Writing

• Exomem vs mem0, Letta, Zep, cognee, and Basic Memory: an honest comparison of memory for AI agents

  An honest comparison of memory for AI agents — Exomem vs mem0, Letta, Zep, cognee, and Basic Memory — on hosting, data ownership, storage format, retrieval, MCP support, license, and price.

  July 4, 2026

• The complete guide to setting up a new Windows PC

  A practical guide to setting up a new Windows PC: reinstall your apps, bring your settings, handle the Store gap, and skip the paid migration tools.

  July 4, 2026

• Share your whole app setup as one file (yes, even MSI Afterburner)

  Your Endstate setup is a portable file, so you can hand someone your exact app config: OBS scenes, MSI Afterburner profiles, your dev setup. Here's how.

  July 4, 2026

• A free, open-source alternative to EaseUS Todo PCTrans, Zinstall, and Laplink PCmover

  Endstate is a free, open-source PC migration tool for Windows, compared to EaseUS Todo PCTrans, Zinstall, and Laplink PCmover on price, method, and trust.

  July 3, 2026

• How to reinstall all your apps with winget (and what it misses)

  winget can export and reinstall your Windows apps from one file. Here's exactly how, plus the apps and settings it quietly leaves behind.

  July 3, 2026

• How to set up a new Windows PC in minutes, not a weekend

  Set up a new Windows PC fast: capture your apps and settings once, then restore them on any fresh install in minutes instead of losing an evening.

  July 3, 2026

• Why winget export skips your Microsoft Store apps

  winget export leaves your Microsoft Store apps out of the file. Here's why it happens, how to see what's missing, and how to capture everything.

  July 3, 2026

• Lazy loading vs. instruction-following

  Skill-triggered context loading reliably outperforms bulk-loaded AGENTS.md. The mechanism behind the benchmark result — two cognitive operations, one binding layer.

  May 24, 2026

• Governance as compression: contracts and skills for AI-augmented codebases

  The pattern I shipped to fix AI-augmented development correction loops, and the controlled three-condition benchmark that revealed which layer does the work.

  May 20, 2026