Skip to content

Choose references for Humagen's working interface

Use Harvey II for workspace structure, Codex for supervising agent work, and Linear for interaction consistency. That is one reference direction for Humagen: a professional place where projects, people, agents and decisions remain connected. Add Claude and Dust for interaction judgment, Figma for managing complex tool states, and Raycast for frequent actions. These are selected for relevance and documented shipped work, not ranked as the world's best designers. The principles below are attributed paraphrases; the Humagen applications are our interpretation. Research checked 26 October 2026.

Product and attributable people Portfolio or credited work Stated principles and original source What to study for Humagen: interpretation
Claude: Jenny Wen. Credited Claude.ai refresh and Cowork vision; current portfolio describes her leadership in the past tense. Portfolio and shipped work Judge the delivered experience rather than process artifacts; adapt methods; treat craft as consequential. Her essay Product coherence, collaboration and removing accumulated interface friction.
Anthropic: Joel Lewenstein and Nate Parrott. Joel leads design; Nate is credited with Claude Code's VS Code experience and Claude Design's origins. Joel's essays/talks, Nate's interactive portfolio. Current team credits Joel: decide what deserves to exist and make alternatives tangible. Essay. Nate: make ordinary jobs excellent; add specialized UI for a concrete need. First-person interview Decide which agent behaviors need persistent controls, and which work well conversationally.
Codex: Ed Bayes; Andrew Ambrosino. Bayes is publicly credited as design lead; Ambrosino leads desktop app development. Neither is its sole credited UI author. Ed's profile, Figma's design-lead credit, Andrew's design-engineering portfolio Bayes: understandable control, working prototypes, balanced discoverability. Original interview. Ambrosino: quality that makes returning for the next task natural. Original interview clip Parallel work, clear intervention points, contextual review and continuity between sessions.
Dust: Édouard Wautier; Clément Aupiais. Public product-design contributors, with no verified sole-author credit for Pods. Wautier's biography/writing, Aupiais's portfolio/résumé Wautier: prototype behavior before polishing; reuse real components; share previews early. Dust essay. Aupiais: accurate, contextual data displays; choose tables or charts by information need. His earlier Algolia essay Human and agent controls over the same work; trustworthy cost, usage and status displays.
Harvey: Shawn Farsai; Bjørn Rostad. Foundational work and published principles, not individually confirmed authors of the October Harvey II refresh. Farsai's work, design foundations case, Rostad's portfolio Preserve professional familiarity; reduce complexity while retaining oversight; give hierarchy and evidence a purpose. Their joint essay Project context, inspectable outputs, actionable notifications and professional confidence.
Linear: Karri Saarinen. Cofounder and attributable design direction, not sole author of every current feature. Personal writing/work, early Linear case Give AI a purposeful work environment and human review. AI workbench essay. Require features to justify their lasting cost. Feature essay Stable project navigation, useful density and restraint as capabilities grow.
Figma: Rasmus Andersson, historical editor/UI2 work, 2016–2020. Detailed shipped Figma portfolio Improve readability and discoverability; reduce inconsistent patterns; modernize without chasing trends. Original 2019 refresh account Permission-sensitive controls, selected-object states and a consistent vocabulary across complex screens.
Raycast: Roy van Rooijen; James McDonald. Roy is a current product designer; James is specifically credited for the refresh's icons. Roy's site, James's Wireframe portfolio, team Prioritize speed, simplicity and delight; expose contextual actions and shortcuts. Official refresh account Fast actions that remain discoverable without prior keyboard knowledge.

