Product builder. https://teapawt.net/t.co/ozk8wAI8tb - AI Job Agent https://teapawt.net/t.co/c6oirM2XLQ - Resumes & Job Search This moment is the only truth.
India
Joined August 2009
- Tweets880
- Following48
- Followers209
- Likes3.9K
Pinned Tweet
You'll meet very few people who don't try to judge your status by your job title, possessions, pedigree or anything else.
Treasure them. They're probably free from the world's noise.
The Finance Ministry @FinMinIndia and @RBI need to better understand the difference between productive and unproductive work.
Instead of reducing paperwork and unnecessary compliance, they keep adding more requirements for everyone, often without any sensible thresholds.
Another example - under FEMA ODI regulations, RBI has apparently directed AD Banks that APRs (Annual Performance Reports) need audited financial statements of the overseas entity, even when the host country itself does not require them under a certain threshold.
If you can't support startups, at least don't hinder them.
cc @sanjeevsanyal
🚨 New RBI rule for anyone invoicing clients outside India, effective 1 October 2026.
Service and software exporters (freelancers, IT firms, consultants, tutors) now need to file an Export Declaration Form (EDF).
#FEMA #RBI #ExportCompliance #Freelancers #IndianStartups
Opus 5.5 is really good. It's like Fable but token efficient and fast.
Feels like we have crossed the threshold where minimal human intervention is required for routine software development/production monitoring work.
Running an AI production monitoring agent with access to logs, server metrics, exception tracker and analytics along with the code base + deploys is a huge unlock.
Highly recommended.
This is a great conversation.
Balanced and thoughtful takes from @illscience on where we are right now, and the consumer AI opportunity.
(The first couple of minutes has the selected highlights).
lovely + dense convo w @lennysan re: "what if it works":
- decentralization + everything is working: this product cycle has more winners and more ways to win, yet many in SV are oddly nursing a dark fantasy about being outside the lightcone
- climbing the ambition ladder: shipping your bad ideas to discover your good ideas + hyper-local expressions of ambition like making the NHS as good as an iPhone
- changing the narrative around machine intelligence: making important things like education and healthcare cheap, unbundling learning from institutions and status from credentials
- what's working in consumer: coding agents, personal agents like Town / Instinct / Grok Bots / ChatGPT Work, and entertainment across creativity / gaming / companionship
- coding as an activity vs an outcome: perhaps coding is more like reading than writing; a fulfilling way to learn / explore / spend time more than a goals based task
loved the conversation + love to hear what you think!
Good to see all the innovation happening at @iitmadras. Thanks for covering it @Astro_Radifah.
indiatoday.in/science/story/…
Here is an example of wasted tokens. It's Opus 5 Max btw.
The efficiency will get better though, these are still the early days.
> "isn’t it kinda crazy that nobody is really using AI Agents?"
> "especially since the models and harnesses are here and ready"
No, they are not ready yet.
We are not at a stage where a regular person can simply talk to them to do a real life task, and it just gets done without having to provide a whole lot of context & information first and then not having to review the final result.
It's getting there, but till they get access to all the scattered information in our lives (a lot of which is in our heads) and can use pre-built systems safely and reliably, there is still a lot to do.
e.g. Am taking X to a swimming lesson, by the time am back can you order these N things from here. Finalize and file my tax returns. Coordinate with Y & Z and make a dinner reservation at that new place in downtown. Btw you spoke with a hiring agent last week, is that company keen to talk to me for that new role?
(And this needs to be a quick voice conversation in any language with back & forth).
Hot take… isn’t it kinda crazy that nobody is really using AI Agents? I don’t mean software engineers or AI early adopters. I mean “college friends talking about it in group chat,” the feeling you got when everyone started using Instagram or TikTok.
These frontier AI models are *insane* (as are the harnesses & tool calls & the like). And every large tech co has an AI agents platform, not to mention all the YC startups doing vertical agents. Yet all of your friends and family outside of tech — who spend all day staring at their iPhones and get paid to work in browser tabs — don’t really care or find themselves using any AI agents yet.
Yes ChatGPT, Claude, etc. are extremely popular… but if you look at the engagement data the vast majority of people are still using these aI chat tools like a glorified Google + Grammarly. That’s why the AGI labs are all pushing desktop apps for Codex, Cowork, etc. so hard to non-technical ppl. And yes exceptions for lawyers and customer service but even those have some asterisks and exceptions to rule.
Look I’m not saying the ChatGPT moment for AI Agents is not coming… it most definitely is! Remember we pivoted from Arc to Dia precisely because we believe computing is going to be radically reimagined around these AI primitives. No doubt. But that’s my point: it’s just so surprising it hasn’t happened yet because all of the tech you’d need is there.
