They publish in public. They think out loud. They prototype the futures the rest of us catch up to. We're directly influenced by all of them.
Staff Research Engineer · GitHub Next · London
The prime example of the kind of design engineer SageOx is becoming. Designer, anthropologist, and self-described "mediocre developer." Previously Lead Design Engineer at Normally, Founding Product Designer at Elicit, Head of Design at HASH. Writes deeply illustrated essays on AI as a new design material, ambient daemons, the dark forest of generative content, and digital gardens. Influences every section of our HARNESS.md.
Notion (previously Senior Researcher, Ink & Switch) · MIT HCI PhD
The clearest current thinker on malleable software and end-user programming with LLMs. Every essay is a working prototype. If you only follow one person on what comes after the IDE, follow Geoffrey.
Thrive Capital (previously Research Engineer, Notion)
Designs UIs over latent spaces, vector embeddings, and continuous parameter manipulation rather than discrete buttons. Stream is the most consistently surprising indie research blog in this space.
- Notes on latent scaffoldingessayInterface as a window onto a model's latent space, not a chat thread.
- StreamnotesContinuous public thinking. Read backwards from today.
Partner, Sutter Hill Ventures · building Intent (prev. GitHub principal research engineer)
Strongest contemporary voice against the chatbot pattern. Designs spatial, gestural, dimensional AI interfaces that treat the model as a material, not a person.
Independent researcher · prev. Khan Academy, Apple
Tools for thought, learning systems, and the design of media that make people smarter. His working notes are themselves a design artifact.
Independent · Datasette · co-creator of Django
The most prolific practical writer on LLM engineering — transparent prompts, tool use, evals, the actual mechanics. If you want to know what works today, this blog is the highest signal-to-noise feed.
- LLM posts archivecontinuousSeveral times a week, deeply hands-on. Subscribe.
llm CLI toolprojectReference for how a thoughtful designer builds for LLM use cases as primitives.
Dynamicland · Oakland
The pre-agentic prophet of direct manipulation, dynamic media, and "inventing on principle." Every modern tools-for-thought person traces back here. Old essays, eternally relevant.
- Magic Ink2006"Most software needs information software, not interaction software." Predates the LLM era by 17 years and feels written for it.
- Up and Down the Ladder of Abstraction2011Interactive parameter exploration. Every Claude Design "knob" descends from this idea.
- Inventing on Principle2012 talkThe most-watched design talk of the decade. Watch it once a year.
applied research lab · the future of software development
The team Maggie Appleton joined for a reason — and the same reason we follow it closely. Operates as an applied research arm of GitHub, prototyping in public the tools that turn into the next generation of how engineers (and the agents that work with them) build software. Copilot Workspace, Copilot Chat patterns, GitHub Spark, and a string of speculative tools have all started here. The closest thing to a public design-engineering studio at our scale of ambition.
- githubnext.comliveThe whole site is the artifact — running prototypes, demo videos, and writeups of every project. Read the blog roll; watch the demos.
- ProjectsongoingA live catalog of what an "applied research" lab actually ships. Use as a benchmark for what proposals/ at SageOx could become.
Game designers — the multiplayer & latency-masking lineage
Cast · team coordination · real-time feel
This is one of the most important and underweighted influences on SageOx. Cast is a multi-player team coordination surface, and game designers have been thinking about cooperation, ambient presence, real-time-feeling-without-being-real-time, and latency-masking interstitials for thirty years. They are the lineage to steal from when designing how a team coalesces and gels through software. Anyone hiring for the Cast surface should weight game-design experience heavily.
- Jenova Chen · Journey, Sky: Children of the Light (thatgamecompany)2012, 2019–The gold standard for designed cooperation — two strangers help each other through a world without chat or voice. The exact emotional register Cast should aim for.
- Hidetaka Miyazaki · Dark Souls, Elden Ring (FromSoftware)2011–Asynchronous multiplayer via breadcrumb hints and message stones — proves a team can feel each other's presence with minimal explicit signal. Cast's ambient presence model owes a lot to this.
- Innersloth · Among Us2018How a tiny social-deduction loop creates intense team cohesion at scale. The "Discussion" and "Vote" phases are masterful coordination UI.
- Riot Games design team · League of Legends, VALORANTongoingTeam dynamics at scale — pings, voice, agent-class balance. Watch their dev vlogs; they're remarkably open about iteration.
- Robin Hunicke · MDA framework, Funomena2004 paper · ongoing"Mechanics → Dynamics → Aesthetics" is the cleanest model for designing emergent emotional experiences. Read the paper; it's six pages.
- Raph Koster · A Theory of Fun, blog2004 book · raphkoster.comWhy a "feature" feels fun. Applies directly to whether a coordination tool feels engaging vs. like extra work.
- Bethesda · Bungie · Valve loading patterns · widely studied2000s–presentLoading-mask techniques: Mass Effect's elevator rides, Half-Life's NPC monologues, Halo's Cortana banter, Bethesda's tip cards. Cast borrows from this: ambient activity hides processing latency so the team feels seamless even when the backend isn't.
Craft cluster — design engineers shipping now
live product work · subscribe to each
A short list of people whose live work is worth tracking. The operators who make their peers want to ship better the next day.
- Rauno Freiberg · rauno.meVercelInteraction prototyper. Every demo is a course in the details no one else notices.
- Emil Kowalski · emilkowal.skiVercelBuilt Sonner and Vaul. The reference for component DX. His five rules are quoted in DESIGN.md §1.
- Karri Saarinen · karrisaarinen.comLinear · co-founderSets the bar for craft at product scale. Linear's whole design language is downstream.
- Soren Iverson · soreniverson.comindependentDaily UI explorations. Publishing a small thing every day compounds.
- Brian Lovin · brianlovin.comCampsite / GitHubLong-running open design notebook. Mobile-first taste.
- Mitchell Hashimoto · mitchellh.comGhostty · co-founder, HashiCorpCoined "Agent = Model + Harness." Builder's perspective on the harness layer named in HARNESS.md.
Most "design canon" lists feel like citation theater — they exist to legitimize the writer, not help the reader. We're keeping this short. These are the older works we still actively reference because the ideas in them inform how we ship today, in an AI-native company, in 2026.
Yale · Graphics Press · 1983–present
Two ideas from Tufte are baked directly into SageOx surfaces in 2026: small multiples (the dashboard pattern that survives the agent era because density wins over single-figure narratives) and sparklines (the word-sized graphic that fits inline next to text, perfectly suited to status-at-the-edge UI). The four-book set rewards rereading.
Honorable mentions
read once · keep on the shelf
Worth knowing exist. Not what we reach for weekly — but each one's central idea shows up indirectly in our DESIGN.md and HARNESS.md. We've intentionally trimmed this list from the usual "design canon" sprawl.
- Don Norman · The Design of Everyday Things1988, rev. 2013Affordances and feedback loops. Universal enough that even agents have absorbed it from training data.
- Stephen Few · perceptualedge.com2004–presentPractical heir to Tufte. Information Dashboard Design is still the reference if you're touching a dashboard, but dashboards are a smaller share of our work now.
- Alberto Cairo · albertocairo.com2012–presentHow Charts Lie is worth handing to a PM or exec. The other two are journalism-flavored data viz.
- Colin Ware · Information Visualization: Perception for Design4th ed. 2020For the day you need to argue with someone about why a specific chart isn't working. Pull the perception research out.