Joined July 2007
- Tweets 489
- Following 329
- Followers 149
- Likes 365
11 Photos and videos
May 13
Super proud to have been selected as the winner for the Vercel Day competition on @ProductHunt!! 🚀
May 12
we ran our first @Vercel Day competition and the results are in
winner getting a pitch to Vercel Ventures: @getqatech
runners up getting $30k in Vercel perks: @specstoryai @calendarpipe @tal_elor @helloaria_ai @arkyhq
congrats to everyone who launched 🎉
1
2
80
Vilhelm von Ehrenheim retweeted
May 5
people need to be pricing in that both the speed of inference and the cost of inference is going to drop exponentially.
it takes years for these breakthroughs to appear, but they ARE coming.
as a result, the rest of the system (which could previously be bad because inference was so slow) is going to have to speed up. you will need to launch sandboxes in <100ms to keep up.
May 5
Introducing SubQ - a major breakthrough in LLM intelligence.
It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA),
And the first frontier model with a 12 million token context window which is:
- 52x faster than FlashAttention at 1MM tokens
- Less than 5% the cost of Opus
Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention).
Only a small fraction actually matter.
@subquadratic finds and focuses only on the ones that do.
That's nearly 1,000x less compute and a new way for LLMs to scale.
1
1
48
6,150
Apr 26
Reading the decision stack by @bfgmartin! This book really resonates w my experience in pretty much all organisations I have worked in. Big or small.
The book draws on analogies w physics - Speed, Velocity and Momentum. Very relevant in engineering today where speed is easier to achieve than ever. Being mindful about how to shape this into velocity and momentum is more important than ever!
Highly recommended!! 🚀
41
Apr 23
Building out better AX @getqatech! 🚀
Verification from inside the cc sessions to sanity check my crazy new features is soo nice!
1
33
Apr 19
Everyone shipping with Cursor or Claude Code today, what’s your actual verification workflow? Reading every diff? Running tests? Trusting it?
Genuinely curious what people actually do, not what they say they do.
@ericzakariasson @bcherny @antonosika what are you seeing?
2
70
Vilhelm von Ehrenheim retweeted
Apr 17
Visual PR Testing with AI: Validate every PR with AI that runs tests for you producthunt.com/products/qa-… by @whtmnk h/t @fmerian for hunting!
1
2
98
Vilhelm von Ehrenheim retweeted
Apr 17
2
21
111
Vilhelm von Ehrenheim retweeted
Apr 17
QA.tech: Ship faster with AI agents that create and run your QA tests by @whtmnk, @getqatech, @while, and more producthunt.com/products/qa-…
1
2
48
Apr 11
Honestly feel like this is the same as we have done in ML for ages. Understand the problem you are trying to solve and set up good evals. With solid train and test splits.
That way you can improve your model and reason about performance. There are no shortcuts. Build solid evaluation strategies.
53
Apr 11
Ramp continue to lead the way in internal AI tooling. Pretty amazing.
I think building tooling for fostering a solid setup for everyone in the org will have a huge impact.
50
Mar 31
Really good breakdown. Regardless if the leak is true or not this is a great blueprint for designing agents.
69
Mar 31
Jade Rubick makes the case that QA should become “Automated Verification Engineers.” Fast, automated feedback on every PR. No gates. No handoffs.
I’d go further. The AVE shouldn’t be a person. It should be an agent.
rubick.com/should-qa-exist/
26
Mar 30
If your infrastructure can’t let a new hire push to production safely on day one, you can’t let an agent do it either.
The need for good engineering practices aren’t replaced by AI. They’re amplified by it.
25
Computer use now available in Claude Code.
Waited a long time for this.
Computer use is now in Claude Code.
Claude can open your apps, click through your UI, and test what it built, right from the CLI.
Now in research preview on Pro and Max plans.
9
4
66
11,306
Vilhelm von Ehrenheim retweeted
If your tests break every time the UI changes…
Are they testing anything?
This deep dive explores agentic testing and a better approach to QA 👇
hackernoon.com/what-is-agent…
22
52
901
4,203,362
Mar 5
Stripe built "Minions." Ramp built "Inspect." Different companies, same architecture.
Cloud sandboxes. Isolated environments. Agents running in parallel.
1,300 PRs/week (Stripe). 30% of all PRs (Ramp). Neither could do this on localhost.
37
Mar 2
I think the most hyped technology in a decade will deliver most of its value on dependency updates, CVE patches, and lint fixes.
Not new features. The maintenance backlog nobody ever got to.
64
Vilhelm von Ehrenheim retweeted
Feb 21
QA tech's @while on what an Agent is:
"The word agent is super old. It's been around for thousands of years and it's actually kind of some of the first words we found in literature."
And the original meaning still holds up perfectly today.
An agent is simply someone acting on your behalf, with your interest in mind.
"Everybody tries to define them in like loops with tools and interacting with environments and things. All of those things are really important, sure. But it doesn't really define what the job of an agent is."
The job is to do something. Take decisions. Act in your best interest.
"That's what a real estate agent is doing. That's what any other kind of agent we call agents normally."
2
1
103
Mar 2
Co-founder of QA.tech. We build AI agents that test software so humans don't have to.
I've been building and thinking about agentic systems for two years. Going to start sharing what I'm learning here as well.
38