Omnigent: Open-Source Meta-Harness for AI Agents | Matei Zaharia
8.9K views · Aug 10, 2026 · Science & Technology
Comments · 7
@GambillDataEngineering · 1 month ago
I Love Omnigent! Thank you all, it has been fantastic. I created a "dreaming" skill that I use to help to distil the "memories" and learning across the context windows into a central "dream" file so it can continue to learn and retain key points across the sessions and harnesses. Omni has really been a game changer for me especially within databricks.
@ssrikrishnan · 1 month ago
Very interesting way to solve problems related to collaboration, cost control and context switching between the models when we run out of quota and a lot more!!
1
@JT91notabot · 1 month ago
State transformation protocol would help so every agent knows what reality looks like in real time. We need a protocol that appends receipts and calculates current accepted state through a deterministic rule.
2
@gordeyvasilev · 1 month ago
👍
@absolute___zero · 1 month ago
yeah, there is a huge problem right now on the AI market. Agents work excellent with small code bases, but when your project becomes larger than 1 million code lines, you just spend lots of time on small refactorings , because the agent has difficulties understanding large code bases. My current project is 1.5 million code lines (C-lang) and I have had to stop development because I am just losing token quota. So am currently writing my own harness, a harness that orchestrates multiple Codex/ Claude Code agents. The key is to decompose complexity, and then you can write 50 million code line projects (like linux kernel) without any problems. OpenAI / Anthropic will never make their agents for this scale because this scale requires more tokens to be consumed for design, but AI companies aren't interested in large-scale customers who write million-line code projects, they are interested in supplying demand for small apps/medium size apps. Therefore, you need to write your own harness. When I finish mine, I will publish it.
1
Up next
Introducing Omnigent: an open meta-harness – Matei Zaharia, Co-founder and CTO, Databricks
Databricks · 34K views
The Agent Cloud: Databricks’ Bet on the Future of AI — Matei Zaharia and Reynold Xin
Latent Space · 40K views
Continual Learning: How AI Agents Get Better With Every Use | Arjun Karanam, Trajectory
Sequoia Capital · 39K views
Why Agentic Systems Need Ontologies — Frank Coyle, UC Berkeley
AI Engineer · 429K views
What Happens to Developers When AI Writes All the Code? (EP5)
Silicon Valleys · 1.1K views
Databricks Omnigent Deep Dive with Matei Zaharia: Collaboration and Control Layer for AI Agents
Josh Bogran · 2.3K views
When to Build Your Own Agent Harness | Harrison Chase, LangChain
Sequoia Capital · 96K views
Building Agents on Databricks with Custom Apps and Omnigent
Databricks · 7.4K views
Building And Structuring An AI Native Company
YC Root Access · 52K views
Building Docs for Agents, Not Humans: Inside OpenWiki
LangChain · 11K views
5 Best Practices for Building AI Agent Skills
IBM Technology and IBM Developer · 79K views
Ajeya Cotra – "This might be the clearest warning shot we ever get"
Dwarkesh Patel · 1.5M views
Yann LeCun on What Comes After LLMs
Unsupervised Learning: With Jacob Effron · 403K views
How Companies Are Building Their Own Intelligence | Sonya Huang, Sequoia Capital
Sequoia Capital · 73K views
Agent Memory EXPLAINED - Complete Architecture
Hugging Face and Alejandro AO · 59K views
Are Agents About to Replace Software Engineering? | Lauren Tan and Roshan Sadanani
MTS · 77K views
Managed Deep Agents explained in 20 minutes
LangChain · 18K views
BREAKING: Trump, Musk, Zuckerberg, Amodei Take Reporters' Questions After New AI Accords Announced
Forbes Breaking News · 1M views
AI FDE at Databricks (E.42)
Free Form AI · 1K views
Why Graph Engineering will 10x your Claude/Codex
Greg Isenberg · 218K views