We thought serving an open model was the easy part. Start vLLM, done.
It wasn't. We got 64% more throughput, one test at a time. Then we realized we'd been running the same loop by hand every time. So we built a harness just like Karpathy's autoresearch.
blog.vertexcover.io/vllm-tun…
1
1
46
Normally the company writes the case study. This one’s written by the customer.
Naman on what happened when Refrens rebuilt how their engineering team works around AI agents. Four months of numbers. We built it with them.
43% more work shipped per sprint. 16% faster turnaround. Juniors taking on complex work.
I’m convinced this is the future. Our job as engineers is less about writing code and more about building the systems that write it.
We tried to make our engineering team 3× more productive with AI Harness. Was the effort and cost worth it?
Here’s the full story, with numbers.
1
36
Ritesh Kadmawala retweeted
We tried to make our engineering team 3× more productive with AI Harness. Was the effort and cost worth it?
Here’s the full story, with numbers.
Funny how AI and agents were meant to make us work less.
Every engineer I know is working harder than ever. And every other post says software is dead.
If it's dead, why is everyone this busy?
56
Anthropic studied 400K Claude Code sessions. Headline: non-engineers succeed at nearly the same rate as engineers.
Everyone read it as "you don't need engineers anymore."
Same week, Meta shipped AI-written, AI-reviewed code with no human in the loop → auth bug straight to prod
1
116
3/ That system has a ceiling. Tests catch a lot. They don't catch whether an auth flow is actually safe.
The real skill now: knowing where your review loop ends and human judgment has to take over.
Where do you draw that line?
1
19
AgentLoop: a daily digest on AI agents & harness engineering. Today's full issue, no spin 👇
agentloop.vertexcover.io/arc…
Subscribe free: agentloop.vertexcover.io/#su…
18
Ritesh Kadmawala retweeted
I’m thrilled to announce we’ve raised $44M to build a new home for product design. Meet @noondesign.
No workflow is more broken and fragmented in 2026 than the product designers’. The very same people who care most about building software don’t have software purpose built for them. @kushagrasinha7 and I have lived this problem first hand as designers ourselves.
That’s why we built Noon. The first product design tool that works entirely on your product code, so you can design not only how a product looks, but also how it works. With AI at its core that works in seconds, not minutes.
For the first time, you can create, iterate, build, test and ship. All in one canvas. No translations or roundtrips to the codebase and back.
Comment “Get Noon” and we’ll get you on the list for early access.
Ritesh Kadmawala retweeted
I built a $10 workflow that turned static winners into animated ads with 30% higher performance than a $10k video shoot.
Here is how:
From 70 mins → 8 mins in GitHub Actions. No Depot, no WarpBuild—just smarter Docker & BuildKit hacks. 🚀 blog.vertexcover.io/speed-up…
1
99
Strot - An AI Agent that scrapes your web api's
github.com/vertexcover-io/st…
blog.vertexcover.io/strot-is…
65
Ritesh Kadmawala retweeted
Just joined the AI Residency @lossfunk and I think @paraschopra has gotten some of India's most cracked people in one space! Ex-
- Guy who hacked his Alto to self drive 80Kms just through a Redmi Note 9
- 19 year old kid who's made 30K USD through NFT when he was 16
- An AI Engg who taught himself law and fought his own case
Ritesh Kadmawala retweeted
Your evals suite should make you feel embarrassed, not proud of yourself.
If you're scoring +90%, it does not mean that your AI/Agent rocks.
Your dataset sucks.
Ritesh Kadmawala retweeted
How do you fit a 250kB dictionary in 64kB of RAM and do lookups? For reference, even gzip -9 cannot compress this file beyond 85kB.
In the 1970s, Douglas McIlroy at AT&T had the same challenge when implementing the spell checker for Unix.
Instead of relying on generic compression techniques, he analyzed the distribution of the data and came up with a compression algorithm which was just 0.03 bits away from the theoretical limit of possible compression. To this day, it remains unbeaten.
The story of how he designed this dictionary lookup from the ground up is a lesson in software engineering. How you consider the resource constraints in front of you and design your solution to run within them.
In my latest article I take you through each step of the design of this ingenious algorithm right from the scratch.
(Check replies for the link)
Ritesh Kadmawala retweeted
Painting the city purple!
@RefrensApp
RT @DaveOCKOP: Wow, this is one of the most unique moments in football. Jurgen Klopp just started a cheer for ARNE SLOT, LA, LA, LA, LA,LA.…
Ritesh Kadmawala retweeted
❤️ #DankeJürgen ❤️

















