The Tech Docket Daily signal on tech & AI — what the world is searching, and why

Today's Docket

July 19, 2026 · 6 entries · about a 60-second read

The day's tech, heard in order: one long read and 5 quick ones — every entry linked to its source, picked from the 12 stories our tracking scored today.

  1. AI Tools & Apps OpenAI Today's read

    OpenAI walks Codex's context window back to 272K

    Our full story on the quiet rollback from 372K: what shrank, who feels it first, and the one number to check before you blame the model.

    The Tech Docket · full article

  2. AI Tools & Apps Short of the day

    One local library now runs 60+ speech-to-text models

    CJ Pais's ggml-based transcribe.cpp spans 16 ASR model families with Vulkan, Metal and CUDA acceleration — on-device transcription, with Python, JavaScript, Rust and Swift bindings.

    Source: workshop.cjpais.com

  3. Policy & Society Short of the day

    A consultant's blunt case that 'AI mania' is breaking decision-making

    The much-discussed Ludicity essay claims the AI projects its team has observed are failing — and that career risk keeps executives from saying so out loud.

    Source: ludic.mataroa.blog

  4. AI Tools & Apps Claude

    Claude Code now runs on a Rust port of Bun

    Bun's creator says startup is about 10% faster on Linux; Simon Willison checked the shipping binary and found Bun 1.4.0 plus hundreds of Rust source files inside.

    Source: simonwillison.net

  5. Chips & Hardware

    2,500 MIDI recorders later: hardware is as hard as you make it

    Chip Weinberger's Jamcorder retrospective flips the cliché — the physical product proved manageable; the 200,000 lines of software behind it were the hard part.

    Source: chipweinberger.com

  6. How-To & Explainers

    Burning a month of AI tokens in 30 minutes, then fixing it

    Bartosz Kotrys hit his subscription cap researching token costs, then restructured the pipeline to run roughly 10x longer on the same budget — and found invoices rarely match token counters.

    Source: quesma.com

How the docket is made: entries are picked from the 12 stories our tracking scored today across independent source types, then written and checked under the same sourced-claims, no-clickbait rules as our articles — see the editorial policy. The docket is a companion to, never a replacement for, the day's full articles. Previous dockets: July 20, 2026.