Autonomy levels
Suggest, draft, act with approval or act alone: the right level for each task, and a way for people to change it.
All services
What we design · Agentic UX
Agentic UX for AI-native products.
We design AI-first products where agents do real work: copilots, autonomous agents and human-in-the-loop flows. Founders and product teams bring us in to make AI-native UX that people understand, control and trust.
See it in action
Agentic UX, in 26 seconds.
- 01Set the goalA person describes the outcome they want: reconcile this month’s invoices and flag anything unusual.
- 02Show the planThe agent lays out its plan before acting, and the person sets its autonomy to “Ask first”.
- 03Act with permissionRead-only steps run on their own. Anything with consequences waits for approval.
- 04Explain the reasoningOne invoice doesn’t match its purchase order. The agent shows its sources and a 62% confidence score.
- 05Human in the loopThe person corrects the amount and approves the payment batch. Nothing is paid until they do.
- 06Make it reversibleEvery action lands in an activity log with an Undo, so trust can grow over time.
Why it matters
When the software acts on its own.
Traditional UX assumes the person does every step. Agentic products flip that: an AI agent plans, decides and acts, and the person supervises. The interface has a new job. It has to show intent before action, let people set how much the agent can do on its own, explain what happened, and make every action reversible.
Many teams add a chat box to an existing product and call it AI-first. We design agentic UX from the workflow up, so the agent takes on real work, people stay in control, and trust grows with every use.
What’s included
What we design.
Previews, permissions and undo
Plans shown before the agent acts, scoped permissions, and an undo for everything that matters.
Explainable AI UX
Sources, reasoning and confidence shown where decisions happen, so people know when to trust the output.
Copilots and generative UI
Copilot panels, conversational UI and interfaces generated around the task, not a chat box bolted on.
How it works
Test every assumption, then build.
- MapJobs & workflows
- DefineAgent roles & autonomy
- DesignHuman-in-the-loop UX
- PrototypeCoded, spec-driven
- TestReal users & evals
Work in this area
Proof, not promises.
Healthcare AI · UX discovery
Freed AI
10,000+
clinicians use the AI scribe we explored a browser extension for.
Read the case study
Retail · AI search
Collagerie
AI search
Redefining product discovery with an AI-based search experience.
See all work
EdTech · 0-to-1
FlowMo
$600K
raised from VCs for the in-app LMS we designed from zero to one.
Read the case study
Ways to work with us
Flexible engagement, senior people.
Fractional
Senior product design and leadership inside your team, part-time, for as long as you need it.
Advisory
Focused guidance at the decisions that matter: audits, roadmap reviews and design critiques.
Retainer
Ongoing design and product support with a team that already knows your product.
FAQs
Good questions.
What is agentic UX?
Agentic UX is the design of products where AI agents take actions on a person’s behalf, not just answer questions. It covers how an agent shows its plan, asks for approval, explains what it did, and lets people correct or undo it.
What is the difference between AI-first and AI-native product design?
AI-first means AI is the starting point of the product strategy. AI-native goes further: the product is built around what AI makes possible, so the core workflow would not work without it. We design both: new AI-native products, and AI-first redesigns of products already in market.
How is designing for AI agents different from traditional UX design?
Traditional UX design plans every step a person takes. Designing for AI agents means designing supervision: setting goals, reviewing plans, approving actions, recovering from mistakes and building trust over time.
How do you design for AI mistakes and hallucinations?
We assume they will happen. Interfaces show sources and confidence, keep a person approving anything high-stakes, and make every agent action easy to review and reverse.
Do you work with AI startups at the MVP stage?
Yes. We design founding MVPs for AI-first startups and work with your engineers from day one, often prototyping in code with Claude Code, Cursor and Kiro.
Can you make our existing product AI-native?
Yes. We audit your workflows, find where an agent can take on real work, and redesign the experience around it without throwing away what already works.