softDev23 @softdeveloper23

AI is making it easier for me to build 10 products this year. Can it write code faster? Yes. Can it help find bugs? Yes. Can it tell me which idea people will actually care about? Not reliably. Can it make anyone use what I build? Definitely not. Execution is getting cheaper, but judgment is becoming the expensive part. What part of building still feels hard to you?

0 0 0 5 0

View Details

softDev23 @softdeveloper23

a day ago

@danellisona Yes, cause I keep thinking that at some point, my posts will start to gain traction. It's how YouTube and SEO work, so I don't really see any difference here. Am I wrong thinking this way?

0 0 1 42 0

View Details

softDev23 @softdeveloper23

a day ago

Exactly. Even with the extra reset they gave us last Saturday, I’m already close to running out of GPT-6 Astra usage again today. My regular limit does not reset until this Saturday. The model is excellent, but it is hard to build reliable workflows around something I cannot reliably access. Another temporary reset is not enough. The usage limits need a real resolution.

0 0 0 69 0

View Details

softDev23 @softdeveloper23

a day ago

@skpnky I think so. I built Think Drink, Pluck, and Brain Dump and made them free, not because they have no value, but because I thought people could genuinely use them without another subscription or paywall. Not everything worth building has to become something you charge for.

0 0 1 16 0

View Details

softDev23 @softdeveloper23

2 days ago

One part of ToolGrad clicked for me: The model creating the training examples does not have to be smarter than the model learning from them. A person writing a good practice worksheet does not need to outperform every student who studies it. That makes “self-improving AI” sound a lot less mysterious. I explain what ToolGrad actually changes, and what it does not: youtube.com/watch?v=EJimIG…

0 0 0 6 0

View Details

softDev23 @softdeveloper23

3 days ago

Teaching AI backwards might be more useful than it sounds. ToolGrad builds working tool steps first, then creates the practice question around them. That’s training material, not your chatbot teaching itself mid-conversation. I explain the difference here: softdev23.com/toolgrad-expla…

0 0 0 11 0

View Details

softDev23 @softdeveloper23

4 days ago

@migarci2 It definitely exists with me, Miguel! 😅

0 0 0 1 0

View Details

softDev23 @softdeveloper23

4 days ago

Electric cars gave us range anxiety. GPT-6 Astra has given me usage anxiety. I keep checking my remaining limit the same way I used to watch my battery percentage on a long road trip. “Can I finish this project, or am I going to be stranded before the next reset?”

1 0 0 14 0

View Details

softDev23 @softdeveloper23

4 days ago

@vikitweetus Eclipse was my first IDE, and my first language was Java.

0 0 0 6 0

View Details

softDev23 @softdeveloper23

4 days ago

That disconnect is real. On X, it feels like everyone is rebuilding their entire life around the latest AI release. Then I talk to people outside this bubble and realize most are still figuring out where AI is genuinely useful. A good reminder that the timeline moves much faster than the real world.

0 0 0 30 0

View Details

softDev23 @softdeveloper23

4 days ago

@Improo_ Happy Birthday, Ramon_Ray! I just followed you. Let's connect. 🤝

0 0 0 2 0

View Details

softDev23 @softdeveloper23

4 days ago

Computer use is the part that has changed my workflow the most. A strong answer helps, but the bigger gain comes when Codex can inspect the actual interface, make the change, verify the result, and keep moving without another manual handoff. Speed matters. Finishing the whole workflow matters more. Now I just want pricing and usage limits that make that workflow sustainable.