Pinned Tweet
Here is a collection of threads that I have posted on Twitter. I will be using it as a central repository for quick access and keep on updating it.
Prashant Singh retweeted
we've left the agents unsupervised for long enough.
i've watched @bettersayAJ and team spend the last few months digging into the weird, messy failures that only show up when real customers start using agents.
they've taken everything they learned and turned it into @neatlogs
it's live today. so so proud of this team!
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Prashant Singh retweeted
TIL# 4
A careful agent should read the complete response, resolve contradictions. But thats not the case.I realize that most coding agents generate response serially.
With in a same message, it will pass
- you a wrong command
- identify it
- and correct it
so all three will be available in same message (like the screenshot). Reply to your prompts are essentially streaming and agent cannot retract those displayed tokens.
If another automation or agent treats every code block as executable instructions, it could run an earlier command before reaching the correction, especially if it processes streamed output incrementally.
Prashant Singh retweeted
The product challenges for AI that records us (and for us)
reasoned.live/p/the-product-…
Prashant Singh retweeted
TIL #2
Design APIs Around Their Abstractions, Not Today's Use Case. What is true for today's consumer and what the API contract is intended to represent.
When evaluating a proposed API change, consider the following questions:
1. What is the natural ownership of the data?
Don't ask, "Where is it easiest to put this field?" Instead ask: Is this an input required to perform a specific operation?
The location of a field should reflect its conceptual ownership, not just its current usage.
2. Design for the abstraction, not the current implementation
Consumers often have assumptions,those assumptions may be valid for the current workflow, but a public API often serves multiple consumers over a much longer lifetime.
A good API models the underlying business capability rather than optimizing exclusively for one client's current implementation.
3. Be careful when introducing exceptions to established patterns
If an API has an established pattern, introducing an exception should be an intentional architectural decision. The question is not whether the exception is wrong, but whether the team wants to establish a new pattern for future development.
4. Evaluate trade-offs rather than arguing absolutes
Many architectural decisions have more than one reasonable solution.The goal should be to understand which trade-off best fits the system being designed.
5.Raise architectural concerns early
The best time to discuss API contracts and architectural principles is during design, not implementation.having alignment early reduces churn and results in better designs.
Good architecture is rarely about winning an argument over a specific field or implementation detail. It's about ensuring that individual design decisions reinforce a coherent set of principles that will continue to make sense as the system evolves and new consumers adopt the API.
Prashant Singh retweeted
Had completely forgotten: I started reporting on the tech industry in India 20 years ago... on August 1st.
Was finding it difficult to get a job and asked friends for help because I only wanted to work in the Internet industry, and there weren't many opportunities around and I didn't know how to get them. I loved the Internet then, almost as much as I do now.
Had interviewed with network18, based on @/codelust 's recommendation, and had been waiting for a month for them to respond. @mrajshekhar suggested I apply for a gig with his friend @rafat who used to run contentsutra.com . Rafat said let's try this out for 15 days. 10 days later, I wanted to do this, and he wanted me to continue. Network18 came back a month later and I said no, because I was having fun.
6 months later, I was running contentsutra, having the time of my live, reporting on a domain and an Industry I love, with the freedom to say whatever I want to say, and meeting fun people, going for barcamps and organising mobile mondays with @pacificleo.
20 years later, still staying true to that: saying what I want to say, doing what I want to do, and what I think is right for the Internet and the Internet in India. The net neutrality campaign happened because it was the right thing to do. The pushback against aadhaar was the same. The pushback against the IT Rules is based on the same principles.
My love for the Internet is at slight loggerheads with my love for AI, but that conflict doesn't stop me from spending 10-14 hours a day (including today) coding and building. Was at a Qwen meetup and an 18 year old reminded me I started working before he was born. I loved it.
I was bored out of my skin (and really unwell with surgeries etc) last year, and policy felt repetitive, stupid and boring, and I turned to AI and reasoned.live to push me to do something new. 8 months later, I'm doing hackathons, building products, building internal tooling, experimenting with hardware, and having more fun that I've had in the last 5 years. Am a kid in a candy store with AI.
All this while, I'm thinking about how AI will change the Internet, and with it, the world.
Thinking, writing, debating, fighting, tinkering, building.
20 years...most days I still love doing this.
