Anurag Nigam (@anuragnigam_) header
Anurag Nigam (@anuragnigam_) profile picture

Senior Software Engineer @spotdraftHQ, Building @zpply_in (https://teapawt.net/t.co/flxxJME2aj). Turning ideas into scalable product. Check this out: https://teapawt.net/t.co/hlCdfHsTbF

Bengaluru, Gorakhpur, Kota
Joined October 2015
Photo by Anurag Nigam (@anuragnigam_)Photo by Anurag Nigam (@anuragnigam_)Photo by Anurag Nigam (@anuragnigam_)Photo by Anurag Nigam (@anuragnigam_)Photo by Anurag Nigam (@anuragnigam_)Photo by Anurag Nigam (@anuragnigam_)Photo by Anurag Nigam (@anuragnigam_)Photo by Anurag Nigam (@anuragnigam_)Photo by Anurag Nigam (@anuragnigam_)Photo by Anurag Nigam (@anuragnigam_)
Now than Jensen is no more CEO of NVDIA and it is 30% low, thinking to add 5% of my investment in the stock - seems good opportunity to take this
48
When you and your coworker discuss PR solution ooppps your agent and your coworker agent #Claude
35
I've been using X for 11 years, and I'm only just discovering that you can take a screenshot of a tweet directly from the Copy Link option. 🤯
22
hey @NetMeds i ordered medicine and it was supposed to be arrive at 5 nov and it is 11 and counting? are you gonna deliver it?
2
1
87
The reason why i looked down upon people who prepare for these jobs or looked down on these bureaucrats. Seeing these people closely I understand India is done deal and no one can do anything
Let me take you back to the Lokpal and Lokayuktas Act, passed in 2013 and brought into force in 2014. Section 44 made it mandatory for every public servant to declare their assets and liabilities, along with those of their spouses and dependent children. Only the IAS lobby opposed this provision with all its strength. You can guess why. The DoPT extended the deadline for submitting property declarations five times. Every time, it declared that no further extension would be granted under any circumstances. Yet it kept extending the deadline. DoPT is also headed by an IAS. Eventually, in 2016, the IAS lobby got Parliament to amend Section 44 in just one minute, without any debate, removing the original mandatory disclosure requirements. The amendment left it to the executive to prescribe how public servants would declare their assets. And since then, the promised rules have never been framed to make that disclosure system operational. Now look at the Whistle Blowers Protection Act, passed in 2014. The government withheld its implementation, claiming that amendments were necessary. When the amendment bill lapsed, the government said no amendment was needed. Yet the law has never been brought into force. This is how transparency and anti-corruption laws are thrown into the dustbin. Even the weakest accountability measures are perceived as threats by the IAS lobby. The RTI Act has been butchered. IAS officers have captured the Central Information Commission. The entire system is designed to loot citizens and provide enormous protection to the corrupt and powerful. Ordinary citizens are left powerless, paying taxes like colonial lagaan, serving bureaucratic lords and living under their arbitrary authority. The IAS system has eaten away at this nation. Even in Pay Commissions, IAS officer occupy key position and secure extra edge in pay among the same upsc batch other cadres, arbitrary NFU to raise their pay, promotions and career progression. They manufacture higher posts, arbitrary hierarchies and privileges, extracting every possible benefit from taxpayers’ hard-earned money. And what do citizens get in return? These generalists occupy leadership positions across departments without specialised knowledge, original work or measurable performance. Wherever they go, they monopolise power, drain public resources and protect their own privileges. India gained political independence in 1947, but the colonial administrative structure survived. British rulers were replaced by Indian bureaucratic lords, while citizens remained powerless before the same feudal machinery. Whoever dismantles this IAS system and replaces it with democratic, professional and accountable public institutions will be a real patriot who gives India its second independence, freedom from the surviving relics of colonial rule. @narendramodi @RahulGandhi @khurpenchh @VivekGa54515036 #BureaucraticReforms #IASReforms #EndBureaucraticRaj #AdministrativeReforms #NFU #PayCommission #8thPayCommission #EndVIPCulture
1
1
1
9
313
Day 1 of cold call approach - Some HR called me to discuss an opportunity she had, after 15 mins of talk I turned the that discussion on user research interview for the agent i’m building. Reverse cold calling it seems🫣
50
my instinct sent me the regular news update which it send in the morning, is everything between meta and instinct sorted or should i ask it to use imessage for now?
