With a small but fully relentless team of developers, we developed a state of the art AI engineering agent - NEO @withneo Experimented and iterated repeatedly until we architected NEO to excel at reasoning and implementing solutions for complex AI/ML problems. 👇

OpenAI releases GPT-5 in august Anthropic releases Sonnet 4.5 in september soon after, Google launches Gemini 3, which may outperform both xAI follows with Grok 5 month later, a chinese open-source model tops them all at a fraction of the cost the cycle continues...

My 2 cents: This is the core problem with vibe coding and nobody is looking deeply at it. This problem needs attention!

Unless you push the LLM to fix the root cause or provide hints, it faces major difficulty fixing and always returns a workaround solution. This is true for all the major foundational LLM providers.

Maybe they are optimized for writing more code, fixing root cause reduces the need for more LLM calls. And workarounds often end up needing even more LLM calls as they are not enough to handle edge cases or major scenarios.

The biggest problem with LLMs for code generation is this: When given a complicated bug to be fixed, it always (literally always) gives a workaround instead of fixing the root cause. Seems like all LLMs prioritize workarounds as quick fixes instead of figuring out the root cause.

I can already imagine Dom’s life after he met his children at the end in Inception. Or Cooper after flying out to the new world in Interstellar.

What we usually see in movies is a small period in protagonist’s life showing a one sided narrative. With AI based film-making, creators would soon be able to explore other shades as well and as a viewer we’d see whole life of the character unrolling in front of us. INSANE!

@withneo Access releasing soon:

Introducing NEO: The first Autonomous Machine Learning Engineer. It works like a full-stack ML engineer that never sleeps: handling data exploration, feature engineering, training, tuning, deployment, and monitoring, end to end. Powered by 11 specialized agents, NEO runs

A 6-person startup with $500k has just outperformed Microsoft! OpenAI's MLE-Bench is a benchmark that tests agents on Machine Learning engineering tasks. NEO @withneo, the first autonomous MLE agent, scored 34.2% vs Microsoft’s 22.4% on the benchmark. This is huge!

Thrilled to share that NEO (@withneo) is now the SOTA Machine Learning Engineering agent on OpenAI's MLE-Bench with a score of 34.2% outcompeting Microsoft's RD Agent. Checkout the official leaderboard: https://t.co/D7HvpHr2im @karpathy @ilyasut @gdb @AndrewYNg

Neo is officially the number 1 fully autonomous Machine Learning Engineering agent. Evolving as we speak to work on more complex AI/ML solutions and tasks across the spectrum. From Gen AI to classical ML 👇

🧠🚀Meet NEO – the AI problem‑solver that turns chaotic ML challenges into clear wins! Want to see how it cracked a tough brain‑tumor classification task in record time? Keep reading for the key moves and results!

Its official Neo is the number 1 autonomous machine learning engineer in the world Leading with 34.2% score on the MLE bench, beating Microsoft’s RD agent. @ilyasut @AndrewLeeMaas @karpathy please check this out. Here’s the GitHub: https://t.co/ToNImWvc4R

See how @withneo tackled the RNA degradation prediction competition on MLEBench and achieved a gold medal 👇

A workflow on how @withneo solved a hard ML problem on MLEBench and achieved a gold medal result on it. Neo can solve complex end to end data science and ML tasks autonomously and collaborates efficiently with human MLEs. Early access:

Meet NEO, the AI problem‑solver that turns messy data challenges into gold‑standard results in record time. Follow the thread to see how NEO cracked a tough RNA degradation prediction competition on MLE-Bench and hit a 🎯 MCRMSE of 0.28491!

NEO is capable of building strong ETA prediction models and perform comparative evaluation between them on your command See @withneo in action:

Meet NEO - First fully autonomous Machine Learning Engineer https://t.co/oZjyctT4Pg @withneo is not just a copilot, it's a multi-agent system that thinks, learn and builds like a real ML engineer on complex data, ML and Gen AI pipelines. Early access: