Chandler Brown
Technical Notes on Data, AI, and Product Systems
Implementation-first writing on architecture decisions, reliability tradeoffs, and lessons from shipping real systems.
Latest Writing
View full archiveaws-is-building-the-control-plane-for-agents
title: AWS Is Building the Control Plane for Agents date: April 14, 2026 tags: [ai agents, aws, bedrock, infrastructure, developer tools] image: /og image.
ChatGPT Workspace Agents Turn Repeatable Work Into Shared Infrastructure
OpenAI's workspace agents move ChatGPT from one-off help to shared, schedulable workflows with permissions, app access, and Slack integration.
Cloudflare Is Building the Control Plane for Agents
Cloudflare's Agent Week shows production agents need memory, identity, private networking, and governance before they need more prompt tricks.
Anthropic's Automated Weak-to-Strong Researcher Shows Both the Promise and Limits of AI Research Agents
Anthropic's latest alignment study shows that agentic research can close a benchmark gap fast, but only inside a tightly controlled sandbox with real failure modes.
Codex Is Becoming the Agent Workspace
OpenAI's Codex update pushes the product past coding help and toward a durable workspace for agentic work across apps, memory, and long-running tasks.
OpenAI Is Standardizing the Agent Runtime
OpenAI's latest Agents SDK and Codex updates move agents from chat prompts toward a real runtime with sandboxes, memory, browser control, and shared tools.
GitHub Is Turning Security Into a Workflow Layer
GitHub's latest updates collapse security, code quality, and remediation into one workflow, with OIDC, issue tracking, and Copilot-driven triage.
Anthropic's Emotion Concepts Are a Safety Signal
Anthropic's emotion-concept paper is useful less as a sentience claim than as evidence that internal model states can predict and shape risky behavior.
Gemma 4 Pushes Google's Open Model Strategy Into Real Deployment Territory
Gemma 4 matters because Google is no longer treating open models as a single product tier. It now spans phones, laptops, workstations, and agentic workflows.
Cursor's Bugbot Is a Preview of Agentic Code Review
Cursor's Bugbot learned rules and MCP support show code review moving from static checks toward persistent, tool-using systems that improve from feedback.
Project Glasswing Shows Where Frontier Security Is Going
Anthropic's Project Glasswing shows frontier models moving into defensive security work, while also making exploit generation cheaper and faster for attackers.
OpenAI's Safety Fellowship Is a Map of Open Problems
OpenAI's Safety Fellowship is less a branding move than a list of safety and alignment problems the company still wants outside help to solve.
TimesFM 2.5: A Practical Read for Forecasting Teams
TimesFM 2.5 is one of the most credible open forecasting foundation models right now, especially for fast multi-series experiments with long context and quantile output.
Axios Supply-Chain Attack: What Actually Happened
In late March 2026, two malicious Axios versions were published to npm and weaponized install-time scripts through a fake dependency, creating a short but high-impact supply-chain window.