Atharva Jadhav

712 posts

Atharva Jadhav banner
Atharva Jadhav

Atharva Jadhav

@ThinkAtharva

Building AI & SaaS in public Shipping a Lovable prompt directory Sharing real builds, bugs & lessons IIT Bombay

Katılım Kasım 2024
189 Takip Edilen125 Takipçiler
Sabitlenmiş Tweet
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
I’m building a prompt directory for Lovable in public. No fake wins. No hype. Just real building: Vercel env vars breaking Supabase auth issues ESLint blocking deploys Missing assets crashing builds Sharing what breaks and what actually works.
Atharva Jadhav tweet media
English
6
0
12
743
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@incentivising This is strong but easy to misread now use frames without letting them use you.
English
0
1
1
145
Incentivising
Incentivising@incentivising·
Never accept the role others impose on you. People will constantly try to define your role, limits, and your place in the hierarchy. Over time, it will drain you of your very essence; you become expendable and cheap, lacking sight or direction unless granted. Refuse the premise entirely. If it's beneficial, then act within their frame, but never internalize it. Always view it as a means to an end rather than something that defines your identity. Successful people will never argue within established frameworks. Be unconventional and unique.
English
10
78
471
12.7K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@TheGeorgePu That is a fast way to burn out a founder now the systems should absorb demand not route everything through one human.
English
0
0
0
12
George Pu
George Pu@TheGeorgePu·
I was in Monaco when the calls started. 5am. Constant. Demanding. Our partnership manager told every customer to 'call George' when they need something. I came back from that trip and fired him the first day back. That's not how I want to run a business. Chaotic. Always on alert. No peace.
English
13
1
20
925
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@PaulSolt Same feeling here the bottleneck shifted to human judgment now the faster this gets the more valuable clarity and taste become.
English
0
0
0
143
Paul Solt
Paul Solt@PaulSolt·
Codex + ChatGPT Pro are getting fast. I’m already struggling to keep up with 2–3 agents. I’m the slowest part of the loop. OpenAI’s CFO described a scenario where a low-latency chip delivers coding at 5× the pace people expect. We’re not there yet, but you can see the trajectory.
Cerebras@cerebras

OpenAI🤝Cerebras openai.com/index/cerebras…

English
9
5
112
11.8K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@svpino This is the part people do not want to hear because AI just turns your existing discipline up to 11 for better or worse.
English
0
0
0
110
Santiago
Santiago@svpino·
I keep seeing the same pattern over and over again: Teams working on a relatively clean codebase, with good test coverage and documentation, are flying with Claude Code. Teams that are yolo'ing things are struggling to make AI work for them. Vibe-coding is great, but good luck getting far in a large, complex, poorly documented codebase. Sometimes, you should consider that the issue isn't with agentic coding but with your codebase. The reason you aren't making progress might not be the model or the prompts you are using. The issue is with the things *you don't have* in place before letting AI touch your code. AI amplifies whatever you already have. If your codebase is well-structured and tested, AI will help you move much faster. If your codebase is a mess, AI will help you create a bigger mess, also faster.
English
77
47
585
54K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@PaulSolt Speed is insane right now the real constraint is not shipping anymore its knowing what is worth shipping.
English
0
0
0
171
Paul Solt
Paul Solt@PaulSolt·
Codex with GPT-5.2 is a cheat code you can use. You can publish an app in a week. Move at the speed of thought! It just works.
English
9
4
97
7.7K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@Prathkum Syntax was never the bottleneck thinking was its like the LLMs just force us to admit the hard part is deciding what should exist at all.
English
0
0
1
106
Pratham
Pratham@Prathkum·
Ever since humans started writing code, they have been trying hard to eliminate this low-level work of writing syntax line-by-line. First we discovered high-level programming languages to make syntax much easier and readable-friendly, then we had low-code tools, and now LLMs that turn out to be the real thing developers have been looking for. By offloading the syntax to LLMs, we finally get to spend 100% of our cycles on the actual problem-solving.
Ryan Dahl@rough__sea

This has been said a thousand times before, but allow me to add my own voice: the era of humans writing code is over. Disturbing for those of us who identify as SWEs, but no less true. That's not to say SWEs don't have work to do, but writing syntax directly is not it.

English
36
23
198
19.9K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@forgebitz Hard to argue with that benchmark because If the tools were magic their flagship products would show it first.
English
0
0
0
61
Klaas
Klaas@forgebitz·
i don't believe any ai coding hype from microsoft if ai coding is so amazing how can linkedin be such complete garbage
English
32
2
70
5.6K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@paraschopra This is the window most people waste playing it safe its like the downside is tiny compared to the upside.
English
0
0
0
76
Paras Chopra
Paras Chopra@paraschopra·
if you're a teenager or young, your biggest strategic lever is that your cost of failure is basically close to zero. so use it to your advantage. take career risks. try building a startup. you'd have such a long adulthood to recover that a few years of failing here and there wouldn't make an ounce of a difference. but if even one of your crazy attempts works out, it'll change your life.
English
55
69
960
42.5K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@FlorinPop17 Exactly building is table stakes now now getting attention and keeping trust is the real moat.
English
0
0
0
10
Florin Pop 👨🏻‍💻
Florin Pop 👨🏻‍💻@FlorinPop17·
In a world where anyone can build anything, distribution and trust are the most important.
English
31
0
49
1.5K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@brankopetric00 This should be stapled to every microservices pitch deck the thing is you did not fix design you just multiplied it.
English
0
0
1
111
Branko
Branko@brankopetric00·
Your monolith isn't the problem. Your code is the problem. Breaking a badly designed monolith into microservices gives you: - 15 badly designed services - Network latency you didn't have before - Distributed debugging nightmares - 47 repos nobody understands You didn't solve complexity. You distributed it. Now instead of one codebase to fix, you have 15 services pointing fingers at each other. Microservices are an organizational pattern, not a performance optimization. If your team can't build a clean monolith, they definitely can't build clean microservices.
English
10
8
90
5.7K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@mattpocockuk Yeah once you have felt that feedback loop everything else feels like ceremony now its like abstractions age fast when automation gets trivial.
English
0
0
3
708
Matt Pocock
Matt Pocock@mattpocockuk·
It's crazy how after discovering Ralph this stuff feels a bit quaint All this advice can be automated away with a few lines in a bash loop
Jon Kaplan@aye_aye_kaplan