Harvey II changes the workspace, not only its styling

The strongest verified match for “Harvey v2” is Harvey II, announced 18 August 2026. The product places documents, parties, tasks, permissions and history in a matter or project Space, with work moving between lawyers and agents. That makes it particularly relevant to Humagen's shared-project model. (Official introduction)

The separately dated 14 October 2026 UI refresh introduces direct Agents access, a redesigned sidebar, a consolidated Library, easier return to recent work and Spaces, and notifications that lead to work requiring attention. These are useful structural references: find work, understand its context, return to it and act. They are documented product changes, not inferred token measurements. (Official release notes, experience release index)

Do not confuse either release with the 16 January 2026 design-system rebuild. Christopher Sim, Stephanie Yeung and Brian Nelson describe shared design/engineering vocabulary, tokens named for intent, and foundations that support additional themes and surfaces. Their contribution is a systems reference. Farsai and Rostad's separate November 2025 essay is the stronger source for product judgment and human oversight. (System rebuild, product-design principles)

Codex and Dust explain how to design agent work

Codex's February 2026 launch organizes parallel agent threads by project, connects output to contextual review, and returns scheduled work through a review queue. Study the transition from execution to human judgment, including continuity with existing work. Bayes's original interview adds the underlying rationale: people need understandable approvals and mode choices, while the composer must reveal capabilities without overwhelming them. (Codex app launch, Bayes interview)

Dust adds a useful principle: human-agent symmetry. A detailed organizer's account of Wautier's September 2026 discussion describes capabilities that people can control through the interface and agents through textual/API primitives. This is attributed discussion evidence, not a published Dust-wide manifesto. For Humagen, the implication is that people should recognize and understand the same operations that agents perform. (TPC discussion summary)

Wautier's first-person February essay is especially practical: run the interaction with realistic content, errors, empty states and keyboard behavior before settling visual polish. Claude's designers provide a compatible judgment standard: prototype cheaply, understand both user needs and the technical medium, and introduce specialized controls when they solve an actual problem. (Wautier, Parrott)

Five Humagen criteria follow from these references

The following criteria are our synthesis, not quotations or official requirements from these companies.

Criterion A concrete question for evaluating a design Research basis
Organize around work Can someone find the project, its context and the active work without understanding the runtime architecture? Harvey II, Linear's workbench
Make human decisions visible Can they distinguish running work, blocked work and completed output awaiting review, then act in context? Codex launch, Harvey release notes
Reveal control where it matters Are permission, approval and stop choices understandable at the consequential step? Bayes interview, Figma's UI case
Use one interaction vocabulary Do people recognize the same action and outcome across projects, conversations and agent operations? Dust discussion, Harvey's system rebuild
Prove the complete interaction Does the design work with waiting, refusal, cancellation, recovery and real evidence, including exact costs? Wautier's method, Aupiais on data trust

Linear is the strongest additional reference for keeping that system coherent as features accumulate. Figma contributes explicit evidence of simplifying repeated controls while preserving effective structure. Raycast contributes a narrower lesson: shortcuts become useful to more people when the interface makes them visible. These are complementary jobs within the recommended direction. (Saarinen, Figma refresh, Raycast refresh)

Keep attribution and evidence limits visible

Product UI and brand work are separate reference tracks. Geist documents Anthropic's launch-era identity and digital work. Harvey's foundational brand case credits Geist's Eric Mortensen and Tim Kamerer, while Farsai describes himself as the internal stakeholder. These sources support studying visual expression; they do not establish authorship of every current product screen. Dust's public PRODUCT.md is likewise explicitly a marketing/brand brief, not a universal product-UI rulebook. (Geist's Anthropic case, Harvey brand credits, Dust brand brief)

Exact individual Harvey II refresh credits remain unverified. The same applies to sole authorship of Dust Pods and exhaustive Codex screen credits. Several portfolio links are professional profiles rather than visual case studies. Jenny Wen's own current bio uses past tense; Ian Silber is a historical ChatGPT/Codex design reference rather than a current leader, and Rasmus Andersson's case covers UI2 rather than today's Figma UI3. (Jenny's site, Ian's departure statement mirror, Rasmus's dated case)

The research verifies public credits, writing and product announcements through 26 October 2026. It does not constitute a live visual audit, usability comparison or accessibility assessment of Humagen or these products. The shortlist is ready for selecting reference interactions; pixel-level rules would require inspecting the actual screens and their states.