Field guide / primary domain
221
sources in this field
Updated September 19, 2026
Current thesis
The shortest path to orientation.
This synthesis records claims and practices from the cited sources; reported outcomes and product capabilities have not been independently verified.
Developer tools standardize on agent-accessible knowledge formats and designer-level aesthetics. Agent Plugins v1.0.0 defines portable package format across Codex, ChatGPT, Cursor, GitHub Copilot, and VS Code. Infrastructure standardizes one-click deployment, real-time token cost visibility, and multi-model dev loops (~$400/month). Vertical slices—building API contract→frontend→services→DB incrementally (100–200 lines at a time)—outperform default horizontal plans since frontier models won't design vertically without explicit steering. Knowledge tooling densifies around Obsidian-as-agent-surface and markdown vaults via MCPs. Terminal emulators redesigned for agentic workflows (Ghostty). Free inference mainstream via NVIDIA. Git-based knowledge systems hit 2.3GB+ walls, forcing SQLite migration. Anthropomorphizing language in AI-generated code review provides audit signals for AI-authored feedback. UI generation now constrains via json-render: Zod schemas guarantee JSON output matches spec, with single definitions targeting 10+ renderers (React, Vue, Svelte, React Native, Next.js, Remotion, React PDF, React Email, Ink, React Three Fiber). Streaming compilation (createSpecStreamCompiler) enables progressive rendering from partial LLM responses. Dynamic prop expressions ($state, $cond, $template, $computed) bind generated specs to app state without imperative code. Pre-built @json-render/shadcn components (36 UI elements) reduce setup cost; devtools provide integrated inspection (spec tree, state editor, action log) via Ctrl/Cmd+Shift+J.
Evidence board
Claims worth carrying forward01A browser-based adversarial testing suite runs massively parallel adversarial checks against each new release, attempting to break it before ship, at a cost described as 'pennies.'
02Parallelizing adversarial/browser-driven test runs across many instances makes exhaustive pre-release stress testing economically viable even for small teams.
03json-render constrains AI-generated UI to a developer-defined catalog of components/actions with Zod schemas, guaranteeing JSON output matches the schema every time rather than hoping the model stays in bounds.
04Single catalog definition targets 10+ renderers from one spec format: React, Vue, Svelte, Solid, React Native, Next.js, TanStack Start, Remotion (video), React PDF, React Email, Ink (terminal), and React Three Fiber (3D/Gaussian splatting).
05createSpecStreamCompiler (in @json-render/core) supports progressive rendering: it processes streaming chunks from an LLM response and returns partial results/patches so UI can render before the full spec arrives.
06Dynamic prop expressions ($state, $cond/$then/$else, $template, $computed) let generated specs bind to app state and branch logic without the AI needing to write imperative code.
07experimental_composeSpec and experimental_createEvaluator (unreleased, source-build only) enable 'Jev' composition — combining/evaluating specs programmatically with a custom catalog; accessible via /playground's Jev (Experimental) mode.
08@json-render/shadcn and @json-render/shadcn-svelte ship 36 pre-built shadcn/ui components ready to drop into a catalog, reducing the setup cost for AI-generated dashboards and apps.
Adjacent fields
Latest evidence
Rafal Wilinski
Rafal Wilinski credits @typesafeai's Jev with building a massively parallel, browser-based adversarial testing suite that stress-tests each release automatically, at very low cost.
Chris Tate
json-render (vercel-labs) is a cross-platform Generative UI framework that constrains AI-generated JSON specs to a predefined component catalog, enabling safe, predictable UI rendering across React, Vue, Svelte, Solid, React Native, video, PDF, email, and terminal targets. It ships streaming compilation, dynamic expression props, state watchers, and devtools, plus experimental spec-composition APIs.
Heinrich
Heinrich launches the early alpha of Ars Umbris, a malleable IDE that treats a knowledge base like a codebase—typed files, composable repos, cross-repo dependencies, and a type engine that gives agents compiler-like feedback on knowledge structure.
Harman
A single developer built Obscura, an open-source Rust headless browser engine positioned as a lightweight, fast, and stealthy alternative to Chrome-based automation for web scraping and AI agents, with drop-in compatibility for Puppeteer/Playwright.
Matt Pocock
Humanlayer's 'show-me' skill (by @dexhorthy) is a markdown-defined toolbox of visual representations—pseudocode, call trees, component trees, file trees, Mermaid diagrams, diffs, and standalone HTML artifacts—used to make code and PR changes easier to understand at a glance.
Trevin Chow
Trevin Chow shares a method for coding agents to attach before/after screenshots and videos to GitHub PRs by uploading directly to GitHub's user-attachments endpoint and embedding the returned URL in the PR body, avoiding committing media files or using an external host.
Vox
A free library converts the design language of 2,000+ top products (Linear, Notion, ElevenLabs, Lamborghini) into DESIGN.md files—structured markdown that coding agents like Codex and Claude Code can read directly to build or align UIs with a chosen product's visual system.
Matt Dailey
Matt Dailey (Ref) outlines a 7-step design process for AI-assisted product design: fixing constraints before jumping to solutions, removing AI-added bloat, iterating in dedicated design tools rather than the live codebase, using component libraries, leveraging preview deploys, stealing proven UX patterns, and deliberately building personal taste.
Daax
A satirical but useful catalog of stock phrases and rhetorical patterns that recur in AI-generated code commentary, commit messages, and technical writing (Claude/Codex-style outputs), useful as a fingerprint for detecting LLM-authored text.
Thinking Machines
Thinking Machines is making its Inkling model freely available on OpenRouter, restricted to agentic harnesses only, for a few weeks to gather real-world agentic behavior data (anonymized) to improve the model.