Whether you are GPU poor or GPU rich, today's release of PyLate has something for you! GPU maxxers: MaxSim kernels greatly speed up training while lowering the memory requirements CPU enjoyers: TACHIOM enables lightning fast multi-vector indexing and search directly on CPU

What happens when multi-agent systems stop relying on a central “controller” agent? Can agents coordinate by sharing results directly with each other? Introducing Decentralized Language Models (DeLM): we let agents coordinate asynchronously through a shared context. Agents claim tasks from a queue and write back compact, verified results as they finish, making progress visible to all workers without requiring a main agent to merge, filter, and rebroadcast it. New paper with @azaliamirh!

Very excited to release late-interaction-kernels (LIK): fused Triton kernels for MaxSim, the scoring step behind ColBERT, ColPali & LateOn. 🚀 Numerically equivalent to PyTorch at a fraction of the memory, with day-0 support in PyLate & colpali-engine. (1/N 🧵)

Just landed nested subagent support in Claude Code Starting to experiment more with agents kicking off agents as a way to better manage context. Capped at depth=5 to start, going out in today’s release. Lmk what you think!

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore. You should be designing loops that prompt your agents.

Argus-Retriever: Vision-LLM Late-Interaction Retrieval with Region-Aware Query-Conditioned MoE for Visual Document Retrieval Introduces a query-conditioned late-interaction visual document retriever. 📝 arxiv.org/abs/2606.04300 👨🏽‍💻 github.com/DataScienceUIBK/A…

We set out to build a better retriever, so we looked for the hardest IR benchmarks. For each, we asked how much headroom remained by running oracle reranking with a frontier LLM. Most had little room left! So we built OBLIQ-Bench to study much harder search queries than before.

Using agents effectively requires embracing their stochastic nature and setting up a framework for them to reason rather than giving them a pile of rules. Harness engineering and knowledge base curation does not (and cannot!) rely on all information being pulled into context—and given things like autocompaction over long horizon work, context is constantly getting blitted so you can't really rely on things being in context either. In general you want to tell agents your expectations on how/when/what type of context they should seek and how the tools in their environment can help them complete their work. You need to tell the agents what they are working on, what parts of it matter, and how they should approach tasks. Tell them about common tasks for the things you’re working on and where they can learn more about them. Make your agents collapse a prompt they are given into a paved workflow.

3 weeks left til @aidotengineer world's fair! if you want to get on this year's map of top ai engineering companies, theres a few spots left we are sold out of: - presenting sponsors - model lab sponsors - platinum sponsors - gold sponsors the big spots left are for the official afterparties - welcome reception, networking night, and world cup quarterfinal bundles if interested - drop a note to sponsorships@ai.engineer detailing size/scale of interest (below is 2025, we are recruiting 2026 now) if you are attending - BOOK YOUR HOTELS BY TMR AS THE ROOM BLOCK DISCOUNT EXPIRES TMR

People are increasingly worried that AI tools make us overreliant. But how do we actually measure this? We introduce Offloading Score, a measure of reliance based on the fraction of cognitive effort offloaded to AI while completing a task. In a controlled user study, Offloading Score detects increased reliance under time pressure, while several common alternatives do not. (1/9)