Atharv

Building cool things, writing code, and sharing what I learn. Traveling on the side.

[Atharv @atharvmudur]

Looking to #connect with people who are actively building products, enjoy conversation, and use AI as a tool, not a substitute for original thought.

I’m new on X, but not new to the space.
Say hi and let’s get to know each other!

[Atharv @atharvmudur]

If you're struggling with Claude or ChatGPT's tone,

tell it what NOT to optimize for.

For example:

“Don’t optimize for sounding impressive. Optimize for being specific and practically useful.”

That one line will make a big difference.

[Atharv @atharvmudur]

Models seem to judge things better when you make them compare two options instead of grading one alone.

For eg:

“Which of these two landing page hooks would hold up better with a cold audience, and why?”

I usually get a better answer from that than from “is this hook good?”

[Atharv @atharvmudur]

If you’re paying for Claude

Instead of asking:
“What are the best practices for onboarding?”

Ask:
“What are the 5 ways onboarding usually fails in practice?”

That one change will make the answer more concrete and the subscription worth it.!

[Atharv @atharvmudur]

Don’t ask Claude to explain a topic directly. Ask it to build up to it.

For example:

“First tell me what someone would need to understand before learning about kubernetes. Then explain kubernetes.”

That usually produces a much better explanation than jumping straight in.

[Atharv @atharvmudur]

Claude gets more honest with your ideas when you:

Ask it for the strongest case against the idea before asking whether it’s good.

“Give me the best argument against building this in Flutter before telling me whether it still makes sense.”

vs asking, “is this a good idea?”

[Atharv @atharvmudur]

Make the model list its assumptions before answering.

For example:

“Before answering, list the assumptions you’re making about why this launch failed. Then answer.”

It changes the output a lot because now you can actually see the scaffolding behind the answer.

[Atharv @atharvmudur]

I’ve gotten much better outputs when I separate criticism from rewriting.

For example:

“Read this like an editor. Tell me what feels vague, repetitive, or unconvincing. Then rewrite it.”

That tends to work much better than just saying “improve this.”

[Atharv @atharvmudur]

@KarenLouu @solarise_webdev I’m yet to share mine! I hope it’s not too late. There’s plenty people out there who are still discovering this each day so keep sharing!
Love the doodle space!

[Atharv @atharvmudur]

A pattern I keep coming back to is asking the model to define what a strong answer should include before it answers.

“List the essentials of a great product analysis, then dive into the teardown.”

That gives a much sharper result than asking for the teardown directly.

[Atharv @atharvmudur]

Prompting is not just about the question.
It’s also about the setting you quietly create around the question.
And sometimes that framing changes the quality of the answer more than you can expect.

[Atharv @atharvmudur]

An interesting thing I’ve noticed with prompting is that when you phrase a message like it’s continuing an earlier conversation, even if there wasn’t one, the model often goes deeper and gives a noticeably better response.

[Atharv @atharvmudur]

“I’m still not sure I fully understand how body recomposition works from our talk earlier.”

That framing seems to push it into continuation mode instead of one-shot answer mode.

[Atharv @atharvmudur]

*on a jog - speaks prompts into the watch*
This is making development such a commodity!

[Claude @claudeai]

You can now enable Claude to use your computer to complete tasks.

It opens your apps, navigates your browser, fills in spreadsheets—anything you'd do sitting at your desk.

Research preview in Claude Cowork and Claude Code, macOS only.

[Atharv @atharvmudur]

@kritikakodes Let’s connect!

[Atharv @atharvmudur]

@TechByTaraa Rust

[Atharv @atharvmudur]

@vivoplt I used to think the same until I got one. I think MacBook is equivalent to a PC setup than a windows laptop but with instant startup time, no need for WSL, and as a bonus the ecosystem. Also you can use Remote Desktop to log into a windows server if you wanted to

[Atharv @atharvmudur]

@adahstwt It’s not instant damage, but keeping your laptop plugged in at 100% all the time can ruin your battery long term. Congrats on your new MacBook. You can use ‘brew install battery’ and set a cap at 80% to keep the battery healthier.
https://t.co/ONiLiDJ5ol

[Atharv @atharvmudur]

@manpo_wang That looks really well made! I appreciate the well designed UI!
Ugh. I have so many bookmarks in folders and sub folders that I didn’t know I had.
You got a +1 user!

[Atharv @atharvmudur]

Looking to #connect with people who are actively building products, enjoy conversation, and use AI as a tool, not a substitute for original thought.

I’m new on X, but not new to the space.
Say hi and let’s get to know each other!

[Atharv @atharvmudur]

@AndyLibavius @RealProductGirl Mostly out of convenience, but I usually stick to the same tool till I complete a task in a project. But if I hit a wall, I switch to see if the other tool can approach the problem is a different way.

I haven’t observed any quality issues with tool switching midway yet.

[Atharv @atharvmudur]

@SunilSun56557 @AlexEngineerAI If you’re a beginner, use AI tools, but after that make sure to question everything and treat it like a teacher to build your basics. That way you learn both coding and prompting together, and you might also end up with a decent MVP along the way.