My top 3 tips for coding with agents: 1. Always start with Plan Mode. It's better to iterate in natural language and then execute once you know what the agent is going to do. This will save you time, effort, and tokens! 2. Start new chats frequently. Remember that your role is to point the Agent in the right direction to make the changes you need. If you change topics, the context window will get muddied. You will also be spending more tokens on longer chats. 3. Leverage AI to do your code review. If you know the failure case, ask a model. One prompt I often use is "scan the changes on my branch and confirm nothing is impacted outside of my feature flag". As a safety net for everything outside this issues-you-expect umbrella, use Bugbot.

English
64
34
872
186.5K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@rywalker This is exactly it the generation is cheap but trust is expensive so you buy it with tests and guardrails.
English
0
0
1
92
ry
ry@rywalker·
people ask me all the time "how do you prevent ai generated code from being garbage at scale" my take: what if you spend 3x as many tokens on defense as you do offense deterministic checks, guardrails, automated testing... let the ai generate, but verify everything in a defensive loop
English
29
12
85
7.8K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@JustJake Yeah that debate always felt artificial now it is just about taste and intent not who touched the code.
English
0
0
1
175
Jake
Jake@JustJake·
My favorite thing about Claude automating coding: We never have to have the “Should designers code” discussion ever again
English
7
1
56
5.9K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@jamonholmgren Same here its the fastest way to stay current without doomscrolling now the value is in curating and sharing not just reading.
English
0
0
2
28
Jamon
Jamon@jamonholmgren·
Love it or hate it, 𝕏 has been the primary way I have consumed ideas and concepts about how to use AI developer tools. I share the best of them with my team on Slack. It's absolutely indispensable for this.
English
22
3
150
6.7K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@aye_aye_kaplan Plan mode is the real unlock now the actual thing is treat agents like juniors who need clear intent before they touch anything.
English
0
0
2
109
Jon Kaplan
Jon Kaplan@aye_aye_kaplan·
My top 3 tips for coding with agents: 1. Always start with Plan Mode. It's better to iterate in natural language and then execute once you know what the agent is going to do. This will save you time, effort, and tokens! 2. Start new chats frequently. Remember that your role is to point the Agent in the right direction to make the changes you need. If you change topics, the context window will get muddied. You will also be spending more tokens on longer chats. 3. Leverage AI to do your code review. If you know the failure case, ask a model. One prompt I often use is "scan the changes on my branch and confirm nothing is impacted outside of my feature flag". As a safety net for everything outside this issues-you-expect umbrella, use Bugbot.
English
66
39
779
343.6K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@anshnanda This is actually useful not just another dashboard actually turning messy reviews into concrete actions is where AI really earns its keep.
English
0
0
0
48
Ansh Nanda
Ansh Nanda@anshnanda·
Just made an app that allows me to analyze all our App Store reviews with AI and uncover deep insights. Highly recommended for all app founders.
Ansh Nanda tweet mediaAnsh Nanda tweet media
English
11
1
26
2.6K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@burkov This is mostly true now once you care about real code quality the differences stop being subtle fast.
English
0
0
0
445
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@RealIanBotes Pretty much yeah. If the tool is outdated my brain checks out and starts thinking how to rebuild it better.
English
0
0
1
8
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
How are we going to do this ?? Sitting in class, half listening, half thinking about how this actually works. Still figuring it out.
Atharva Jadhav tweet media
English
1
0
4
82
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@rakyll This plays perfectly with AI driven tooling its like small sharp pieces compose better when generation and refactors are cheap.
English
0
0
0
455
Jaana Dogan ヤナ ドガン
Jaana Dogan ヤナ ドガン@rakyll·
Developing small, composable programs is once again yielding an advantage. The UNIX philosophy prevails.
English
39
112
1.4K
82K
Atharva Jadhav
Atharva Jadhav@ThinkAtharva·
@haider1 That shift feels real for a lot of us now the leverage moved from typing code to shaping intent clearly enough that machines dont screw it up.
English
1
0
1
113
Haider.
Haider.@haider1·
confession time: i haven't written real code since claude code and codex came out, except for a few config changes and i probably won't write much code anymore. maybe a few one-liners here and there, but with the right context and a solid workflow, cc or codex can handle almost everything now so my main job has shifted to turning business requirements into inputs they can actually execute
English
34
9
136
8.7K

Keşfet

@incentivising @TheGeorgePu @PaulSolt @svpino @Prathkum @forgebitz @paraschopra @FlorinPop17