AI Engineer

Agents, codebases, and teams — Aditya Khandelwal, Amazon AGI Lab

Taking Reinforcement Learning Cross Datacenter — Nan Jiang, Modal

RL post-training efficiency hinges on decoupling the trainer from the rollout fleet to overcome the constraints of tightly coupled RDMA clusters. By shifting from full checkpoint synchronization to sparse, bit-level weight diffs, rollout workers can operate on elastic, distributed compute across global regions. This ap...

Multiplayer agentic engineering — Arjun Singh, Superconductor

Multiplayer agentic engineering focuses on integrating AI agents into team workflows to enhance productivity and collaboration. To achieve this, teams should adopt a model-agnostic approach, allowing for seamless switching between models based on performance and cost data specific to their own codebase. By centralizing...

Guide, Verify, Solve — Anirban Chatterjee, Sonar

AI-driven software development requires a shift from mere experimentation to rigorous engineering to ensure code reliability, security, and maintainability. While AI tools boost initial productivity, they often introduce "verification debt"—a persistent accumulation of code quality issues and security vulnerabilities—d...

Velocity Sickness: What Happens When Your Whole Team Gets 10x Faster — Matt Dailey, Ref.

Velocity sickness—the stress caused by rapid, AI-driven output that lacks genuine impact—stems from outdated workflows that prioritize implementation over strategic decision-making. As AI agents handle coding tasks, engineering teams must pivot from isolated, ephemeral chat-based environments to shared, durable documen...

Evolution of agentic surfaces — Gagan Bhat & Isabella Kai He, Anthropic

Claude Managed Agents provides a production-grade infrastructure designed to bridge the gap between rapidly evolving AI model capabilities and the rigid, often stale, harnesses used to deploy them. By decoupling the agent’s "brain"—the reasoning loop—from its "hands"—the tool execution environment—this architecture imp...

Benchmarking Coding Agents on New vs Legacy Codebases — Denys Linkov, Wisedocs

Refactoring a legacy AI pipeline from multiple repositories into a single monorepo significantly enhances shipping velocity and developer engagement. At Wisedocs, this six-month initiative addressed critical issues, including slow performance and unmaintainable code, ultimately enabling the team to support larger files...

The New Primitives: Building AI Native Software — Kwindla Kramer, Daily

The evolution of digital computing, from Vannevar Bush’s 1945 vision in "As We May Think" to the current era of AI agents, reveals a consistent trajectory toward more intuitive, multimodal human-computer interaction. Just as the transition from static web pages to dynamic mobile applications redefined software utility,...

Open Source Is Dead. Long Live Open Source. — Saoud Rizwan, Cline

The open-source community faces a critical decline as AI-driven automation fosters skepticism, security risks, and a flood of low-quality contributions. While closed-source AI labs currently subsidize high inference costs to lock developers into their ecosystems, this strategy is unsustainable. Businesses are increasin...

Anthropic's CCA Exam as a Field-Guide for Agentic Engineering — Frank Coyle, UC Berkeley

Realtime multiplayer, automation, and you! — Idan Gazit, GitHub

Always-on agents run production without the on-call tax — Justin Smith, Resolve AI

Codex, Behind the Harness — Dominik Kundel, OpenAI

The Codex harness serves as an open-source framework for building agentic AI, utilizing the app server protocol for UI communication and the responses API for LLM inference. Efficient context construction remains critical, achieved through deferred tool loading and capped skill lists to manage token budgets and prevent...

Local Models: Trust, Control, Optimization — Carter Abdallah, NVIDIA

Compression at the Edge — NVIDIA, Unsloth, HuggingFace, Ollama

The State of Model Routing — NVIDIA, Cognition, OpenRouter

Model routing optimizes AI production by dynamically delegating tasks to the most appropriate model based on complexity, cost, and domain specificity. Rather than relying on a single frontier model, developers achieve higher performance and cost efficiency by using a primary orchestrator to manage sub-agents or "sideki...

Gadgets: Personal app vibe coding that is actually safe — Kenton Varda, Cloudflare

Building Turbopuffer: Gergely Orosz (@pragmaticengineer ) × Simon Eskildsen (CEO)

MCP Apps: Extending the Frontier — Ido Salomon & Liad Yosef

MCP Apps revolutionize the agentic web by enabling services to transmit interactive, branded UI components directly into AI chat interfaces, replacing inefficient text-based interactions. By leveraging the Model Context Protocol, this framework allows applications like Shopify or PostHog to deliver functional widgets t...

MCP Tasks (async): Why Aren't Any Agents Supporting Them? — Cornelia Davis, Temporal

MCP Tasks represent a critical evolution in agentic workflows by enabling long-running, asynchronous processes that survive infrastructure failures and human-in-the-loop delays. The current lack of widespread client support stems from the high complexity of implementing durable state management, as demonstrated by the ...