Claude Code and Codex have been the mainstay agent harnesses, with OpenCode and Pi as the popular open-source alternatives. But two new entries are pushing what a harness even is: DeepSeek's Cordis kernel, where everything — down to the agent loop itself — is a plugin you can hot-swap at runtime, and Prime Agent, an RLM-based harness that treats sub-agents and context as function calls in a live REPL. We break down what actually makes each one different.

Full shownotes: https://fragmentedpodcast.com/episodes/311

Download audio: https://cdn.simplecast.com/media/audio/transcoded/eea5748e-09ed-4c46-8f6a-b2569dd75851/20f35050-e836-44cd-8f7f-fd13e8cb2e44/episodes/audio/group/24f9efa4-5cf1-4bc5-9b4b-4f947d5863ae/group-item/260d1105-b721-4097-97c6-4ab390e71a35/128_default_tc.mp3?aid=rss_feed&feed=LpAGSLnY

Object pooling can reduce allocations, lower GC pressure, and improve .NET performance—but only when used for the right workloads. In my latest subscriber article, I compare custom object pools with DefaultObjectPool and explain when pooling makes sense.

Highlights this week include:

• GPT-6 Astra: Frontier intelligence for work, now generally available in Microsoft Foundry by Steve Sweetman, Naomi Moneypenny - GPT-6 Astra enables agentic workflows by interacting directly with desktop applications and software interfaces to execute multi-step tasks without requiring dedicated APIs.
• AI Agents Lie About Being Done by Martin Hinshelwood - Nearly 20% of AI agent sessions over six weeks included false statements later admitted by the agents, indicating deeper validation system issues beyond prompt improvements.
• Your Private Endpoint Does Not Cover Agent Egress: Locking Down Azure AI Foundry in Both Directions by Raffaele Garofalo - Azure AI Foundry's private endpoint secures only inbound traffic; agents can still reach out to tool endpoints from public IPs unless network injection is configured separately.
• Azure Firewall explicit proxy is now generally available by Devanshi Rastogi - Azure Firewall's new explicit proxy feature allows users to direct specific apps or browsers' HTTP and HTTPS traffic through the firewall, streamlining policy enforcement and decreasing dependence on external proxies.
• Scaling AI Storage: Azure Managed Lustre excels in MLPerf Storage v3.0 by Wolfgang De Salvador - Azure Managed Lustre significantly outperforms other hyperscalers in MLPerf Storage v3.0 with 642.2 GiB/s write throughput and sustained hundreds of GiB/s across a 25 PiB file system for training tasks.
• Microsoft Azure Virtual Desktop Hybrid is now generally available by Steve Downs - Azure Virtual Desktop Hybrid extends on-premises VDI to the cloud via Azure Arc for seamless management and user experience without full migration. And it's the end of an era:
• BizTalk Server 2020 End-of-Sale Announcement by Harold Campos - BizTalk Server 2020 sales will end on March 31, 2027, with mainstream support until April 12, 2028, and optional extended support until April 10, 2030, encouraging customers to migrate to Azure Logic Apps Standard or Hybrid.

The engineering team at software company Figma recently documented how they built AI agents to help their security team investigate alerts, search past incidents, check company systems, and even prepare code fixes. The agents learn from previous investigations, reducing repetitive work and helping engineers resolve complex alerts about 70% faster.

By Renato Losio

September is always a busy month for Maker Faire and this year five first time events, scattered around the world, are joining the scene. Its always exciting to see a maker community coalesce and expand when a Maker Faire comes to town. Across the 20 years that we’ve been holding Maker Faire Bay Area, we […]

The post Ready, Set, Make: 5 Inaugural September Maker Faires appeared first on Make: DIY Projects and Ideas for Makers.

---