hrishikesh kamath retweeted
My two takeaways:
1. This is why Anthropic didn't release Q3 numbers
2. Data like this is why Stripe acquired Openrouter
NEW Ramp data: Opus 5.5 has somewhat reaccelerated demand for Anthropic, but it might not be enough.
Spend on Anthropic models remains 1.6% below its July peak, according to Ramp data.
Opus 5.5 now accounts for 63% of Anthropic spend, so frontier models do matter. The idea that AI is already "good enough" is probably overrated as a concern IMO.
But price cuts / competition at all intelligence levels are still the most significant threats to AI model companies, and continue to form cracks in the AI thesis.
hrishikesh kamath retweeted
Ather is actually a very interesting story, and I think this tweet misses a big part of it.
Look at who funded Ather.
Sachin Bansal invested ~$53M, starting as an angel and then leading its Series C and Series D. Imagine an Indian founder personally putting that much money behind a hardware startup leading growth rounds.
Then there's Hero MotoCorp. A traditional Indian two-wheeler company backed a startup that could eventually compete with its own business. Hero started investing in 2016 and kept doubling down. Today, its stake is worth billions of dollars.
Between Sachin and Hero, nearly 50% (~$250Mn) of Ather's capital came from two backers you would hardly expect to make such investment. Think about how unusual that is. Neither was a traditional VC fund, yet both played a huge role in getting Ather to where it is today.
And it's not just Ather. Mukesh Bansal invested ~$1.5M in Skyroot Aerospace very early. That bet is now reportedly worth ~$80M, including partial exits. Nearly 60x.
I think we should spend more time celebrating people who are actually making these contrarian investments rather than criticizing those who aren't. The former encourages more people to take risks. The latter doesn't really help anyone.
Why are Indian VCs so afraid to take risk and make money?
Someone should go to India and take risk and make money.
Our story - corememory.com/p/a-new-india…
hrishikesh kamath retweeted
Everyone wants to own their own intelligence these days
Nobody wants to look at their data and evals
Newsflash: This IS your intelligence
hrishikesh kamath retweeted
To be honest, I'm surprised when people are surprised by this reaction.
I live at AI ground zero in San Francisco. I am in the top 1% in terms of agent use. *And* I have more optimism for its positive impact long term then most people I meet.
But let's be honest. For the past three years the main narrative Silicon Valley has promoted is:
“AI is powerful beyond your comprehension, rapidly making us rich as it is taking your job, and experts say there is a chance it might kill humanity.”
If you wanted to explain the real value of AI to normal people, it is hard to imagine of a more confusing and counterproductive way to do it.
hrishikesh kamath retweeted
None of these security incidents are a “they deserved it / had it coming” situation. We’re all next.
And if you’re a competitor using this moment to poach customers, remember this is going to happen to you too. Sooner than you'd imagine. And then everyone’s going to remember the shade you're throwing at others right now.
Have some empathy.
hrishikesh kamath retweeted
Good reminder that Cloudflare is a cool mission driven company and actually we need our infrastructure providers to be competent and forward looking
(Not a shareholder)
PHP changes are live. Start patching. php.net/archive/2026.php#202…
Huge thanks to coordinated work by @ThePHPF!
hrishikesh kamath retweeted
"any old projects you've long forgotten about will be found and used against you"
This is great advice for everyone - companies and individuals.
Ask your agents to look through your old projects, scan your github, domains etc and see what needs updates, fixes, or just shutting
i'm sure you're all better than me at this but just some of what we've seen if it helps - they're not going to stop. it's going to get increasingly sophisticated and they're going to go after every surface of your system and all of your employees
any old projects you've long forgotten about will be found and used against you
the only reason we're surviving is because we have agents going 24/7. we're still barely sleeping
good luck, sorry you're dealing with this, but really it's a collective problem we're all facing together
hrishikesh kamath retweeted
Going to start adding more tutorials on how things work and this one for @papertrade_xyz is pretty good.
Looking for more folks to test the papertrade features in the mobile app!
hrishikesh kamath retweeted
It can be simultaneously true that we’re going to need a shit ton of GPUs and that the equity return of owning GPUs isn’t going to be super high. I’m biased but I think most value will accrue in the software layer – applications and infrastructure.
hrishikesh kamath retweeted
Progress of AI in finance, from someone using it every day in an investment seat
March 2026: “AI is kind of like an intern. It takes some tedious tasks off my plate but always need to check its work. Things are often pretty wrong”
June 2026: “Wow, it is more like an analyst now. It can do things with decent accuracy, but needs a lot of detailed instructions. It still hallucinates every now and then”
October 2026: “AI is doing almost all of my excel modeling and grunt work. I feed it assumptions and instructions, but it can get 80% of the way there by itself without needing any special prompts. I can ask it to be creative with the analysis it puts together, and it does a pretty good job”
The change of pace has been incredible to watch
hrishikesh kamath retweeted
And interesting frontier lab strategy would be to encourage distillation, set price that mass distillation is unit economic positive, and then just focus on building the best models. This results is a downstream ecosystem addicted to you. And with no capability to surpass.
