for Claude Code, ChatGPT Work, Codex, Hermes, OpenClaw, Grok Bot and other agents

Put a design method next to your agent

Humane Agentic Design turns a real product question into evidence, a prototype, a tested interface, a next learning step, and—when the work is ready—a final product

Job → evidence → prototype → review → learning
Open source, MIT ↗

The interface is the last thing the agent sees

Most agent workflows begin with a request and end with a screen. Humane makes the space in between visible

It begins with a person in a situation, not with a component. The agent helps name the job, the evidence, and the question that is still open

It makes the cheapest useful answer to that question, before effort hardens into a polished guess

It gives the emerging interface rules that can travel — type, language, layout, imagery, and the decisions behind them

It walks the result as a person would, distinguishes evidence from inference, and says plainly what remains unknown

Then the work leaves a trace: what changed, what improved, and what the next question should be

A method becomes useful when you can use it

Humane is a collection of focused skills. Each one holds a part of the work without turning the whole process into a black box

A live map of the method’s handoffs. Select a point to see where it belongs.

Explore a skill

Hover or select a node to see when to use it, what it hands off, and the exact skill name an agent can run.

Start with the progress, not the feature

Find the change a person is trying to make, the forces holding them back, and the language they already use for the problem.

A clear brief for what deserves to exist

Make images that belong to the work

Turn an approved visual direction into illustrations and photographic prompts that carry the same material, tone, and judgement as the rest of the product.

A reusable image recipe instead of isolated decoration

Let people change their mind before the build

Turn the first idea into a sketch, a clickable prototype, or a small working page—only detailed enough to answer the live question.

A cheap thing to react to, before an expensive thing to maintain

Give language the same care as the interface

Write the words that help someone act: the invitation, the setting, the confirmation, the moment when something goes wrong.

Product language that does not leave people guessing

Make a visual system that can travel

Choose type, colour, spacing, and imagery as a shared vocabulary, then make the rules legible to people and to the code.

A design system with a point of view, not a pile of variables

Meet the system you already have

Fit a decided interface to the real components, tokens, and constraints of a project without quietly changing the product to suit a library.

The existing stack becomes a useful boundary, not a design brief

Look at it through someone else’s eyes

Ask different people to read the same page, message, or proposition independently, then pay attention to where their readings converge and split.

A check on whether the work lands beyond the room that made it

Walk the path, not just the screen

Take a real task and attempt it step by step. The important question is not whether the interface looks coherent, but whether someone can finish what they came to do.

A record of where the experience actually breaks

Borrow a point of view

Let an engineer, a sceptical buyer, an operator, or another stakeholder read the work from their own practical concerns before it has to survive a real meeting.

Constructive friction before the cost of commitment

Check the familiar things people still miss

Inspect a screen against the usability principles that make a system feel understandable, reversible, and calm when something unexpected happens.

A disciplined critique of what the interface asks people to carry

Make the layout survive use

Check the structural details that undo good intentions: dense information, small screens, touch, contrast, URLs, loading states, and the awkward edge cases.

A practical guard against polished but brittle interfaces

Stress-test the decisions

Review a design against familiar usability principles, structural failure modes, contrast, and the details that are easy to miss in a beautiful first pass.

Specific shortcomings, with their severity and their evidence

Show the change a person can feel

Describe the before and after in human terms: the tension, the relief, the new capability—not only the output a product happens to produce.

A transformation story strong enough to test against reality

Build a visual voice, then use it

Explore an identity before committing it to rules; when the direction is clear, make illustration and image work that belongs to the same world.

A coherent visual language from first reference to finished asset

Choose type in the product’s own words

Test actual headlines, long reading, dense tables, symbols, weights, and languages before calling a typeface the right one.

A type decision based on use, not a specimen pangram

Find the next move

When the path is unclear, Humane reads the handoffs between its tools so the next step follows the work instead of a generic playbook.

A method that can explain where it is going and why

Prepare the ground

Check what a project can actually use—its context, sources, tools, tokens, and review surfaces—before asking an agent to promise work it cannot verify.

A working environment made explicit before the first decision

Make interaction feel present

Build a small family of tactile sounds for the moments a person can feel: a panel catching, a choice landing, a task completing—quietly, without turning the interface into a game.

A sound palette that belongs to the product, not a layer of noise

The method grows through use

Humane lives alongside a wider practice of making thoughtful technology together. Follow the work, exchange experiments, and find the people behind the tools at Salient Community.
Visit Salient Community ↗

Put the method next to your agent

For Codex and other agents that read Agent Skills

Add the marketplace once, then install the Humane plugin

Copy this when you want an agent to take care of the setup

Gleb Kalinin seated in a studio beside a camera and lighting equipment

Made by Gleb Kalinin

Humane comes from my product and service design practice, as well as from being an artist and, lately, a developer. My approach takes language, visual judgment, and the lived situation of the person on the other side seriously. I don't think empathy can become algorithmic, but it definitely can be taught. I am looking for a future where default accessibility of apps and services — for humans and agents — will be high by default.

Most of my tools are open source, because a program or a method is only stronger when it can be inspected, challenged, and adapted.

Visit glebkalinin.com ↗

Goal-Driven Design

A practice laboratory by Gleb Kalinin, creator of Humane Agentic Design. It is where judgment stays human: live practice, critique, references, and the discipline to decide what an agent should not decide for you.
Goal-Driven Design sponsors the ongoing development of Humane

The agenda

StartBring a live product question that matters now.

PracticeUse evidence, prototypes, critique, and references to give that question a shape.

ReturnLeave each session with a decision, a trace of why it was made, and a next move.

Bring a real question

Not a hypothetical brief. Start with a product, an audience, and a decision you genuinely need to make. The group gives you enough friction to see what your first answer missed.

Build judgment through practice

Use the toolkit, but do not outsource taste. References, critique, prototypes, and evidence become a working way of deciding.

Join the lab this September

Bring the real work you want to move forward. We will make it clearer, more testable, and more yours.

Join Goal-Driven Design ↗
Supported by ↗