product engineering

a toolkit of skills, agents, workflows, resources, and apps for building products.

philosophy

building a product is a context-decision loop. context informs a decision, and that decision generates new context for the next one. the loop never stops turning.

the loop is powered by work, which generates artifacts — research, conversations, designs, tradeoffs, documentation, feedback. they might live across different tools, different teams, and different moments. when you look for the full picture behind a decision, you find fragments. not because anyone organized them wrong, but because there just wasn't something to connect them. that means you either spend time rebuilding context that already existed, or move forward without it.

the loop keeps turning, but each turn has less to work with. this is context degradation. the further you get from a decision, the harder it is to trace why it was made. eventually you stop looking and start assuming.

product engineering exists to reduce that degradation — a way to organize work that aims to keep context connected and findable, so the next decision builds on the last, not from scratch.

the build loop

the build loop is a way to organize work — diverging and converging steps that move from open exploration to focused execution, then package everything for the next turn.

context
explore*
shape*
ship*
compound*
decision
*context-decision loop within each step

explore

diverge. go wide across the problem space. play with possibilities and discover what you don't know.

shape

converge. mold what you explored into something concrete enough to build. narrow from idea to plan.

ship

execute. execute with the living context of why this matters, who it's for, what success looks like, and how to do it.

compound

package. close the loop. package what you shipped, learned, and heard so the next cycle starts with more context, not less. this is compounding in motion.

who this is for

builders who blur the lines between product, design, and engineering — and want a composable system focused on reducing context degradation.

\u2190 home h

awesome

an opinionated list of tools and resources we find worth knowing about.

this is not meant to be exhaustive. every entry is something we've given a fair chance and have first-hand experience with. if it's not on the list, it's because we haven't used it long enough to have an opinion — not necessarily because it isn't good.

orchestration interfaces

tools for running and coordinating multiple coding agents (claude code, codex, opencode, etc.) in parallel — usually with isolated workspaces, unified review, and some opinion about how humans stay in the loop.

Conductor

Mac app for running multiple coding agents (Claude Code, Codex) in isolated workspaces, with a unified interface for monitoring progress and reviewing diffs. The “calmest” of the bunch — feels like a PM dashboard for your agents.

Superset

Desktop code editor built around parallel agents in isolated git worktrees. Leans more “editor with agents inside” than “agent dashboard,” which is nice if you still want to drop into code yourself.

Subspace

Mac app that consolidates terminals, editors, browsers, and docs into one keyboard-first workspace, with persistent memory across agent sessions. Most opinionated about being your single working surface.

Paseo

Desktop, web, and mobile clients that run coding agents as local processes with a unified cross-device interface. Distinguishes itself by not being Mac-only and by following you off the desk.

Maestri

Native macOS infinite canvas where every terminal is a node. Drag agents around, wire them together, let them talk via PTY. The most spatial take on multi-agent work.

design tools

tools sitting at the design/code boundary — prompts to designs, designs to code, or visual surfaces for directing what gets built. plus the workhorses everyone already uses.

Figma

The default collaborative design tool. Worth a mention here specifically for Figma Make, their prompt-to-code surface that turns design intent into working prototypes from inside the same file.

Storybook

Component workshop for building, documenting, and testing UI in isolation. Still the cleanest answer to “where does the design system actually live?” once a project gets non-trivial.

Claude Design

Anthropic Labs product for collaborating with Claude on prototypes, slides, one-pagers, and UIs. Inline comments, live adjustment knobs, and a clean handoff bundle to Claude Code when a design is ready to ship.

Decode

“Whiteboard for coding agents.” Connects agents over MCP and lets you direct them with diagrams, sketches, and live previews of every screen at once — like reviewing a Figma file, except every frame is your real running app.

Pencil

Canvas-based design tool that converts visual designs into code, aimed at compressing the design-to-dev handoff into a single surface.

Lovable

AI app builder focused on rapid prototyping. Describe what you want, get a working full-stack app deployed end-to-end — no infra setup, no scaffolding. Press a button, have a page on the internet.

Open Design

Open source Claude Code alternative — a coding agent CLI you can self-host and extend. Same shape as Claude Code, but the implementation is yours to fork.

plugins

plugins for coding agents (primarily claude code) that bundle skills, commands, and agents into a coherent working methodology rather than ad-hoc prompts.

Superpowers

Jesse Vincent's agentic skills framework and software-development methodology. A composable set of skills (brainstorming, TDD, debugging, plan-writing, verification-before-completion, etc.) plus the meta-skill that makes the agent actually use them. Cross-platform (Claude Code, Codex, Gemini, Cursor, Copilot, OpenCode). The opinionated default for “how should my agent work.”

Compound Engineering

Every's plugin built around the idea that ~80% of the value is in planning and review, ~20% in execution. Ships ce-brainstorm, ce-plan, ce-work, ce-code-review, ce-debug, ce-compound, and a roster of persona-based reviewers. compounding in plugin form.

terminals

the actual surfaces where you (or an agent) type, run, and inspect things. most orchestration interfaces above ship their own embedded terminals — these are what you reach for outside of one, or what you wish those embedded terminals were when they fall short.