60
I’ve been thinking for quite some time about sharing what i am building and how I’m building it, mostly with AI. But I’ve always been camera shy, and that is one of the reasons I never started. Now that I feel i’ve spent enough time in my comfort zone, I’m going to start sharing short vlogs on what I’m building, what I’m learning, what worked along the way, and what didn’t. Not sure how good I’ll be at this initially, but I guess the only way to get comfortable is to start. Let’s see how this goes!
1
3
76
"never say a number first" is good advice right up until the recruiter says "we can't move forward without your current salary" here's what actually works when you can't dodge it if they ask your expectation first: "what's the budgeted range for this role?" most recruiters will just tell you. to them it's not a secret, it's a filter if they insist on your current salary, give it. making it up is a bad idea when payslips can come up later but own the next sentence: "that's what my current role pays. for this scope, the market is X to Y" revealing your number isn't the mistake. letting them do math on it is "30% on current" is their frame "market for this level" is yours honest caveat: none of this works without leverage. one other process running, even an early round, changes how every one of these lines lands what's the worst salary question you've been asked on a screen?
1
1
55
I’m not sure what’s going on, but the past few weeks, especially the last two, have been literally chaotic. I’ve been seeing a lot of security attacks across companies. The way DevOps teams are handling these attacks is honestly incredible. I haven’t come across a single company that hasn’t faced some kind of attack in the past week.
37
I’m here at OpenAI DevDay Exchange. Come say hi
2
91
If i say I am disappointed in @ChatGPT and codex, that would be a understatement, I literally do not use it much now a days while i think i have been plus user from 2022-23. And now a days when i start using it they hit limit reached literally every 5 minutes like bro what do you want? like seriously every 5-7 minutes is joke. Why should i pay $20/month for something you cannot do effectively. For someone who says that this is how things work, i do pay claude $100/month and use it to the fullest
1
1
115
I don’t know where does this indiranagar guys and baddies go for chai, i went to one of the best rated chai place and it was such a mid. I can count 10s of chai places in HSR and they all are amazing
48
Last time I shared how I spent $1,500 on AI agents that built a product over 44 hours. Many of you asked where the money went. So I measured it from the session transcripts. The answer: context, not the model. Every step an agent takes re-sends everything it has read so far. One agent ran 952 steps and ended with almost a million tokens in context. Its late steps cost about 19x its early ones. That one run cost $148. What filled the context: - A 145k-token design doc, read 165 times. Each task needed one section. - Agents doing several jobs in one run. - Opus on 54% of code reviews. Only 7 of my last 60 PRs touched auth, payments or migrations. - 93% of steps made one tool call at a time. A review took 111 round trips. What I changed: - A script that returns one section of a spec, not the whole file. - One task per agent run, with a hard stop at 150k tokens. - A short tool list per agent. Starting context dropped from about 60k tokens to 20k. - Opus only when a PR's changed files are risky. Five days in, cost per step is down 21 to 46% across every agent type. Whole-doc reads went from 51 to zero. Code reviews aren't re-measured yet. Numbers in the table below. The bill tracks how much your agents re-read, not how much they write. What's eating your agent budget? #AgenticAI #AIEngineering
69
One shot build this video for @zpply_in from opus 5.5. reply to get the prompt
41
Job applications are a pain for everyone. Job seekers fill the same forms again and again. Recruiters get so many copy-paste applications they can't find the right people. We are building something to fix this, for both sides. Coming soon.
You saw the job at 11:48 PM. Everyone else saw it at 9:05. Something's coming for that gap. Guess what it does 👇 Waitlist link in bio. #jobhunting
53
I spent over $1,500 in Claude API usage on one experiment. I shared a rough product idea with my AI system and let it work for 44 hours. It came up the technical design decisions, HLD and LLD and by the end, I had a working product which required 160+ PRs. The agents built the frontend and backend, tested real user flows, and fixed bugs as they found them. This was not one agent doing everything. I set it up like an engineering team, with different agents planning, building, reviewing, and testing the work. A Principal Engineer agent made technical decisions. A Senior Engineer agent created the implementation plan. An SDE 2 agent divided the work and coordinated SDE 1 and testing agents. I stayed involved in key decisions. The agents saved their progress so they could pick up where they left off, avoid working on the same files, and fix failed tests. What I learned: getting AI to write code is the easy part. The hard part is keeping it on track for 44 hours and making sure the final code is worth shipping. I will share how I built it, what it cost, and what failed. What should I share first? #AgenticAI #AIEngineering #SoftwareEngineering
2
2
122