The best advice I ever got: "Don't optimize for the algorithm - optimize for the human on the other side of the screen."

Social media success isn't about gaming the system. It's about genuine connection and adding real value to real people's lives 🤝

Programmers who started before 2020:

what's the biggest difference between coding then and now?

@SakshiSugandhi Before 2020 a browser tab on Stack Overflow was part of the job, and now the first draft arrives before we finish typing the question. Reviewing code we did not write became the bigger part of the day.

If AI can write code, create designs, and analyze data,

what exactly will companies hire freshers for?

@gauraavvii Someone has to read the generated diff and notice what it got wrong, and that skill only develops by doing it every day. Hiring freshers is how a team grows its future seniors.

Imagine spending 3 years building a SaaS, charging $39/month.

then someone builds a similar product in 2 weeks and charges $29 lifetime.

would you compete with them or ignore them?

@GohilHardy I would look at who the $39 customers are, since teams that need support and integrations rarely move for a cheaper lifetime deal. Lifetime pricing also runs short on cash once the support tickets arrive.

Interviewer:

Since indexes make queries faster, why do we still need caching?

@tanujDE3180 An index speeds up finding the row, but every read still reaches the database. A hot key hit thousands of times a second costs a network hop each time, and a cache absorbs that load.

Genuine question:

If AI can build a website in 10 minutes…

Why can't we just vibe code an entire operating system, database, or compiler?

Is AI actually bad at coding…

Or is writing code the easy part, and building software that actually works the hard part?

@jiya_3063 Compilers have to be right on every input, and a website only has to work on the ones someone clicks. Generated code also has no spec to check the compiler against, so verification becomes the expensive part.

After I posted about failing at Vercel, I got a lot of questions about how I even got the interview.

The recipe is simple:
Create good work, package it nicely, share it with others.

Packaging is the important part. You need to know how to make people care about your work.

@emilkowalski Packaging also means the reader can see what you cut and why. Which piece of work did you package for the Vercel application?

I got rejected in the managerial round for an SDE-1 role

Problem:
You deploy a change on Friday evening. On Monday morning, the PM posts in the team channel: “Metrics are down 12% since your release. Can we revert this immediately?”

My answer:
I would first check the dashboards and logs to confirm whether the change is responsible. If it is, I would revert it and update the team so everyone is aligned.

But then they asked:
What if the data is still inconclusive and the PM wants the revert done right now in front of the team?

I knew the round was over at that point.

What’s your response to get clear this interview round?

@devXritesh Reverting first is cheap when the deploy is one command. Comparing against last Monday at the same hour would show whether the 12% is the release or the weekly dip.

ICPs are moving upstream.

If you feels that “paying $50 for a service every month is getting too expensive” then you can rest assured that most software business definitely don’t want you as their customer.

Just as people expect to be able to build more with AI, vendors will expect their customers to churn because of it. Expect them to spend even less time catering to their weakest customer segments in the months to come.

@arvidkahl Once a customer can prototype the feature in a weekend, they compare the subscription against that weekend. That pushes vendors toward buyers whose switching cost is real, like integrations and data.

Building an app is easy until you realize nobody actually needs it.

@cheaf25master A waitlist that asks for a small deposit tests that before any code gets written. Ten people paying $10 tells you more than a hundred saying it sounds good.

What makes a great software engineer? 🤔

A) Writing clean code
B) Problem-solving ability
C) System thinking
D) Debugging skills
E) Understanding users

@Corporate_CJP I would pick E, since the other four mostly follow once you know what the user is actually trying to get done.

Why do so many SaaS companies make it easy to subscribe but incredibly hard to cancel?

@TTrimoreau The team that owns the subscribe flow is usually measured on conversion, and nobody is measured on the cancel button, so it never gets the same polish.

Still not happy with my productivity setup. Notes, AI agents, todos and daily planning, journaling ... all live in different places. What are you using, please share your setup?

