Aerithic

Design and research studio / Sydney

When should your AI act, ask or stay quiet?

A go, change or stop answer
from 10 weeks
For product and R&D teams

We find out with real people before you ship. Aerithic prototypes AI that acts without being asked, tests it in real rooms, and tells you what to do next.

A hand at a window in a dark room, lit amber from outside

The threshold, shown as an illustration, not data

Ask

It checks first. People stay in control, and they notice every time.

Stay quietAct

Services

Start at the stage you are at.

A smooth stone form on creased linen

Frame

From A$7,500

Where should our AI act on its own, and where should it ask?

A hand approaching a dark stone disc

Test

From A$24,000

How much autonomy will our users trust, and where will they take control back?

A wall washed by a strip of warm light

Build

From A$100,000

Can we see it working in real rooms, with real hardware?

All services

Signals / what moved recently

All signals

21 October 2026
Material

A touchscreen sleeve that survives the washing machine

NUS researchers publish a woven display of 54 light-emitting fibres: 96 by 54 pixels at 30 fps, 50 detergent washes and 1,000 bends.

Proof that cloth can carry a display. Low resolution, prototype only, and it still points the wrong way if all it does is show video.

30 September 2026
Civic and consent

The OAIC publishes AI transparency fact sheets

Guidance ahead of the 10 December 2026 start of Australia’s automated decision-making transparency obligations (APP 1.7 to 1.9).

Notice has to be specific and written down before the system runs. Ambient systems need a consent surface that works in a room.

23 September 2026
Device and IO

Meta Ray-Ban Display goes international, with an agent that sees what you see

On sale in the UK and Canada from 23 September and in France, Italy and Germany from 13 October. Muse acts on what the wearer is looking at.

The post-screen era arrives on faces. The announcement covers price, styles and the agent, and says nothing we could find about bystanders.

28 August 2026
Civic and consent

Harvard JOLT Digest: agentic AI deserves a presumption of authorisation

Zefang Wu argues software agents acting for a user should be presumed authorised, with the Perplexity cases as the live test.

We take no view on the law. The design question is who said the agent could do that, and how anyone nearby would know.

Stones and a shell laid out on cloth like specimens, one marked by a pixelated detection patch
Plate 02 / Specimens

Field notes

All essays

Projects

Talk to the studio

Bring us a question. We will tell you where to start.

Book a workshopBring us a questionSee prices