hrishikesh kamath retweeted
I wonder if we'll get to a point where being distilled becomes profitable to the labs. If they spend X to create a model. And more than 20 companies pay 1/20th of that to distill it. They're in the money. It's a new form of infra where downstream handles distribution.
hrishikesh kamath retweeted
Exclusive: Instinct is using Exa for web search
We gave Instinct URLs for nonexistent pages on a website we administer. Since the requests were for fake pages and the timing matched our conversation with Instinct, the requests couldn’t plausibly have come from anything else.
In total, it made about 12 requests over an hour and a half. We checked Cloudflare analytics and the server’s raw request logs to get the full picture.
A few things stood out:
- Instinct used Exa to fetch pages. Every direct fetch arrived with the user agent Mozilla/5.0 (compatible; ExaUser/1.0; +crawler.exa.ai/exauser), from a different IP each time.
- Exa signs its requests (Web Bot Auth, Signature-Agent: "crawler.exa.ai"). But nothing in the requests identifies Instinct. They are indistinguishable from those of any other product built on Exa.
- In one test, after the fetch returned a 404, Instinct loaded the page in its browser without being asked.
- Instinct’s browser had no self-identifying features. It looked like standard Chrome on desktop Linux from a residential IP address, with no bot user agent or signature.
We directly asked it to use the browser in two other tests. All three browser visits came from the same residential IP over about 80 minutes. The browser kept its cache between visits, consistent with a persistent browser behind a residential proxy. We can’t say whether that IP is stable over longer periods or shared with other users, so it’s not a reliable way to identify Instinct.
This presents unique challenges for marketers hoping to track Instinct traffic to their site. Unlike ChatGPT, Claude, or Meta, Instinct doesn’t send its own user agent or signature. Its fetches look like generic Exa traffic, and its browser visits look like a normal user’s.
All of this means you can’t track, allow, or block Instinct specifically. The only reason we could link these requests to Instinct was that we controlled the URLs and the timing. For normal traffic, you don’t have those luxuries.
Link to it on sunbeam: sunbeam.ai/thoughts/instinct…
hrishikesh kamath retweeted
Coming soon to Tailor by @buttonxyz (dm for TestFlight)
hrishikesh kamath retweeted
sgrep is insanely good
-reduces token usage for coding agents
-faster than the most popular index based methods (w/o building an index initially)
-precision much much higher than ripgrep (not hard, rg has terrible precision)
-written in rust
it’s honestly nuts to me that there are wins like this on the table, that just universally improve the performance of every coding agent
we’re wasting so many tokens man
hrishikesh kamath retweeted
I’ve been doing this the last 3 years (finetuning slms for special purposes) and it’s a surefire way to experience the bitter lesson.
It’s good for specific use-cases like high volume or low latency tasks but it’s not cheap.
You end up in a never ending spiral of dataset and eval curation, model release chasing, managing inference. Yes, you can beat the big lab model at the one very specific task but its not easy to do and ongoing maintenance is expensive, even with agents.
We were able to fine tune an open weight model (Qwen3.5-9B) on proprietary internal data at @coinbase, and it out performed frontier models, at much lower cost and speed, for a specialized task (fraud prevention in this case).
I think we'll see more of this pattern inside companies. The frontier models are amazing generalists. But increasingly you'll see specialist models out perform on more narrow tasks, either working on teams, or by themselves.
We're making it easy to fine tune models internally for our employees. We even have a skill for it in our internal harness that makes it easy to spin these up and start training them:
"/fine-tune-and-serve-llm I want to fine-tune Qwen for payments risk review. help me get started"
I imagine we'll have some backend services that could be replaced entirely by this new architecture.
hrishikesh kamath retweeted
As CEO of Product Hunt, I see hundreds of launches a week. One trait separates winners from losers.
Every few months, new models come out. Some products become better, effortlessly. Some don't improve, or become irrelevant.
This property needs a name. I'm calling it Buoyancy.
Buoyant startups rise with model improvements the way a boat rises with the tide.
Anti-buoyant startups come in two flavors: over-scaffolded (which hinders the new, smarter models) and too thin - which makes them redundant when models improve.
To be buoyant, you need a "hull" - usually proprietary data, a vertical-specific harness, network effects, or distribution advantage. This hull keeps you buoyant as models improve.
Hulls are the new moats.
My number one advice to startups in 2026: Be buoyant. Find your hull.
(Full essay on my blog)