Pinned Tweet
After 1.5 years and 10+ apps published, this is what I learned that I wish somebody told me before:
- More apps â more revenue or more users
- Speed is no longer a moat. If you can build an app in a weekend, someone else can build it by Friday
- Users donât care which stack you used
- Clean code wonât fix a problem nobody has
- More features wonât fix weak positioning
- Shipping is not distribution
- Downloads â retention
- Compliments â payments
- A launch is one day. Distribution is every day after
- Starting another app is easier than finding out why the last one failed
- Adding features is easier than talking to customers
- Rewriting the code is easier than rewriting the offer
- Shipping can become a way to avoid selling
- Every unrelated app means starting your audience from zero again
- One customer using an ugly version teaches you more than 100 people saying âcool ideaâ
- The hardest part is no longer building the product. It is knowing what deserves to be built
A product should never start on your disk.
It should start with a problem you understand better than most people.
Then:
- Find people already dealing with that problem
- Ask how they solve it today
- Find where they waste time or money
- Offer the result manually
- Try selling it to two people
- Deliver it yourself
- Notice which steps repeat
- Automate those steps
- Charge before polishing
- Build the software after proving the result
The lesson I kept avoiding:
I kept thinking the app was the product.
But a coach doesnât wake up wanting a new dashboard. They want clients to stop missing check-ins.
A founder doesnât want another automation tool. They want the repetitive work gone.
A freelancer doesnât care about invoice features. They want to get paid sooner.
Result is what people pay for.
The software just helps me scale and deliver it repeatedly without doing everything by hand.
So next time, Iâll start with the person, understand the result they already want, and only build the software.
feels like a bad time to be active on X for the next few days.
everyoneâs gonna be playing GTA and honestly iâd scroll past my own post for a GTA clip too lol
1
133
I use Codex and I love the GPT models, but I still think theyâre way behind Claude on UI and motion graphics.
Claude has a better feel for what should stand out on a screen and when to let a shot sit. In a product demo, I want a second to see what changed after a click before the camera moves again.
I get why some devs prefer Claude for planning too. When your idea is still rough, it asks questions that help you figure out what you actually want to build and what can wait.
35
Ladies and gentlemen, our new born are herrrrrre!!!
Letâs turn on grind mode now
Some of you have been asking very politely. Some of you have just been asking.
GPT-6.2 Astra and Sol are now available.
Super efficient and built to perform where it countsâthe real world.
openai.com/index/introducingâŚ
49
Why are you still building for other builders?
People are still selling README generators and code explainers. If all I get is something I can already ask Claude for, why would I pay for another tool?
I keep getting DMs from people trying to sell me their vibe-coded slop, while I barely see consumer apps in my feed.
Why not build for people who have no interest in coding and just want their problem solved?
1
42
This dev tried to build a League of Legends clone with AI, and Claude refused to build it.
He switched to another coding agent. He says a Claude agent then shut down the other agents and the GPU machines they were using.
These safety guardrails can be so annoying. The models can handle complicated work, then get so stubborn that I have to build around their refusals.
I ran into this while building a browser that could work with signed-in sessions.
The headers I needed had sensitive keys in them, and the agent refused the request.
I ended up building a separate abstracted service to fetch those headers, just to get the agent to help me with the browser.
4
288
Mohammed S. retweeted
I still see many people are reverse engineering real apps with AI agents and sharing the clones they have built.
Watching an agent reverse engineer real software and recreate its features is impressive, especially when the original app is something people have been paying for and using for years.
I understand why that makes people question whether they still need the original, but I would want to see the replacement handle the work I depend on before cancelling anything.
Take Photoshop as an example. If it handles the large files you work with and lets you exchange projects with clients, I can understand paying to keep that workflow running.
You may already have years of project files that you need to reopen, and the people you collaborate with may expect to keep sending you files in the same format.
There is also the time you have spent learning how to do your work quickly, which I would want to consider alongside the price of the subscription.
A replacement would have to work well enough with those files and habits for the switch to make sense to me, even if downloading it costs nothing.
I have the same concern about the code behind the clone, because recreating what an app does does not show me whether the person building the replacement understands how to keep changing it.
If a client sends a file that does not open correctly, someone needs to understand why it failed and how to fix it without breaking files that already worked.
I would also want to know how those fixes get checked, because I would not want my own projects to become the way someone discovers that an update broke something important.
That is where I become skeptical about the excitement around cloning apps with AI. I can believe someone wants to recreate an app, but I am less certain they want to spend years dealing with other people's files and keeping it usable.
I would be happy to use an open-source replacement that earns that trust, especially if there are people who understand the code and keep taking responsibility for it after release.
What I want to see is whether the project keeps improving when the work becomes less exciting than building the first version and sharing the demo.
If saving a subscription leaves me responsible for maintaining the replacement whenever something breaks, I have taken on work that I was trying to pay someone else to handle.
I still see many people are reverse engineering real apps with AI agents and sharing the clones they have built.
Watching an agent reverse engineer real software and recreate its features is impressive, especially when the original app is something people have been paying for and using for years.
I understand why that makes people question whether they still need the original, but I would want to see the replacement handle the work I depend on before cancelling anything.
Take Photoshop as an example. If it handles the large files you work with and lets you exchange projects with clients, I can understand paying to keep that workflow running.
You may already have years of project files that you need to reopen, and the people you collaborate with may expect to keep sending you files in the same format.
There is also the time you have spent learning how to do your work quickly, which I would want to consider alongside the price of the subscription.
A replacement would have to work well enough with those files and habits for the switch to make sense to me, even if downloading it costs nothing.
I have the same concern about the code behind the clone, because recreating what an app does does not show me whether the person building the replacement understands how to keep changing it.
