The recent meltdown in the indie hacker community about how no one buys software anymore really shows how much a bubble there is on X 😅
Still tons of opportunities if you solve real problems, esp if you go into niches.
But the easy “I had an idea yesterday” is gone
True. I’ve been going on and on about this for years 😂
I consistently underestimate how long the tail or normies & enterprises that exist.
It can feel “too late” to get into AI already but probably 0.1% really use agents 😂
Traveling in Bavaria / South Germany is blowing me away. Feels like Switzerland just very cheap, no marketing, never seen IG reels here, no people/crowds and very family friendly.
Are Germany gatekeeping this??
Lars Holdgaard retweeted
Replying to @danielrakh @bengold
Engineers, 2 designers, 1.5 product managers and me
Hate or love Dubai, this is not happening.
Left in 2022 and was back in January. It’s growing FAST.
It’s only getting started.
Mark my words give it a few years and Dubai will feel like an empty desert filled with buildings.
The charm of Atlantis The Palm is fading fast. Visitor numbers are reportedly down by nearly 40%, and the place often feels almost empty now.
It’s become a classic case of “been there, done that.” Dubai has turned into a one-time visit destination overpriced, not very family-friendly, and heavily dependent on luxury appeal.
Once people check it off their list, they rarely feel the urge to return. Over the years, the excitement has worn off.
Today, travelers value money more than ever, choosing simplicity over excess and thinking twice before spending big on luxury.
@ATLANTIS @visitdubai @VisitDubai_IN
This is really THE question.
Right now a million features are not being built because devs are giga busy and roadmaps are full of stuff.
Will AI just mean we get to raise the floor of products OR industry is fucked ?
When asking developers if they use AI:
On X: “I’ve 8 terminals running independently in worktrees in a Ralph loops 24/7!”
Normies: “he he GitHub copilot once made one unit test for me”
This could very well be the a huge moment for non-coders!
I showed this to my wife who runs a home inspection business in Portugal, where making the reports is the most time consuming
Pretty sure she can have a prototype of automating those reports now running in few hours.
Of course this was possible before; but the UI/UX was missing in these tools. I showed this to her and she was like "okok now I get it!".
(she absoliutely didn't get my love for the Claude Code in the terminal :D )
Pretty crazy - didn't tweet for 8 months.
My last tweets was about trying OpenAI Codex/Devin agentic coding.
Now I run 6 Claude Code agents in 6 seperate worktrees most times when building.
How the fuck did the world change to quick lol
(github commits to brrrrrrrr)
First project is setup with OpenAI Codex agentic AI coder!
Very excited!!
But "feel the AGI"? Yet the setup is running crazy scripts & a horrible editor for secrets. Looking forward to this being some questions after reading my deployment files & upload of secrets file
1/4: Update on Devin AI from after using a few hundred dollars.
HOW IT FEELS TO USE:
These types of agents will become game-changers, but for now feels like a junior dev on some very aggressive drugs.
It feels like underneath @cognition_labs rely on a cheap / old core model.
3/4: THE PROBLEM:
Because the underlying models seems to be a very low-quality LLM, much of the code generated is incredibly retarded or naive.
If it can rely on existing examples? You get good results
If it has to re-invent the wheel? It just makes up the most naive code
Time to start trying out Devin AI by @cognition_labs - skeptical to say at least, but let's see if we can get some pull requests auto-created!
Ok. Trying on a .NET 4.8 solution that has to run on Windows.
Devin can spin up an Ubuntu computer/browser and probably build the application, but won't work on Windows.
So far the results are horrendous. Basically an endless chain of me saying: "it doesn't build? wtf?"
Ok, for any big decision in my life:
1) ask chatgpt to break down the questions that should be researched
2) start a deep-research for each
3) wait 30 min for reuslts
4) copy paste all the research
5) ask o3 to summarize and make a full compact plan
Game-changer









