When hiring your first 10-15 engineers, forget ‘10x engineers,’ focus on ‘engineers who give a sh*t’ - who care about building it right, think about the next 4 releases, learn to become domain experts, eager to hear user feedback, care about the business and their part in supporting it, will burn the midnight oil to get it done without being prompted, are excited to write the docs, push those around them to be better, want to be part of the interview circuit, keep linear/jira clean and up to date, start to dig into the competitors, add that extra polish because they can’t imagine doing it any other way.
Companies are representations of the people that build them.
When building a company, find team members who really ‘give a sh*t,’ it’s the only way.
We're using AI agents to automate everything. Except the engineering of AI agents themselves.
That seemed backwards to us.
What if you could train your own specialized AI agent with just 15 minutes of your time instead of months of AI engineering?
Would you keep renting the same expensive generic models as your competitors, endlessly patching prompts and harnesses until you hit a wall?
Today, we're launching the agent that builds, trains, and continuously improves AI agents. 🚀
Meet FAiE, our Frontier AI Engineer. Describe what your agent should accomplish, and FAiE designs evaluations, diagnoses failures, curates data, selects training strategies, and runs experiments.
15 minutes of your effort. A few hours of automated AI engineering. Your own specialized agent, with its model weights trained for your task.
Already have an agent? Keep your harness. Connect it to @oumi_ai and train the model behind it with SFT, DPO, or GRPO (Private Preview).
The agent that builds your best agent.
Excited to see what you build: platform.oumi.ai/
Live demo today at 10 AM PT: oumi.ai/events/fall26
We're using AI agents to automate everything. Except the engineering of AI agents themselves.
That seemed backwards to us.
What if you could train your own specialized AI agent with just 15 minutes of your time instead of months of AI engineering?
Would you keep renting the same expensive generic models as your competitors, endlessly patching prompts and harnesses until you hit a wall?
Today, we're launching the agent that builds, trains, and continuously improves AI agents. 🚀
Meet FAiE, our Frontier AI Engineer. Describe what your agent should accomplish, and FAiE designs evaluations, diagnoses failures, curates data, selects training strategies, and runs experiments.
15 minutes of your effort. A few hours of automated AI engineering. Your own specialized agent, with its model weights trained for your task.
Already have an agent? Keep your harness. Connect it to @oumi_ai and train the model behind it with SFT, DPO, or GRPO (Private Preview).
The agent that builds your best agent.
Excited to see what you build: platform.oumi.ai/
Live demo today at 10 AM PT: oumi.ai/events/fall26
Introducing voxl
extremely fast remote desktop streaming written in rust
→ 60 fps + low latency
→ macOS, Windows and Linux
→ One WebSocket, so it works through proxies and sandbox portals
→ Agent APIs: /screenshot, /record, /wait
Huge news: @Vanguard is acquiring @Altruist 🚀🚀🚀. Congrats to @jasonwenk & the Altruist team!! It’s been awesome working together for the past 8 years as you’ve revolutionized wealth management 🙏🙏. May you have many more “Altruist moments”!! @Venrockwsj.com/business/deals/vangu…
I am regularly asked what we @Venrock look for in frontier & deep tech teams at the Series A:
1. technology is past scientific possibility
2. customer commitment is real, not just interest
3. unit economics strengthen with deployment
4. team is uniquely credible for the next bottleneck
Most rounds get stuck on #2.
🎙We’re back with a conversation between one of the greatest minds in AI research @MelMitchell1 of the @sfiscience and @ethanjb. Tune in for perspectives on AI capabilities, performance benchmarks, and the path forward that’s grounded in evidence: bit.ly/3PFyYAF
🎙We’re back with a conversation between one of the greatest minds in AI research @MelMitchell1 of the @sfiscience and @ethanjb. Tune in for perspectives on AI capabilities, performance benchmarks, and the path forward that’s grounded in evidence: bit.ly/3PFyYAF
🚨 The era of general-purpose AI is over.
Today we're launching Oumi. 🚀
The platform that lets any team build custom AI models — in hours, not months.
Just describe what you need. Oumi builds it. #VibeML
Higher quality. Lower cost. Fully yours.
The core idea behind Newton’s physical world model is learning about physical dynamics directly from sensor data (e.g. vibrations, pressure, magnetic signals) rather than relying on human-generated artifacts like text.
Language models, on the other hand, have a broad understanding of how we talk about the world, but they’ve never sensed a vibration or temperature change.
In our new research at @PhysicalAI, we introduce a lightweight, parameter-efficient method to bridge these two worlds. This results in a model that can describe and reason about what sensors are detecting and provide qualitative interpretations: the water is about to boil, the machine is behaving abnormally, or the room needs ventilation.
These interpretations can directly guide actions (e.g. turn off the stove, schedule maintenance, open a window etc) bringing physical intelligence and human language together.
Perhaps surprisingly, aligning Newton with language is lightweight. Training took just 20 minutes on a single H100, thanks to the emergent structure in Newton’s physical encoder, which naturally clusters similar physical behaviors, even though it was never trained on these labels.
Our paper presents a proof of concept system that can answer free-form questions about local physical conditions like road surface type, ride comfort, or anomalies, while retaining the general world knowledge of the LLM.
This work demonstrates a promising path toward interpretable, language-grounded physical intelligence, where machines can explain what they sense and humans can reason with them intuitively.
Huge thanks to my co-authors Hasan Doğan, Laura I. Galindez Olascoaga, Muhammed Selman Artıran, and O. Serdar Gedik.
Read more & find the paper here:
archetypeai.io/blog/language…
$15-25 per PR
No Zero data retention
20min for a review?
If you want
orders of magnitude cheaper with really high quality and low noise
always zero data retention (even for managed saas)
< 2 min for a review
just install gitar.ai
What does a Defense Neo-Prime look like?
When companies like Anduril Industries, Shield AI, or Epirus emerge, three things are usually true very early:
1) A clear step-change capability
e.g., autonomous targeting, microwave drone defeat, autonomous ISR.
2) Immediate operational demand
units actively trying to buy the system.
3) System-level control
not just a component, but a platform that owns the architecture.
The limiting factor is no longer how many operations a chip can perform. It is how quickly data can move to and from those compute units — and how much energy it costs to do so.
open.substack.com/pub/ethanj…
Three bottlenecks are converging.
Bandwidth density.
Even with HBM, large models saturate memory channels. More stacks mean more packaging complexity and tighter yield constraints.
Energy per bit.
At scale, moving data often consumes more energy than computing on it. In dense racks, this becomes a thermal limit.
System integration.
Performance deltas now come from topology, interconnect, and memory coupling — not just transistor counts.