Keep human-readable "source of truth" that you review yourself. Give up control for the AI internal docs, and ask AI to sync their changes to your docs.
November 5, 2025 at 12:26 PM
Always guide the AI with manually reviewed docs.
Let AI maintain internal docs (it's more verbose anyway).
Keep human-readable "source of truth" that you review yourself. Give up control for the AI internal docs, and ask AI to sync their changes to your docs.
Let AI maintain internal docs (it's more verbose anyway).
Keep human-readable "source of truth" that you review yourself. Give up control for the AI internal docs, and ask AI to sync their changes to your docs.
Claude Code cooked again.
I love choosing from options. It helps me narrow the possibilities.
If I'm professional and see the options are wrong, then I can just write on my own.
What a pleasure to code.
I love choosing from options. It helps me narrow the possibilities.
If I'm professional and see the options are wrong, then I can just write on my own.
What a pleasure to code.
October 23, 2025 at 11:00 AM
Claude Code cooked again.
I love choosing from options. It helps me narrow the possibilities.
If I'm professional and see the options are wrong, then I can just write on my own.
What a pleasure to code.
I love choosing from options. It helps me narrow the possibilities.
If I'm professional and see the options are wrong, then I can just write on my own.
What a pleasure to code.
When vibe coding feels too easy, prepare for refactoring.
You don't need to read code, but you need to read docs and give the AI crystal clear instructions.
Otherwise, you'll drown in the AI's assumptions and sea of unaligned code.
You don't need to read code, but you need to read docs and give the AI crystal clear instructions.
Otherwise, you'll drown in the AI's assumptions and sea of unaligned code.
October 21, 2025 at 7:00 AM
When vibe coding feels too easy, prepare for refactoring.
You don't need to read code, but you need to read docs and give the AI crystal clear instructions.
Otherwise, you'll drown in the AI's assumptions and sea of unaligned code.
You don't need to read code, but you need to read docs and give the AI crystal clear instructions.
Otherwise, you'll drown in the AI's assumptions and sea of unaligned code.
Store context everywhere: commits, files, issues.
You're the context manager. Guide the AI:
"Read previous commits and upcoming issues, then propose a plan."
Aim for simplicity.
You're the context manager. Guide the AI:
"Read previous commits and upcoming issues, then propose a plan."
Aim for simplicity.
October 20, 2025 at 5:00 PM
Store context everywhere: commits, files, issues.
You're the context manager. Guide the AI:
"Read previous commits and upcoming issues, then propose a plan."
Aim for simplicity.
You're the context manager. Guide the AI:
"Read previous commits and upcoming issues, then propose a plan."
Aim for simplicity.
AI won't make you achieve your goals.
You will.
Planning is 70% of the work. Execution just follows the map.
You will.
Planning is 70% of the work. Execution just follows the map.
October 18, 2025 at 2:33 PM
AI won't make you achieve your goals.
You will.
Planning is 70% of the work. Execution just follows the map.
You will.
Planning is 70% of the work. Execution just follows the map.
PRDs are a pain, so I built a digital PRD.
I aim for balance between chaos and control.
I'm building it live with Lovable, and it's shining again for small internal apps.
Can’t share PRD? I prompt "add live sharing"
Need to mock screens with AI? I prompt "add export"
I aim for balance between chaos and control.
I'm building it live with Lovable, and it's shining again for small internal apps.
Can’t share PRD? I prompt "add live sharing"
Need to mock screens with AI? I prompt "add export"
October 7, 2025 at 9:41 PM
PRDs are a pain, so I built a digital PRD.
I aim for balance between chaos and control.
I'm building it live with Lovable, and it's shining again for small internal apps.
Can’t share PRD? I prompt "add live sharing"
Need to mock screens with AI? I prompt "add export"
I aim for balance between chaos and control.
I'm building it live with Lovable, and it's shining again for small internal apps.
Can’t share PRD? I prompt "add live sharing"
Need to mock screens with AI? I prompt "add export"
Dear developers: if you don't know something, just say it.
Experienced devs can spot fake confidence from miles away.
Experienced devs can spot fake confidence from miles away.
October 6, 2025 at 2:28 PM
Dear developers: if you don't know something, just say it.
Experienced devs can spot fake confidence from miles away.
Experienced devs can spot fake confidence from miles away.
Do tools like Claude Flow actually speed up your development?
My gut says you end up spending more time setting up flows and verifying output, vs. just writing clear specs and letting a single agent code.
Someone deep into the space, help me!
My gut says you end up spending more time setting up flows and verifying output, vs. just writing clear specs and letting a single agent code.
Someone deep into the space, help me!
October 4, 2025 at 8:00 AM
Do tools like Claude Flow actually speed up your development?
My gut says you end up spending more time setting up flows and verifying output, vs. just writing clear specs and letting a single agent code.
Someone deep into the space, help me!
My gut says you end up spending more time setting up flows and verifying output, vs. just writing clear specs and letting a single agent code.