Prashant Singh retweeted
I’m seeing people from the tech/startup/design circles being quiet about the protests and continuing to post about their work, etc. Being apolitical is a choice. And it’ll be remembered.
Take a stand — for or against. Get blocked and unfollowed. ATP, being apolitical is worse.
Prashant Singh retweeted
This order is illegal and unlawful and there’s no law as far as I know that allows the Government of India to do this. They’re acting as a law unto themselves.
Let me explain:
1. Bitchat itself has not violated any law. It’s a piece of software that people use to communicate without the Internet. If a threat was created using Microsoft Word OFFLINE, would you hold Microsoft responsible for it. If that is the principled rationale here, then every software in the world is susceptible to the vagaries of actions from the Indian government.
2. Such an act of disproportionate blocking needs to be backed up with proof: merely being able to communicate without surveillance, is not a violation of any law. Neither the platform nor the government of India can track what is being done with it. On what basis is this order being issued?
4. Decentralised technologies are critical to the functioning of the web: they allow users to take control over their own Internet usage, without dependence on a single platform. Maybe that’s exactly it: the Indian government controls our speech and our communications THROUGH concentrated large platforms like x, Instagram, Facebook, and the idea of decentralisation, where they can’t order a single player to take something down, seems unmanageable to them.
4. An act of blocking has to be proportionate: there is legitimate speech potentially on the platform itself. We don’t know what is the actual speech which it thinks (doesn’t know) is illegal. By taking down these platforms you censor/prevent people from having the opportunity to communicate privately, and legitimately.
5. Directing Github to remove this app is also deeply problematic: not only is Bitchat not illegal, a platform that allows you to download an app or code for your own personal usage, for legal purposes, is also not illegal. Github is at risk because it’s a centralised platform btw.
6. No law in India mandates an app to have mandatory user registration, phone number verification or centralised logging of communications. There is no ban on anonymous coordination. Surveillance isn’t mandatory. In face, many existing communications apps do not do this. THIS as a basis for issuing a blocking order is itself illegal - no law supports this action.
7. The order says that the platform is capable of being used for unlawful activities: I’m afraid that’s no reason for blocking something. A messaging app can be used for illegal activities, it can be used for legal activities. A knife can be used to till or cut fruit. You don’t ban knifes.
Again, this order is illegal and unlawful and there’s no law as far as I know that allows the government of India to do this. They’re acting as a law unto themselves.
Prashant Singh retweeted
Stratechery's @benthompson
"I got a fun email from former Apple executive, Nest founder, and one-time Stratechery Interview subject Tony Fadell in response to yesterday’s Update about Apple suing OpenAI (published with permission)"
Honoured to be recognised as Fintech Startup of the Year by @BwDisrupt. 🏆
Here's to everyone who's making #FinancialFitness a movement!
Prashant Singh retweeted
Crossed 20 signups! This is fun. Thanks for organizing this @GrowthX_Club @udayan_w @NousResearch
Folks: sign up at pragent.app/waitlist . Launching today
Prashant Singh retweeted
At the highest levels of the craft, analogies are a great explanation tool and a poor decision-making tool. Elegant for explaining the output of your thinking, treacherous as input to your thinking.
Analogies are the map you draw after the journey, not one by which you navigate.
Prashant Singh retweeted
Re-introducing myself for everyone here :)
Hi, Radhika Dani here 👋🏽
I’m an interdisciplinary product designer and former founder, interested in how people understand, decide, and behave - through design, psychology, and increasingly, AI.
I’m pursuing that curiosity more formally through a master’s in psychology.
Something you can expect: context, ideas and thoughts on the UX of experiences, data and AI: how information is framed, how bias shapes what we see, and how technology can amplify - or intentionally counter - it.
Alongside that: mini experiments, prototypes, and things I build while exploring tools like Claude and Codex.
Also: an amateur hiker and a foodie with a curse of lactose intolerance.
❤️✨
Prashant Singh retweeted
we're building a small research team at @neatlogs around self-healing environments for agents. we've done some early work on identifying critical events in a trace and now want to invest in the research in parallel.
DM me if that sounds interesting. if you're a @lossfunk alum, we skip the screening process and you go straight to an interview

























