Sushil Kumar

1/ I’ve been building a bunch of things lately after getting back to coding in the last few weeks. But one problem kept sitting in the back of my mind: agent workflows were still too wasteful on code search/retrieval.

@petergyang Shipped products with zero docs. Also burned months on specs nobody opened. The honest part? Docs were never the work. They were the excuse to not ship. Agents just remove the excuse. Now you have to actually know if the thing is worth building.

I hope AI agents will help us all stop obsessing about: - Specs - Designs - Issues - Roadmaps - Strategy docs - Spreadsheets and instead let us all obsess about improving the end user product and rapid iteration loops with users. Product development is much more fun that way.

@karpathy LLMs made dependency hell worse, not better. Now you don't even read the code you're running. You just "pip install openai" and trust 47 strangers. This attack was inevitable. The only surprise is it didn't happen sooner.

Software horror: litellm PyPI supply chain attack. Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords. LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm. Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks. Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages. Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.

@garrytan The tools that quietly power most of my automation have zero cool factor. Plain text files. Shell scripts. Makefiles. They're boring, interoperable, and still running 5 years later.

In the AI revolution, low status and useful is where the alpha is. Markdown files look like shit and have basically zero status. But they are insanely useful. Humans can read them, models can read them, agents can write them, diff them, transform them, chain them.

@ctatedev The future is background agents doing the heavy lifting around the clock. Humans shouldn’t have to babysit workflows. They should open one unified UI, review the results, and step in only when an agent needs input. One system. Parallel agents. Control when you need it.

~100% of my dev is done in sandboxes in the cloud Highly recommend it: - Unlimited parallel agent sessions - My local machine stays safe - Can work from anywhere - Can close laptop - Lap stays cool Interesting idea to visualize with Kanban

@levie Every technological wave has its chorus of "this time is different." Focus should be on building for that transition, not for fear.

“AI exposed jobs may increase hiring and attract higher wages. It all depends on a) elasticity of consumer demand and b) number of AI exposed tasks in a job.” This is a key point. We’re going to see lots of AI automation emerge that has the opposite effect that we expect, because the cost of doing something goes down and greater demand for that service exists at lower prices. Take a *very* simplistic example in agentic coding to see what happens when you can dramatically increase output per $ of engineering budget. Before AI, a mid-sized company or team within a large company has a project they want to build software for. It takes 50 engineers to fully resource the effort, but the project doesn’t provide the ROI to fund it compared to other initiatives. Or the company knows its expertise isn’t in building software so it’s not even worth starting. So they hire 0 engineers, and don’t start the project. Now, AI agents make it possible for this to be a 10 engineer problem. All of a sudden the ROI calculus immediately changes on starting up the project. So now instead of hiring 0 engineers to do the project, the company hires 10 with AI agents. This has endless implications in coding, in particular, because coding can now have impact for anything from doing internal workflow automation, systems integration, data analysis, as well as customer-facing product innovation. By bringing down the cost of writing code, we can just begin to use it for far more. This will likely play out in a number of other job families as well, where lowered costs or higher output will lead to more demand. Now, not all of this will be smooth. For instance, there may need to be some reallocation of talent across the economy to move from some places of excess supply to places of lower supply. This could be bumpy at times, but the dynamic holds.

@Annu_NexraAI A. Every time. You don't find product-market fit in a lab, you find it in the wild with real users telling you what they actually need. The "polished" version usually has features nobody asked for.

Honest question for builders: Do you prefer to A) ship fast (V1 → V2 → V3) and improve with feedback or B) build the full product first and launch when it's polished? Curious what actually works better in the real world.

@OmriBuilds Maybe. But the moat won’t be ‘10 agents running’. everyone will have that soon. Moat is taste + distribution + speed of shipping when the agents break. AI gives leverage, not inevitability.

The next $1B startup might be built by a 18-year-old running 10 AI agents on OpenClaw.

Formative years of my career were at Unacademy. Wild ride, ups & downs, all of it. Glad the Board carved out an ESOP buyback even in tough times. It means those I built alongside, including me, will get something tangible out of it.

6/ If you want to try it: pip install know-cli Repo: github.com/sushilk1991/kn… PyPI: pypi.org/project/know-c… Ask your agent to install it and use it in real tasks for a while, then ask for feedback. The difference shows up pretty quickly.

5/ What started as a side utility is now a core part of how I work with agents: fewer tool calls, better retrieval, less token waste, better output quality. Still shipping fast and improving it week by week.

1/ I’ve been building a bunch of things lately after getting back to coding in the last few weeks. But one problem kept sitting in the back of my mind: agent workflows were still too wasteful on code search/retrieval.