VS Code

The default editor for most of the world, with a perfectly serviceable integrated terminal. Worth listing because it's still where the largest extension ecosystem and the most agent integrations land first.

Zed

Minimal Rust-built editor crafted for speed and collaboration with humans and AI. The fastest-feeling editor I've used, with first-class multiplayer and a clean built-in terminal.

Ghostty

Mitchell Hashimoto's GPU-accelerated, native, zero-config terminal. The “just a great terminal” choice — no AI, no platform, no opinions beyond “be fast and correct.”

Warp

AI-native terminal and dev platform. Inputs are blocks instead of a stream, and Claude/Codex agents are first-class citizens rather than something you bolt on.

cmux

Ghostty-based macOS terminal with vertical tabs, agent-aware notifications (OSC 9/99/777), and a CLI/socket API for scripting workspaces, panes, and a built-in browser. Purpose-built for babysitting many coding agents at once.

daily drivers

apps that live in our dock — ai clients, capture tools, and small utilities that we'd notice immediately if they weren't there.

Claude Desktop

Anthropic's native desktop client for Claude. File uploads, Projects, MCP support, and a lot less friction than the browser tab equivalent.

ChatGPT

OpenAI's native desktop client. Voice mode, system-wide hotkey, and tight OS integration on macOS and Windows.

Codex

OpenAI's coding agent — available as a CLI, a web app, and increasingly embedded across their surfaces. The Claude Code counterpart from the OpenAI side.

Wispr Flow

Voice-to-text app that converts speech into polished, edited prose across any app on Mac, Windows, iPhone, and Android. Genuinely good enough to replace typing for most prose — once it clicks, going back feels slow.

Monologue

Every's voice dictation app. Same family as Wispr Flow — effortless voice-to-text aimed at getting thoughts down faster than you can type them.

Screen Studio

Mac screen recording app that produces polished, animated videos out of the box: auto-zoom on the cursor, smooth transitions, clean exports. Default choice for product demos and PR demo reels.

Amphetamine

Free Mac utility that keeps your Mac awake. Sounds trivial; isn't. Long-running agents, recordings, builds, downloads — anything that breaks when the Mac sleeps benefits.

platforms & services

the hosted layer — where code runs, data lives, jobs execute, and behavior gets measured.

Cloudflare

Edge compute, storage, and networking under one roof: Workers, R2, D1, KV, Durable Objects, Queues, Pages. Hard to beat on price and global latency for greenfield work.

GitHub

Obviously. Worth calling out specifically for the trio that does most of the heavy lifting: repos (still the canonical home for code), Issues (lightweight task tracking that doubles as a public knowledge base), and Projects (a surprisingly capable kanban/spreadsheet view layered on top of those issues).

Turso

Edge-distributed SQLite built on libSQL. Cheap, fast, embeddable, and gives you database-per-tenant economics that don't make sense with traditional Postgres.

Electric

Real-time partial sync from Postgres to your local app. The cleanest path to local-first apps without rewriting your backend.

Temporal

Durable execution engine that turns long-running, retryable workflows into ordinary code. The grown-up answer to “we need a queue, retries, scheduling, and saga handling,” and not much else.

Inngest

Event-driven, durable workflow platform with a developer-first DX. Write functions in TypeScript and get retries, scheduling, fan-out, and step durability for free — a lighter counterpart to Temporal.

OpenRouter

Unified API for accessing 100+ LLMs (OpenAI, Anthropic, Google, Meta, etc.) through a single endpoint, with automatic failover and one bill. The pragmatic answer to “I want to try N models without N integrations.”

PostHog

Open-source product analytics with session replay, feature flags, experiments, and surveys in one suite. Self-host if you want to keep the data; cloud if you don't.

frontend & libraries

code that ships in the app itself — frameworks for rendering, libraries for the hard ui bits.

Astro

Content-first web framework with an islands architecture that ships zero JS by default. Best-in-class for marketing sites, docs, and content-heavy apps that occasionally need interactivity.

TanStack

Tanner Linsley's collection of headless, type-safe libraries: Query, Router, Table, Form, Virtual, Start, and more. Tons of stuff worth a shoutout — start here and explore.

Monaco Editor

The same code editor that powers VS Code, packaged for embedding in a browser. Default choice when you need a real code editor inside a web app instead of a glorified textarea.

React Flow

Library for building node-based editors, diagrams, and flow UIs in React. Handles pan/zoom/drag/connect so you can focus on your domain instead of reinventing graph interaction.

auth & dx

the supporting cast — auth, authorization, and release tooling.

better-auth

TypeScript-first, framework-agnostic auth library. Sessions, OAuth, MFA, orgs, plugins — all with sane defaults and a much better DX than rolling your own or wiring up a hosted provider.

Permit.io

Authorization-as-a-service. Externalize permissions logic (RBAC/ABAC/ReBAC) with a policy editor and SDKs, instead of growing a tangled if (user.role === ...) mess in your codebase.

Changesets

Monorepo versioning and publishing tool. PR-based changelog entries, semver coordination, automated publish workflow. The standard for any monorepo with multiple packages.