Ofer Mendelevitch

Yesterday I received my author copies of the "Hands on RAG for production" - it was very exciting to hold in my hand the actual physical copy, after so many months of work. amazon.com/Hands-RAG-Prod… @OReillyMedia

Boris Cherny, Peter Steinberger, and Andrew Ng have all been circling the idea of loop engineering - stop prompting the agent, start designing the system that prompts it for you. The tooling for a single loop and a single agent is basically solved at this point. But when we get to 10 loops, we're living in chaos. Context doesn't travel between sessions, nobody can see the whole board, and your attention becomes the bottleneck - five terminals demanding a decision instead of one. That's the problem we built Jam to solve: a shared surface where teams of coding agents coordinate work together in Claude Code, with persistent agent identities, a room where they talk to each other, and a live board so you're not reconstructing the state of the project by hand. Full piece from our own @ofermend: band.ai/blog/loop-engi…

Prompt engineering was about telling an agent what to do. Loop engineering is about giving it a goal and letting it keep working. But complex goals rarely need just one agent. They need a team. Meet Jam: loop engineering, powered by collaborating agents. band.ai/blog/loop-engi…

@dorsa_rohani You can do this already with Jam (also cross machine) band.ai/use-cases/jam

We made Claude Code multiplayer! Use it with other people in the same terminal, and connect your Claudes so they can all talk to each other.

@catmanyau Claude / Codex are quickly becoming the main tools of work daily, and not for vibe coding but for real projects, not vibe coding (although that does exist a lot too). I really meant it when I said it's becoming the new normal...

@ofermend i actually think the 'well, with Claude' qualifier is a useful signal—it distinguishes between a weekend prototype and a years-long build.

A lot of developers - when they say, "I built X", feel the need to qualify: "…well, with the help of Claude Code" (or Codex). You don’t need to apologize. We know. It's what we expect. That's the new normal.

We're so excited to be hosting the first SF Bay Area Multi-Agent Systems meetup, with @DrataHQ. It’s time to start talking about what happens when your AI agents actually collaborate with each other - not just work as orchestration diagrams. Join us on Thursday, July 30 in San Francisco for a real conversation about how agents communicate, coordinate, and work as teams. We’ll be holding a live panel discussion with open Q&A, and a room full of people actually building this. Plus delicious food and drinks (don’t worry, we got you!) RSVP: luma.com/hs3x4zt6

Claude Code and Codex are awesome, no doubt. But do we need a "Coding Agent Detox Day" - say, once a week - to ground us back in how we used to work? If that idea makes you uncomfortable, that's the signal.

so preparing for France vs Spain on Tuesday, I asked Atlas about the history between the two countries. Turns out they only played each other once - in 2006 Guess who won? atlasincredible.com/story/46

Excited for this hackathon next week in SF - join to build next gen agent to agent collaboration with band.

The default answer to "how do agents collaborate" is an orchestration graph: define nodes, define edges, hope it holds when you add agent #5. Omer Fogel spent last week at @aiDotEngineer World's Fair and he argues the real answer looks more like a mesh. band.ai/blog/ai-engine…

Build production-ready AI applications with Hands-On RAG for Production by @ofermend & Forrest Sheng Bao! Published By: @shroffpub & @OReillyMedia MRP: ₹2,600 Link To Buy: tinyurl.com/3j5u58nz #HandsOnRAGForProduction #RAG #GenerativeAI #AI #LLM #MachineLearning #OReilly

@credibledatahq Getting ready for the game tomorrow #Fıfaworldcup2026

Super cool from @credibledatahq - it's amazing how easy asking questions from data has become with LLMs and a good Malloy semantic model. Get ready for your watching parties :) #FIFAWorldCup2o26

Happy to share that the book is now (finally) also available as paperback amazon.com/Hands-RAG-Prod…

Excited to be speaking at the @Snowflake event with @HetzVentures - I'll be sharing cool ways to build agent-to-agent collaboration with Snowflake agents using @band_hq RSVP here: luma.com/2p7b6lbl

Amazing day 3 at @aiDotEngineer with lots of great discussions at the @band_hq booth about agent to agent collaboration, and especially coding agents collaborating through jam. And always great to run into @swyx, who makes it all happen…

@aiDotEngineer Less orchestration. More collaboration. Even the orcs get it. More orc adventures tomorrow at the @band_hq booth.

we put the orcs in agent orchestration

@rasbt congrats @rasbt - I just got my author copy a few weeks ago and it's a great feeling to have the physical book in your hand. Can't wait to read yours.

After 18 months of writing, coding, and experimenting, Build a Reasoning Model (From Scratch) is finally out! My first copies just arrived! 📚 440 full-color pages. Inference scaling, reinforcement learning, and distillation from scratch.

AI Eng world fair is this week in Moscone center. Looking forward to sharing @band_hq with everyone and interesting discussions about agent to agent collaboration at scale. And always fun to see old friends @seldo @steveonjava @charles_irl