@gauravmunjal This kind of honesty is rare. Having seen parts of this journey up close, it’s been quite a ride, and knowing you, the real "past work will look like a footnote" phase is still to come. :)

But more importantly, we are back to following the mission that we started on Unacademy with, which is just making great educational products and making them accessible. In hindsight, the best year for Unacademy was not 2021, it is 2025. We take pride in the fact that Unacademy changed online education for good and invented a product and a YouTube marketing playbook that became an industry norm. A few bad strategic calls set us back by a few years, but I am glad that we are stronger that we ever were.

Unacademy turns 10 today I thought it's a good occasion to jot down some thoughts on this day about the wild ride that we have gone through. Ten years ago, we started with a simple mission: to empower great educators and make their content accessible to everyone. This is when the company was started. Unacademy started as a YouTube channel in 2010, 15 years ago. When I was in the 3rd year of my college. And I started making videos on computer science to help my friends. It continued as a side project with @RomanSaini joining in 2014, and us blitzscaling the YouTube channel. His UPSC videos would get millions of views that helped us in becoming the number one education channel in the country. We thought that the biggest problem then was that the smartest people are not teaching, and we need a platform where we'll get the best educators who'll create videos for learners for free. Kind of like what Twitch had done to gaming, we wanted to do that to education. Create a YouTube-like platform. And on 10th December 2015, Unacademy was launched. For the first four years, we kept adding more and more great educators to the platform, which would lead to millions of learners joining the platform. There was a point in 2018 when we were doing millions of views per month on our own platform, with Unacademy being the number one education app on the Play Store. And it was all organic growth, because we had a rule to not spend money on performance marketing. These were exciting times. Roman would personally go to the houses of potential educators to convince them to start teaching on an Unacademy. That's how some of the best educators joined the platform. From the beginning, we always thought of ourselves as a tech company operating in education. Unlike every other player in the market which behaved like an education company using tech to enable them. That's why we had features like streaks, knowledge hats, and a lot of gamification built in for educators and learners. Educators would get massively addicted to the knowledge hats which the learners would gift them on crossing certain milestones. Even till today, some of the biggest educators that we have are the ones who grew on the free platform that we had built as our first product. Then, in 2019, just one year before Covid hit, we launched a subscription product where learners could buy a subscription and get access to live classes from the best educators for their exam. The product was an instant hit. From zero revenue in January 2019 to $1.8M revenue in September 2019, we were one of the fastest growing companies to get to almost $20M Bookings ARR in nine months. The next one year was crazy because our revenue kept scaling and especially post-Covid, we were growing even faster. Almost 1M Paid Subscribers were now enrolled on Unacademy. There were three back-to-back funding rounds. From being a $100M valuation at the start of 2019, we were at a $1.5B valuation by September 2020. We had spent less than $50M to reach a valuation of $1.5B. But soon the distractions began. In the next few months, we did another round, ended up raising another $440 million. Totalling the fund raise to more than $700 million in just less than two years. We thought that what we are seeing in COVID would sustain forever and that's the new reality. And started burning a lot of money to acquire a lot of market share, without realizing that education business is slightly different from a normal tech business, and here your first transaction is basically your LTV. We had become the number one test prep brand in terms of recall. We were the number one test prep player in terms of online business, which had scaled to almost $100M in revenue.

@shobhitic @monali_dambre @swaroop_g1 Bhai please, not all Biharis worship sattu. Some of us took the pain so you don’t have to 😤😂

@monali_dambre Is it @Sushilk91 @swaroop_g1 ?

Someone should make protein kachori please 😩😩

Happy to announce that @sumjain is now the CEO of our Test Prep Business. I have known Sumit since 12 years when he was the CEO of CommonFloor and he acquired my first company Flatchat. Sumit joined us as Co-Founder in 2020 has been instrumental in significantly improving the Unit Economics of our Test Prep Business and more importantly the Learner Experience. Looking forward to many more years of Partnership :)

🌍 Our mission: make 1:1 learning accessible to every learner on Earth. Today, Airlearn launches the world’s first Iconic AI Tutor. Starting with languages, and soon, everything 🚀

Startups have to rediscover zero bug bounce Every few sprints get to zero P2 bugs before you start new feature work. Otherwise your product will be a low quality mess forever. One part of great craft is getting to zero known bugs regularly.

Ladies and gentlemen, I've officially posted my way to the top: I'm joining @X as Head of Product. 𝕏 is the most important social network in the world. It's where internet culture originates and where the world's most influential people convene. Finding my community and building an audience on 𝕏 has impacted my life more than any single thing: it's unlocked friendships, professional opportunities and it's even where I met my girlfriend. While I already spend every waking hour on this app, I'll now be spending that time helping others unlock that same value. And we'll certainly be leveraging the power of Grok to create hyper-relevant timelines and help people understand everything that's happening. Thanks @elonmusk and the 𝕏 team for inviting me to work on the product that I live and breathe.