Marko Kraemer (@markokraemer)

Pinned post

Introducing the open-source @kortix CompanyOS:

Your entire company as code: agents, skills, connectors and rules in one repo you own. A workspace for your team, a framework for your agents.

Sovereign, multiplayer, extensible AI.

1:51

Marko Kraemer @markokraemer 2h

is there moat in humans or can every task be compute-loop-maxxed? Where does human edge matter today & will it shift in the future – is there a defining thing that keeps MOAT, competitive edge alive?

Today the general consensus is TASTE which can actually be better generally described as JUDGEMENT is still a moat & u can see this matter extremingly for open-ended tasks, while close-ended tasks become incredibly gameable compute loop maxxed

When it comes to something like openkernelchallenge.dev I am incredibly curious how much actual unique HUMAN knowledge went into the @Yuchenj_UW @databricks kernel factory,

And even if the core skill here is to setup the COMPUTE LOOP MAX factory, the loop today – if thats the core differentatior – is that inself a task, the designer/director that will stay unique or it wont

Anything creative today
- Vibe coding / design engineering, ux/ui
- Vibe coding / System design!
- .. so much more subsets not too list it all
- AI movie making higgsfield.ai/@aist/projects…, higgsfield.ai/@aist/projects…
- SLOPCORE, Audio/video x.com/anabology/stat…
- more..

has come so far – we have entirely crossed the uncanny valley. BUT A LOT of HUMAN orchestration & direction under human TASTE/JUDGEMENT goes into making it feel good. The diff between slop & SLOPCORE

Will that stay unique to us and humanity`s role is going to be that of a director – or is the AI Director better, or its not even quantifiable cuz its different... and subjective? There is objective truths for good TASTE/JUDGEMENT of any sort, but these truths in some cases are dynamic & change with trends over time especially when looking at design (in the visual sense)

This "Directing", "Orchestrating" – long-context, long-horizon, deep-judgement is likely also going to be automated and the level of abstraction (& autonomy) is going to increase even further

may the SI be with u

openkernelchallenge.dev Muse Spark Open Kernel Challenge Make Meta’s Muse Spark Open run faster. Start by optimizing its GPU kernels, then speed up how SGLang serves it. The fastest win $100,000 in prizes.

Yuchen Jin Verified account @Yuchenj_UW 6h

Databricks just took 2nd place in Meta's Muse Spark Open Kernel Challenge, out of 300K+ submissions!

Our kernels achieved an average 2.5X speedup over SGLang kernels.

We built a Kernel Factory where AI agents autonomously hill-climb their way to better kernels. We found Opus 5.5 to be the strongest model for GPU kernel optimization.

These kernels go straight from our Kernel Factory into production, powering Databricks Unity Gateway and AI inference for our customers.
View quoted post on X

Marko Kraemer @markokraemer 2d

long @theresidency, was my first home when I came to San Francisco. Forever grateful!

the residency Verified account @theresidency 2d

new source of funding on the market

$120k from the residency

for early stage, deep tech founders, working on a post-agi future

tell your friends, the singularity is here.

🔗 applications for winter 2027 close nov 29th
apply.livetheresidency.com/apply

View quoted post on X

Marko Kraemer @markokraemer 2d

bro @UseCorgi sales-ops are insane, literally 1 minute after signing up I am getting a call & get booked for a call

Marko Kraemer @markokraemer 3d

How to agent scale-max:

1. Run the agent in a @Cloudflare DO / @deno_land CELLD / @rivet_dev Actor

2. Add the full heavy sandbox when u need a full machine. Ideally its colocated and close to the above. @daytonaio @e2b @boatdotdev & co.

3. Have a decouple shared FS that can be mounted anywhere (@archildata style)

Larsen Cundric Verified account @larsencc 3d

I was wrong about agent sandboxes.

In my article I said pattern 1, where the agent runs on your infra and only calls a sandbox for tools, doesn't work at scale. I moved everything to pattern 2 and put the whole agent inside a VM.

At the time that made sense but today I'd pick pattern 1.

For your own personal agent, a coding agent, or a small project, isolating the whole agent in a VM is still a great default. I would keep doing it.

At real multi-tenant scale it falls apart.

There will be more agents than humans. Every human gets one computer. Giving every agent its own VM running 24/7 means billions of CPU-hungry machines, and we're already short on compute.

Most of what an agent does doesn't need a machine at all. It thinks, calls APIs, reads pages, and waits. Nobody needs a 4 CPU VM sitting idle so an agent can run a simple JS script once an hour.

What I'd build now:
> the agent loop runs on shared infra that scales
> it gets a sandbox only when it has to run untrusted code
> sandboxes spin up in seconds and die right after
> zero state on the machine, everything lives outside it

VMs are a tool the agent calls, but they shouldn't be where it lives.
View quoted post on X

Marko Kraemer

Latest posts 5