Someone deep into the space, help me!
How do you actually check the quality of a dev subcontractor?
Code is easy to generate so impressive GitHub means nothing.
Asking about concepts and best practices works, but takes a lot of time and energy.
Code is easy to generate so impressive GitHub means nothing.
Asking about concepts and best practices works, but takes a lot of time and energy.
October 2, 2025 at 6:30 AM
How do you actually check the quality of a dev subcontractor?
Code is easy to generate so impressive GitHub means nothing.
Asking about concepts and best practices works, but takes a lot of time and energy.
Code is easy to generate so impressive GitHub means nothing.
Asking about concepts and best practices works, but takes a lot of time and energy.
/context with Claude Code is a lifesaver.
When context is over 80k:
1. Ask Claude to summarize the thread
2. Review it by hand
3. Start a fresh chat with the cleaned summary
From my experience, Claude gets unstable once you near 100k tokens.
When context is over 80k:
1. Ask Claude to summarize the thread
2. Review it by hand
3. Start a fresh chat with the cleaned summary
From my experience, Claude gets unstable once you near 100k tokens.
October 1, 2025 at 8:17 AM
/context with Claude Code is a lifesaver.
When context is over 80k:
1. Ask Claude to summarize the thread
2. Review it by hand
3. Start a fresh chat with the cleaned summary
From my experience, Claude gets unstable once you near 100k tokens.
When context is over 80k:
1. Ask Claude to summarize the thread
2. Review it by hand
3. Start a fresh chat with the cleaned summary
From my experience, Claude gets unstable once you near 100k tokens.
Tired of posting consistently to X?
...
Well, why are you posting?
Remind yourself of your goal, and move towards it.
...
Well, why are you posting?
Remind yourself of your goal, and move towards it.
September 30, 2025 at 12:05 PM
Tired of posting consistently to X?
...
Well, why are you posting?
Remind yourself of your goal, and move towards it.
...
Well, why are you posting?
Remind yourself of your goal, and move towards it.
Claude Code acting weird?
Read your CLAUDE. md. It could be outdated and have conflicting instructions to your codebase.
Read your CLAUDE. md. It could be outdated and have conflicting instructions to your codebase.
September 30, 2025 at 8:30 AM
Claude Code acting weird?
Read your CLAUDE. md. It could be outdated and have conflicting instructions to your codebase.
Read your CLAUDE. md. It could be outdated and have conflicting instructions to your codebase.
After a week of syncing dev work, the lesson is clear once again:
Planning is the most important phase of building.
Planning is the most important phase of building.
September 28, 2025 at 8:08 PM
After a week of syncing dev work, the lesson is clear once again:
Planning is the most important phase of building.
Planning is the most important phase of building.
Effective AI coding prompting 101
I'm creating a new page:
1. State the goal, reference style & data structure
2. Build the interface
3. Connect with mock API
👉 Then start fresh chat for backend.
Context is too polluted to continue with backend work.
I'm creating a new page:
1. State the goal, reference style & data structure
2. Build the interface
3. Connect with mock API
👉 Then start fresh chat for backend.
Context is too polluted to continue with backend work.
September 22, 2025 at 11:55 AM
Effective AI coding prompting 101
I'm creating a new page:
1. State the goal, reference style & data structure
2. Build the interface
3. Connect with mock API
👉 Then start fresh chat for backend.
Context is too polluted to continue with backend work.
I'm creating a new page:
1. State the goal, reference style & data structure
2. Build the interface
3. Connect with mock API
👉 Then start fresh chat for backend.
Context is too polluted to continue with backend work.
Vibe coding hack of the day:
Define strict data shapes for the UI components.
AI can make a mess in between, but your interface will always render.
Want even less errors? 👇
Define strict data shapes for the UI components.
AI can make a mess in between, but your interface will always render.
Want even less errors? 👇
September 16, 2025 at 4:40 PM
Vibe coding hack of the day:
Define strict data shapes for the UI components.
AI can make a mess in between, but your interface will always render.
Want even less errors? 👇
Define strict data shapes for the UI components.
AI can make a mess in between, but your interface will always render.
Want even less errors? 👇
Don't let users wait >150ms without feedback.
Even short action of 200ms you can break it up:
0–50ms → click feedback
50–200ms → subtle color change
200ms → start page animation
The app isn't faster, but it feels faster.
Even short action of 200ms you can break it up:
0–50ms → click feedback
50–200ms → subtle color change
200ms → start page animation
The app isn't faster, but it feels faster.
September 15, 2025 at 1:24 PM
Don't let users wait >150ms without feedback.
Even short action of 200ms you can break it up:
0–50ms → click feedback
50–200ms → subtle color change
200ms → start page animation
The app isn't faster, but it feels faster.
Even short action of 200ms you can break it up:
0–50ms → click feedback
50–200ms → subtle color change
200ms → start page animation
The app isn't faster, but it feels faster.