Again if you stop for a second and think about it… for all the press and money and hype and models and crazy ARR numbers… this “AI Agent” moment does not *feel* like the other breakthrough tech moments we’ve lived through (e.g. think the shift to Stories via Snapchat & Instagram, or shift to on-demand via Uber/Airbnb/Doordash).
Which is a long way of saying: if you can figure out the answer to “why” most people don’t care about AI agents yet (and have no enduring interest in using them) — especially since the models and harnesses are here and ready — the answer to that question will allow you to capture a lot of marketshare and make a lot of money in 2027.
Theoretically, the tech is ready for AI Agents to totally transform how we work and live our lives… but alas the general public dgaf… that’s the generational puzzle to solve for the next 12 months for anyone not working on the models themselves.
I think all the following things are happening together in this AI cycle right now.
1. Insane excitement in the tech community looking at things AI can do N times faster than they could ever do. (Mind you, it still can't do all of that independently without it being steered by humans).
2. Executives/management racing to showcase "AI transformation" initiatives to keep stock market/their bosses happy.
Building solutions/systems; and *solving actual problems that impact top or bottom line* are not the same thing, and unfortunately many people miss that point (including technical folks).
3. There is a doomer set of people that think AI will become the superlord. In their opinion, humans will have nothing left to do at all, and the world will end. It's a conviction they probably hold anyway and now it feels inevitable to them.
4. Reallocation of capital in massive numbers is having its downstream effects. It is pricing in a future that will take years to materialize.
What can an average individual/company do?
I guess be more productive using these tools and/or build new things, else the ground will shift faster than anticipated for most. This also includes folks sitting on capital in equity markets thinking they got it covered, but perhaps they aren't safe either.
Some nice ideas here, especially if you're building solo and want to use AI agents to speed up your workflow.
Not sure if any of this is a good idea or not, but here's my "stack" for AI-assisted SaaS building:
1/ I use both the Claude Code and Codex native apps. I like both for slightly different reasons. In particular, Codex's preview browser with annotation markup is really helpful.
2/ also have CC and Codex installed on a remote VPS with a hand-off skill to hand off tasks from desktop app to VPS as needed.
3/ I use Fable primarily to orchestrate larger feature builds. It's set up to delegate coding tasks to Codex or Opus/Sonnet agents as needed and then reviews and iterates. Really big features start with it building a clickable mockup for me to approve.
4/ smaller, more direct tasks, bug fixes, etc., usually directly in Codex using 5.6 Sol.
5/ PRs get batched into a review queue, and then Fable reviews in a batch, runs tests, CI, and deploys to a staging environment.
6/ full manual browser QA in staging before release. I'll often have the agent compile a PDF of screenshots of every step of the QA so I can check for any weird UI bugs.
7/ Once it's good on staging, promote staging to main. After every deploy, Fable is instructed to keep an eye out post-deploy for any immediate bugs, errors, or issues.
8/ I also have Hermes installed on a VPS and a post-deploy hook where Hermes does a non-destructive browser QA of the production app to look for any immediate issues.
9/ a scheduled Codex task checks Sentry multiple times per day for any new errors and has a triaging workflow to either immediately create PRs and add them to a monitoring queue. In the case of extremely urgent, problematic issues, it can push a PR hotfix to main and notify me immediately in Slack. This Codex session has its own internal dashboard, which is an HTML artifact that it keeps up to date with:
recurring issues, what can be learned to improve testing , what it pushed, etc
10/ For example, it appropriately nudged me to upgrade our Hetzner server, which it pretty much handled for me. It has a recurring task to check the overall health of our servers and databases, ensure backups are running properly, do dry runs of the backup restore, make any recommendations based on usage and capacity
11/ our team has a Slack channel for reporting bugs and feature requests. Codex skill checks that Slack channel for me periodically and proactively investigates and creates PRs for me to review. It's scheduled to close the loop and reply back in Slack when things have been resolved or features have been added.
Lack of attention to detail and the rush to ship are the main reasons so much software feels broken today.
Software wasn't perfect before either. I think we are just using a lot more tools now.
Gave Claude Code (Opus 4.8 Ultracode) a detailed rebuild plan and locked mockups.
After a context reset (compaction?), it ignored the plan entirely and built a polished MVP that looked nothing like the spec.
It had resumed from the session summary, not the actual plan.
@bcherny @ClaudeDevs
Aditya Babbar retweeted
Looking for a team of 10-12 cracked researchers & engineers to build a new AI Lab in India.
Funding and compute secured.
DMs open.