After a long 2-week vacation, here's finally a new video! Let's finally explore why bigger context windows make your agents worse, and, as a bonus, 10 context management techniques that actually work! Watch it here: https://
youtu.be/PQglg4N_jxo
AGENTS
-
Why larger context windows can degrade AI agent performance
By
–
-
Anthropic conference: agents orchestrating other agents
By
–
Hier 19 Mai, Anthropic a tenu une grosse conférence à Londres.
— Jouhatsu | AI Influence Operator (@Jouhatsu_ai) 20 mai 2026
6 ingénieurs qui ont créé Claude ont partagé ce qui va changer ta façon de builder pour toujours.
Gardez-la précieusement en signet 🔖
Et personne n'en parle encore:
→ Des agents qui se pilotent entre eux… pic.twitter.com/Qae7TYuopMHier 19 Mai, Anthropic a tenu une grosse conférence à Londres. 6 ingénieurs qui ont créé Claude ont partagé ce qui va changer ta façon de builder pour toujours. Gardez-la précieusement en signet Et personne n'en parle encore: → Des agents qui se pilotent entre eux
-
LangSmith Sandboxes GA: Agents Get Isolated Runtime Environment
By
–
ICYMI: LangSmith Sandboxes are GA Agents get a real filesystem, shell, and package manager. Isolated from your infra. Works with Deep Agents, Open SWE, or your own code. Auth with the same API key you already have. No new runtime to build or manage.
-
Query-Aware Context Compression in RAG Systems
By
–
Context compression isn't new in RAG. Our contribution is making it query-aware, citation-preserving, and fast enough for orchestration. Read the full research blog:
-
Embedding AI into Business Apps Matters
By
–
The next AI winners in business won’t be the ones with the most tools. They’ll be the ones embedding AI inside the apps where work already happens. That’s why the latest Power Apps updates matter: Copilot in business apps App skills Agent supervision in context
-
Virtual Town Experiment Compares Agent Behavior Across Different LLMs
By
–
Emergence AI built five identical virtual towns and gave each one 10 agents.
— The Rundown AI (@TheRundownAI) 20 mai 2026
All had the same rules and starting conditions. The only thing that changed was the model running the agents.
15 days later, Claude Sonnet's town had zero crimes.
GPT-5 Mini's agents didn't break… pic.twitter.com/xUKw8Gc42YEmergence AI built five identical virtual towns and gave each one 10 agents. All had the same rules and starting conditions. The only thing that changed was the model running the agents. 15 days later, Claude Sonnet's town had zero crimes. GPT-5 Mini's agents didn't break
-
AI agents and recursive performance optimization in code training
By
–
This is the part of AI progress that will sneak up on people. Agents grinding away at training code, kernels, architecture tweaks, and benchmarks until they start finding real performance gains. Coding agents are struggling at this, but this breakthrough looks promising.
-
New Course on Building Self-Iterating AI Agents for Media Generation
By
–
New course: Build AI agents that generate images and videos — an under-explored frontier. A key to performance is having the agent evaluate its own output, and iterate to improve quality. This short course is built together with @googlecloudtech and taught by Katie Nguyen and… pic.twitter.com/xxuEvZs8HP
— Andrew Ng (@AndrewYNg) 20 mai 2026New course: Build AI agents that generate images and videos — an under-explored frontier. A key to performance is having the agent evaluate its own output, and iterate to improve quality. This short course is built together with @googlecloudtech and taught by Katie Nguyen and
-
SambaCloud platform for deploying frontier AI models and agent pipelines
By
–
SambaCloud is SambaNova’s cloud platform for deploying frontier AI models.
— SambaNova (@SambaNovaAI) 20 mai 2026
It runs models like @MiniMax_AI M2.7 at extremely low latency thanks to purpose-built AI hardware—making it a strong option for production agent pipelines where speed is critical.
Start building:… pic.twitter.com/mUYKhx8rMZSambaCloud is SambaNova’s cloud platform for deploying frontier AI models. It runs models like @MiniMax_AI M2.7 at extremely low latency thanks to purpose-built AI hardware—making it a strong option for production agent pipelines where speed is critical. Start building:
-
Improving Grok’s reasoning capabilities for agentic tasks
By
–
As Jason mentions, it’s a top priority. Grok was initially trained to give quick answers in chat mode, rather than think hard for as long as needed to get the task right, which is what matters in agentic mode. We are fixing that.