This is what I read about and tried this week.
AI
A graph-based tool routing library for LLM agents. Instead of asking your LLM "what tool should I use next?" at every step, Self-Healing Router models your tools as a weighted directed graph and uses Dijkstra's algorithm for deterministic routing. When a tool fails, edge weights update to infinity and the path recomputes automatically. The LLM is only called as a last resort — when no feasible path exists. - link [tool] - ( Added: 2026-03-09 11:12:40 )
More about sandboxing agents. Worth reading - link [opinion] - ( Added: 2026-03-08 18:37:02 )
Claude Code introduces evals for skill creation meaning you can now measure and improve your context - link [best-practice] - ( Added: 2026-03-07 15:22:11 )
Very nice tool that allows you to build voice based applications using ultra low latency response times - supports cloud based and local models - link [code-repo] - ( Added: 2026-03-06 14:22:02 )
From Alibaba, an open source project that provides a new specification for creating sandboxes for your agentic applications. Supports some agents/frameworks/tools - link [tool] - ( Added: 2026-03-06 11:28:08 )
A runtime MCP layer that provides some enterprise level capabilities to your agentic applications - link [tool] - ( Added: 2026-03-04 20:58:03 )
Even if LLMs continue to get exponentially more powerful, there will be core engineering techniques that make LLM-powered software more reliable, more scalable, and easier to maintain. If you want to know what those are, check out this repo - link [opinion] - ( Added: 2026-03-04 20:35:14 )
The Agentic Engineering Guide. 10 parts, 33 chapters. Context engineering, authorization, cost control, evaluation, adoption playbooks. Rather than a hands on tutorial, this is more about the decisions you face when adopting agents, in the order you face them. - link [opinion] - ( Added: 2026-03-04 19:58:51 )
A new project that provides you with options for running your agentic ai in isolated sandboxes, both local and prod scale - link [tool] - ( Added: 2026-03-04 19:47:59 )
From the folks behind the warp terminal, comes Oz he orchestration platform for cloud agents. Spin up unlimited parallel cloud coding agents - programmable, auditable, and fully steerable. - link [tool] - ( Added: 2026-03-04 18:26:42 )
Worth reading if you think that keeping Claude in a container is all you need when it comes to sandboxing strategy. Yeah, this included me too! - link [best-practice] - ( Added: 2026-03-04 09:21:55 )
If you are looking to get some hands on training on Claude, then you can do worse than review this selection of courses. - link [tutorial] - ( Added: 2026-03-04 09:19:23 )
Very cool new library that has a really nice UI to help you keep track of what your agentic applications are doing. Instruments everything and allows you to replay and view all previous transactions via the nice UI - link [tool] - ( Added: 2026-03-03 17:26:33 )
A really great post from Mike Chambers that looks at "on demand application generation" and how the current state of where we are allows you to build this whilst looking at emerging new tools/frameworks that are making it simpler than ever to build apps "on demand" - link [opinion] - ( Added: 2026-03-03 10:21:40 )
OSS
- Latest post on the fun and games that are copyright and AI. Some good references in this post, that keeps you up to date. - link [opinion] - ( Added: 2026-03-08 09:14:14 )
CODING
A new way of think about how to orchestrate ai coding assistants - link [opinion] - ( Added: 2026-03-06 14:22:48 )
Ghost is a CLI that flips the git workflow: instead of committing code, you commit prompts. An AI coding agent generates the artifacts; the commit captures both the intent and the output. Your git history becomes a chain of prompts + their results. - link [tool] - ( Added: 2026-03-04 09:20:54 )
A plug-and-play framework for generating projects with Claude Code CLI — optimized for speed, token efficiency, and consistent output across any language or framework. - link [tool] - ( Added: 2026-03-04 09:20:16 )
A new tool for users of Kiro CLI on macOS, that provides a nifty menu bar toggle to manage your MCP servers - link [tool] - ( Added: 2026-03-03 08:04:00 )
AWS
Demo code that shows an example of what you can do with Durable Functions in an an event driven architecture flow - link [code-repo] - ( Added: 2026-03-06 11:13:34 )
If you are looking to take AWS certification tests, then you can get vouchers which are valid for 12 months across Foundation, Associate, and Professional tiers - link [best-practice] - ( Added: 2026-03-04 09:25:38 )
Generated on 2026-03-09 by Bookmark Manager | 3rd Mar 26 to 9th Mar 26 | Total bookmarks: 21
Top comments (2)
The self-healing router concept is clever — using Dijkstra for tool routing instead of asking the LLM at every step removes a whole class of latency issues. Have you tried any of these in a multi-agent setup?
No, its on my todo list but when I do get around to it I will most likely write up my observations.