Blog

Notes on Building and Shipping

Practical writeups covering implementation details, product decisions, and tools that matter in real projects.

Page 1 of 4

  1. aws-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.

  2. 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.

    ai-agentsopenaienterprise-softwareautomation
  3. 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.

    ai-agentscloudflareinfrastructuresecurity
  4. 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.

    ai-safetyanthropicagentsresearch
  5. 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.

    openaiai-agentsdeveloper-toolscodex
  6. 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.

    ai-agentsopenaideveloper-toolsinfrastructure
  7. 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.

    githubsecuritysupply-chain-securitydeveloper-tools
  8. 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.

    ai-safetyinterpretabilityanthropicllms
  9. 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.

    googlegemmaopen-modelslocal-ai
  10. 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.

    ai-codingcursorcode-reviewdeveloper-tools