ASUS ProArt × Berkeley Innovation · In progress

Designing how ProArt creators work alongside AI agents.

ASUS is bringing agentic AI into ProArt’s creative tools. Our team is designing the interface between early-career creators and those agents: how a creator hands off work, stays in control, and keeps building their own judgment.

Role
Project Mentor
Timeline
Fall 2026
Scope
Product Strategy, Human–AI Interaction
Team
Mikayla Acosta, Kylie Chu, Vit Do, Daniel Park
The question

When an agent can carry out the steps for you, which ones should a creator still do themselves?

What we’re doing

Research with 134 survey responses and 20 interviews. Early signal: planning friction rises with AI use, not falls.

Status

In progress. Research report and prototypes go to ASUS in December.

The thread: Agents remove friction, but early-career creators need some of it to grow. The work is deciding which friction to keep.

The challenge

Agents remove friction. Early-career creators need some of it to grow.

An agent doesn’t just answer questions. It takes a goal and carries out the steps: opening tools, moving files, making judgment calls. That’s powerful for production work, and risky for creators still learning what good looks like.

The creator is no longer prompting a tool. They’re delegating to one, and no interface pattern exists yet for handing creative work off and taking it back. Early-career creators build taste through exactly the decisions an agent would skip.

Hardware categories
11
Creative professions
6
Agentic AI tools
5

The question

Remove all friction, and you remove judgment.

How might we keep the productive friction that builds a creator’s judgment, when the whole point of an AI agent is to take friction away?

Research so far

We asked creators where friction helps, and where it just slows them down.

Survey responses
134
Active AI users
95
Interviews
20
Age range
17–60+

Early signals

Planning friction rises with AI use
  • Light AI use33%
  • Moderate50%
  • Heavy AI use80%

More AI didn’t mean less overhead.

Authorship falls with AI use
  • Light AI use67%
  • Moderate0–20%
  • Heavy AI use0–20%

Heavy AI users talk far less about the work being their own.

Exploratory coding of 20 interviews. Not population prevalence.

Where we’re headed

An interface where the creator sets the agent’s role.

Research pointed us away from a smarter agent and toward a clearer relationship with one.

The creator decides

How involved the agent is, task by task: do it, guide me, or hands off.

Advise by default

The agent consults and flags trade-offs. It builds only when asked.

Intent stays visible

The creator’s goal and style travel with the project.

My role

I lead the project, from scope to direction.

I guide the research, synthesis, ideation, and prototyping leads, pick the opportunity we focus on, run meetings to keep ASUS aligned, and ask the questions that narrow the brief.

So far

What I’m learning by leading it.

  1. The brief narrows through questions. My job is less about having answers and more about asking the ones that cut scope.
  2. Not all friction is bad. The same steps an agent would skip are where early-career creators build taste.
  3. Early signals aren’t findings. Twenty coded interviews point a direction; they don’t prove a pattern, so we say that out loud.

Progress

Three sprints in.

The full case study, with concepts, decisions, and prototypes, lands when we deliver to ASUS.

Want to talk about it sooner? Email me ↗