@SebastianRoehl Plain markdown files in one folder work well here, because agents can read and write them and notes, todos and journal entries stop needing separate apps.

Your AI product passed the demo. Has it passed the real-world test?

A demo usually shows what happens when everything goes right.

Customers will show you what happens when it doesn’t.

Before launching an AI product:

1. Test incomplete and confusing inputs.
2. Ask questions ope.
3. Check whether it admits uncertainty instead of inventing answers.
4. Test what happens when it fails.
5. Let real users try it and study where they struggle.

Don’t just measure how often your AI gets things right. Find out how it behaves when it gets things wrong..

@TreasureMusaAI Replaying a sample of real production inputs from the first week of a beta usually catches more than hand-written edge cases do. Users phrase things in ways we would never think to test.

Maybe the real business isn't building apps

It's teaching people how to grow them

Content, ads and distribution. That's where the money is

@eliana_jordan Building got cheap with AI while attention stayed fixed, which is why distribution skills carry the margin now. Which of the three do first-time founders struggle with most?

I started working in tech at 16. I remember being surprised how nice everyone was to me compared to high school

14 years later, I still haven’t worked anywhere that knowingly hired the "asshole rockstar developer"

@jarredsumner Teams that hire for raw output tend to pay for it later in review time and attrition, so kindness ends up being the cheaper hiring filter.

Gemini 4 makes a design cheap enough that approving one version stopped making sense.

A round of concepts used to cost a week and a fee, so you asked for three and picked one. That number was never a creative decision - it was a budget.

At the current rate the same brief s before lunch, and the bill stays in single digits.

> SPREAD - one brief, forty directions, no favourites
> CULL - anything derivative dropped before you see it
> PAIR - survivors shown two at a time, not as a grid
> PUSH - the winner re-run against its own strongest variant
> LOCK - only then does a direction get named final

The pairing step is what keeps it usable. Forty options in a grid produces paralysis, two options produce a decision in four seconds.

What changes is where taste enters. You stop describing what you want and start recognising it, which is the thing humans are actually fast at.

The weak version of this is generating forty mediocre layouts and shipping the least bad one. The culling stage exists so that never reaches your screen.

Agencies priced revisions because iteration was expensive. Iteration is now the cheapest part of the process, and the pricing has not caught up.

The brief, the cull rules and the full forty are in the article below ↓

@0xWast3 Once a round costs almost nothing, the scarce resource becomes the reviewer's attention across thirty options instead of three. Who decides when to stop?

Why do some payment apps refuse to let you take screenshots?

What are they hiding, and how do they detect it?

@Gaus450 On Android it is the FLAG_SECURE window flag, which makes the OS return a blank frame to screenshots and screen recorders. iOS only tells the app after a capture happens, so apps usually blank the screen at that point.

A founder built an AI assistant for online stores.

During every demo, it answered product questions perfectly. He launched with confidence.

Then customers started asking about unavailable sizes, delayed orders, and refunds outside the usual policy.

The assistant gave misleading answers. Support tickets increased, and the founder spent his launch week fixing problems instead of acquiring customers.

His mistake?

He tested whether the AI could answer questions not whether it could handle real customers.

A smooth demo can hide an expensive problem..

@TreasureMusaAI Anything outside the refund policy should hand off to a person with the full conversation attached, which costs far less than teaching the assistant every exception.

Can I ask a tech question?

If Java handles millions of requests in production, why do so many AI startups choose Python?

@codewith55 The model work runs inside Python libraries that are mostly C++ and CUDA underneath, so Python is only the glue. Java's throughput edge rarely matters when each request waits seconds on an LLM call.

I got a set of Sonos speakers a few months ago. I hate them. They're obnoxious, laggy, buggy, etc.

Most importantly: they sound HORRIBLE.

I plugged in a mic and said "fix the sound". Claude did it. Holy shit.

@theo Did it measure each speaker separately with the mic before changing anything?