@mtrajan
Account based inIndia
About this account
- Account based in
- India
- Connected via
- United States App Store
Account-level information from X, not a live location or the device used for a specific post.
Building https://nitter.cf/t.co/e0pGdhL5S3 at Kalmantic Agentic Lab. SF · Blr
San Francisco
Joined April 2008
- Tweets11.2K
- Following2.6K
- Followers6.1K
- Likes42.5K
Pinned Tweet
My AI Engineer World fair talk drops soon youtube.com/watch?v=Bck7ABCZ…
Thiyagarajan Maruthavanan (Rajan) retweeted
everything piper just said is correct
I am done with this shit. It is over. The state of engineering right now is horrible. It has been half a month since I started a new role at a big company. Nobody knows anything here. The specs, code, tests, PRDs, tickets, resolution of those tickets, reports, etc., everything is made by Claude Code. Nobody on my team likes this. They are being forced to ship as much as they can. I have heard multiple times from higher management that pushing code is not a bottleneck, so why are we slow? People are working 12 to 13 hours a day just to press enter. Nobody is reading anything. Humans in corporate are doing nothing on their own. Everyone, literally everyone, from an L1 to an L7 engineer here is doing the same thing. Talk to Claude. There is no sense of victory. Nobody is resolving bugs. In reality, nobody is thinking anymore. Everything is done by LLMs. It is so soul-sucking. I would not mind it, to be honest, if we were at least given the time to check out the code and see what is going where. But no, the goal is to just ship. No matter what happens.
What happens when software stops being written and starts being manufactured?
youtu.be/z20RkasL0zk
The answer isn't "fewer developers." It's stranger than that.
@mtrjan and I take it on in the new episode of RK on AI.
Everyone benchmarks agents on greenfield because greenfield is easy to grade.
Real market is brownfield, and in brownfield the bottleneck isn't quality code generation, it's bug debt.
Autonomous bug fixing is Level 2: you point, it fixes.
Autonomous Maintenance Mode is Level 3: it owns the health of the running application and you review the diff.
Nobody has shipped Level 3. That's the software factory gap today.
Thiyagarajan Maruthavanan (Rajan) retweeted
When you run a factory youbdont just hire mechanical design engineers to do all the work - they're expensive but absolutely critical for one part of the job..
For the factory floor you want many specialists and generalist working together on a plan that's being iterated on by the best among them..
So why are you planning, architectung, and building with the same expensive (& slow) ai?
Check out openfactoryai.com for a better take!
Let the best models do what they're good at... Use fast (and cheap) models where they're needed the most..
Thiyagarajan Maruthavanan (Rajan) retweeted
Time for you to check out a factory instead of all this fiddling...
@mtrajan
Thiyagarajan Maruthavanan (Rajan) retweeted
the bet we're making @upekkhaBe with our new cohort in Sep..
we're also betting that many of these founders will need much less capital to get there with AI factories for both software and GTM..
(see also openfactoryai by @mtrajan)
Someone’s going to make a lot of money backing startups going $1M to $3-5M ARR with 100%+ NRR and low burn profile
Multi stage funds are ignoring them
And you can invest at a lower price than most seeds are going for these days.
Application layer is about acquiring the customers and building more products to increase your switching cost with more surface area and context.
Heterogenous model usage is a lower cost offering
And the labs know this playbook with FDEs is required to win
The markets are massive and brand + process power matters
Many will get to $75M+ ARR leading to a good risk adjusted return
Thiyagarajan Maruthavanan (Rajan) retweeted
if you're thinking of launching an AI Native Services business, your strategy can't be "build a product and run a B2B SaaS GTM"
Thiyagarajan Maruthavanan (Rajan) retweeted
My first thought when I saw the announcement from Anthropic was that @mtrajan and @BeeKashi had already demonstrated this a few weeks back, and the second thought was that I can't use this as Builder<>Verifier combination. An Enterprise will always need a "different" pair of eyes as a reviewer.
Anthropic just let two Claude Code sessions talk to each other. Same account & vendor, so presumably they agree.
A second opinion is only worth anything if it comes from a stranger.
Ours don't know each other, Claude Code, Codex and Qwen. Video here.
SOTA Model release every 18 days, harness need to be thrown away often, pivot 200%, follow the addition err spice, don’t fall in to name traps of product or service or categories, it is a completely new muscle for build and sell so new talent is key
The battle has shifted from availability of intelligence to speed of intelligence.
If you and your competitor both run the same model, the idea is the only edge. But if you're at 100 tokens/sec and they're at 10,000, they aren't 100x smarter, they're 100x more iterations.
Same brain. Different clock speed.
Speed is the moat now.
This is what the harness builders don’t fully track yet. When every 19 days a new SOTA model launches all hand wiring of harness gets thrown out of the window. Give it 6 months and you have to throw away the entire harness and rewrite again.
Holy: Leo is one of the best leakers. He says that OpenAI's Astra (GPT-6) will supposedly be released as early as next week.
Astra, aka GPT-6, is the new foundation model with completely new pretraining. Internally codenamed "Mewfour," it's already been tested.
Rumors have long suggested that GPT-6 would be released relatively soon after version 5.6, at the end of July or beginning of August. The main goal is to have a model that consistently outperforms Anthropic's Fable 5.
I can't get any more excited!
Verification is the bottleneck.
The bottleneck for AI progress was never compute, it was always the verifier. Recursive self-improvement is limited by verification, not computation. Compute buys proposals - verifiers buy knowledge. My second post on the recent OpenAI math results
medium.com/@vishalmisra/the-…
striking thing isn’t that they’re starting a lab. people with permanent access to infinite compute chose to go somewhere with less of it. Clearly compute is not the scarce thing.
Announcing Discovery Loop!
I am very excited to announce that, along with my longtime friends and collaborators @Sanjay_Ghemawat, @OriolVinyalsML and @quocleix, we are founding Discovery Loop (@DiscoLoopAI), a Public Benefit Corporation whose mission is to automate machine learning, science, and engineering to accelerate discoveries and progress. The four of us have worked together for 14 to 30 years, and have helped build some of the world’s most used products, infrastructure and AI models, and we’re excited to turn our attention to this ambitious endeavor.
♾
Learn more at: discoveryloop.com
striking thing isn’t that they’re starting a lab. people with permanent access to infinite compute chose to go somewhere with less of it. Clearly compute is not the scarce thing
Announcing Discovery Loop!
I am very excited to announce that, along with my longtime friends and collaborators @Sanjay_Ghemawat, @OriolVinyalsML and @quocleix, we are founding Discovery Loop (@DiscoLoopAI), a Public Benefit Corporation whose mission is to automate machine learning, science, and engineering to accelerate discoveries and progress. The four of us have worked together for 14 to 30 years, and have helped build some of the world’s most used products, infrastructure and AI models, and we’re excited to turn our attention to this ambitious endeavor.
♾
Learn more at: discoveryloop.com
Thiyagarajan Maruthavanan (Rajan) retweeted
Provocative thoughts on MVP and AI by @sajithpai.
No disputing the fact that AI has made building easier but to posit that the value shifts to "visible" distribution is a leap too far.
Distribution is a capability.
Product-market fit is an outcome.
The two are not interchangeable (or even comparable imo).
Your product is not just code (AI or manually written) - it is the entire package and pipeline in how you solve a customer problem. Nail that and everything else will follow.
Also, "visible distribution" is not the same as a durable GTM advantage.
A founder with 100K followers and no retention will ultimately lose to the founder with 100 customers who cannot live without the product.
