Akshar Patel

Senior Product Engineer

[Akshar Patel @aksharpatel_in]

Skills are useful. I use a lot of them.
⠀
But the biggest unlock I've found with AI isn't a better prompt.
⠀
It's handing it a sample project.
⠀
Prompts describe. Samples define.
⠀
Stop writing 400 words. Just paste the reference.

[Akshar Patel @aksharpatel_in]

Belief in AI gets you average output.
⠀
Debate with AI gets you leverage.
⠀
When's the last time you pushed back and got a better answer than the one it called "final"?

[Akshar Patel @aksharpatel_in]

AI isn't built to be right.
⠀
It's built to be helpful.
⠀
Helpful means agreeable, until you make it think harder.
⠀
The best prompt isn't a question. It's a counter-argument.

[Akshar Patel @aksharpatel_in]

Challenge it once and you'll often get:
⠀
"Your reasoning is actually very solid."
⠀
That's the receipt.
⠀
The "Final answer" was never final.
⠀
It was just unchallenged.

[Akshar Patel @aksharpatel_in]

Most people use AI wrong.
⠀
Not the prompts. The posture.
⠀
They ask, AI answers, they accept.
⠀
The best output comes after you push back.

[Akshar Patel @aksharpatel_in]

AI made starting easy.
⠀
It didn't make finishing easier.
⠀
The hardest 20% — the debugging, the judgment, the call only you can make — is still 100% on you.
⠀
Most people burn the day on the easy 80% and call it work.

[Akshar Patel @aksharpatel_in]

The people actually shipping right now aren't the ones with the best prompts.
⠀
They're the ones who know when to close the tab.
⠀
AI is a multiplier on doing.
⠀
Not a substitute for it.

[Akshar Patel @aksharpatel_in]

AI gives you a cheap completion hit at the start of a task.
⠀
You feel like 80% is done.
⠀
But the start was never the hard part.

[Akshar Patel @aksharpatel_in]

A few things I keep reminding myself:
⠀
Benchmarks measure capability. Workflow measures usefulness.
⠀
The best users push back on the model.
⠀
If your gut says your answer was better, it probably was.

[Akshar Patel @aksharpatel_in]

The frontier isn't a benchmark score.
⠀
It's whether the model can out-reason you on your actual problem.
⠀
Not yet. Not consistently.

[Akshar Patel @aksharpatel_in]

Frontier models aren't where the hype says they are.
⠀
Actual response from GPT 5.5 yesterday:
⠀
"Your answer was clearly better than mine. I gave an overcomplicated, generic response instead of identifying the simplest effective solution."
⠀
This happens often.

[Akshar Patel @aksharpatel_in]

Experience used to mean shipping faster than everyone else.
⠀
Now it means knowing what not to ship.
⠀
AI didn't kill seniority. It moved it up the stack.

[Akshar Patel @aksharpatel_in]

AI can do everything now.

But you still decide: low, medium, or high reasoning.

High reasoning on everything? Enjoy the bill.

Low on complex work? Enjoy the garbage output.

Your experience didn't disappear.

It just moved to model selection.

[Akshar Patel @aksharpatel_in]

The gap between "impossible" and "routine" is shrinking fast.

What models couldn't do 12 months ago? Easy now.

What they can't do today? Give it 12 months.

[Akshar Patel @aksharpatel_in]

Judging AI by what it can't do today is like reviewing the internet in 1995.

"It's slow. It'll never replace anything real."

We know how that aged.

[Akshar Patel @aksharpatel_in]

The era of on-the-fly apps is already here.

Try this in Claude: "Show me how compound interest works"

It doesn't explain it.

It builds you an interactive tool.

No app store. No install.

[Akshar Patel @aksharpatel_in]

The gatekeeping era of software is ending.

Not because vibe coding is perfect.

Because it's good enough and getting better fast.

[Akshar Patel @aksharpatel_in]

Most engineers optimize a component.

Product engineers see where the system breaks.

The difference isn't skill. It's scope.

[Akshar Patel @aksharpatel_in]

The AI trap no one talks about:

The easier AI makes things, the easier it is to stop learning.

But AI only amplifies what you already know.

Zero times anything is still zero.

[Google Research @GoogleResearch]

Ever wonder if a computer can simulate brain activity?

Google Research just curated a new #NotebookLM notebook asking just this.

It features sources on:
✅ Predicting neural activity with AI
✅ Nanoscale brain mapping
✅ Synapse-level reconstructions

Start exploring the frontier of neuroscience today:
https://t.co/oQJ5mMHv6C