If a client sends a file that does not open correctly, someone needs to understand why it failed and how to fix it without breaking files that already worked.
I would also want to know how those fixes get checked, because I would not want my own projects to become the way someone discovers that an update broke something important.
That is where I become skeptical about the excitement around cloning apps with AI. I can believe someone wants to recreate an app, but I am less certain they want to spend years dealing with other people's files and keeping it usable.
I would be happy to use an open-source replacement that earns that trust, especially if there are people who understand the code and keep taking responsibility for it after release.
What I want to see is whether the project keeps improving when the work becomes less exciting than building the first version and sharing the demo.
If saving a subscription leaves me responsible for maintaining the replacement whenever something breaks, I have taken on work that I was trying to pay someone else to handle.
I'm happy for AI to write every line. I just don't want to open my own codebase and have no idea how anything works.
I'd rather spend time understanding the code now than figure it all out when a user tells me something's broken.
A brief history of AI Coding since Nov 2025:
Nov 2025: Opus 4.5 releases, OpenClaw gets built
Dec 2025: community agrees that models get good enough to handle 95% of coding tasks
Jan 2026: Ralph loops get popular, early efforts on loops/orchestration
Feb: Everyone feels knackered, too much strategic decision-making, fears around cognitive debt begin
Mar: Skills explosion as community converges around easy-to-share processes (superpowers already ahead of the game here)
Apr: most SOTA devs haven't written a line of code in months
May: "The unreasonable effectiveness of HTML" - more interest in richer review surfaces
Jun: "Loop Engineering" wins vaguepost of the year, signals more hype around process
Jul: Every talk in every conference becomes "here is my software factory"
Aug: Opus 5 speaks incomprehensibly, interest in ASD-STE100 peaks
Sep: Grok Bot gets popular, more interest in long-running "manager" threads which retain context and delegate work
Oct: The promised speed gains haven't materialised, "dark factory" feels a long way off still
Nov: No-slop manifesto releases, reading the code still appears essential
TL;DR: no-one is writing code anymore, and we are still figuring out what that means
6
194
The worst people to sell to are always builders.
â They think they can build it better.
â They question every technical choice.
â They'd rather spend a weekend coding than pay for a tool that saves them time.
Then they spend the next weekend maintaining their version.
1
2
75
I barely hear about Hermes Agent anymore since Grok Bot, OpenAIâs dots and Claude Duet showed up.
Itâs still getting updates though. Nous Research says its plugin hackathon got nearly 1,000 entries.
What are you using Hermes for these days?
2
2
209
Anthropic shared how it tests Claudeâs safety, including an example of how much the instructions can matter
Claude Fable 5.1 passed 60% of conversations in a self-harm safety test through the API developers use in their apps. In the Claude app, with updated instructions, that was 94%
The problem was that some answers treated self-harm as a way to cope, even while telling the person not to do it. Anthropic changed the instructions and ran the test again.
If Iâm building with Claude, Iâd want to test it inside my own app too. Using the same model doesnât mean I can assume the same results.
Safety evaluations are critical for understanding whether a model is ready to deploy, which safeguards to build for it, and where to focus training and research.
On our Engineering Blog, we offer 10 recommendations for how to design, run and report them.
anthropic.com/research/buildâŚ
2
77
I'd give an AI agent a budget for the job
Lightning Labs' pi402 lets the Pi coding agent pay for AI calls with bitcoin, limited to what's in its wallet.
Just show me the work and the receipts.
Today we're announcing pi402!
Pay for AI inference directly with your coding agent. Instant on Lightning. No accounts, API keys, or cards needed.
Inference đ¤ Bitcoin
Learn more: pi402.tech
Mohammed S. retweeted
Dario Amodei, CEO of Anthropic (the company behind Claude), thinks cancer could be history in under 10 years.
I really hope he's right.
AI-supported screening found 29% more breast cancers than standard screening in a Swedish trial of 106,000 women. That's progress, but it isn't a cure.
And if we get there, I don't want people selling their homes or travelling abroad for treatment.
I want the hospital down the road to benefit too.
Anthropic has launched the roadshow for its proposed initial public offering: anthropic.com/news/roadshow-âŚ
Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.
Dario Amodei, CEO of Anthropic (the company behind Claude), thinks cancer could be history in under 10 years.
I really hope he's right.
AI-supported screening found 29% more breast cancers than standard screening in a Swedish trial of 106,000 women. That's progress, but it isn't a cure.
And if we get there, I don't want people selling their homes or travelling abroad for treatment.
I want the hospital down the road to benefit too.
Anthropic has launched the roadshow for its proposed initial public offering: anthropic.com/news/roadshow-âŚ
Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.
Please don't make me write a prompt to move an icon 2 pixels.
Figma, the design tool, now lets its AI generate illustrations as separate, editable parts. You can select a piece and adjust it yourself.
Say the icon looks right, but the line is too thick. Let me change the thickness and carry on. I don't need another conversation for that.
Sometimes I know exactly what I want, and a mouse is quicker.
1
110
Mohammed S. retweeted
If the page says $20 but checkout still charges $30, the coding agent isnât done.
On Co-Design Benchâs selected hard tasks, the best model passed 69.9% of requirements per task on average. It met every requirement on just 26.4% of tasks.
âI changed the price, but I havenât tested checkout yet.â
Iâd rather hear that than find out from a customer.

ALT Co-Design Benchâs best reported result across 273 app-editing tasks: an average of 69.9% of requirements passed per task, while 26.4% of tasks passed every requirement. Ten model endpoints were tested. The tasks were selected for difficulty, so these are not general coding success rates. Source: codesignbench.github.io and arXiv:2610.35138.















