AI product design for products that shouldn’t look like a dashboard

Halaska is an AI product design studio. We design and prototype AI-native products: new ones from zero, and existing ones that a founder can tell look generic but can’t say why. The work is led personally by Chris Halaska, who spent six years at Google designing products for a billion people and now spends most of his time on the interaction problems AI has created and nobody has solved yet.

Halaska keeps pricing simple and publishes its floors so you can assess the fit before the first call.

The problem we get hired for

Most AI products land in one of two defaults. A chat box, because that is what the model exposes. Or a standard SaaS dashboard with AI features added, because that is what the component library ships. Both make the product look like every other one. Founders tell us the same thing in different words: it looks like HubSpot, not like an AI-native product; chat isn’t the right pattern for this; the AI does something impressive and the interface hides it.

That is a design problem, and it has no best practice to copy yet. It is the problem we want.

What AI-native means in a product

Intent first. The user says what they want; the interface works out the steps.
The workspace adapts. Layout, controls and context change with the task instead of sitting in a fixed grid of cards.
Visible intelligence. The user can see what the AI did, where it got it from, and where it needs them.
Human control that feels natural. Approve, edit, undo and redirect without fighting the system.
A distinctive visual language. Typography, motion and density that belong to the product, not to a template.

How we work

Define

A foundation sprint, two or three concepts

A short foundation sprint takes in your product, your model’s real capabilities and the founder’s vision, and ends with two or three interaction concepts, not one. The best AI products we’ve worked on came from the second concept, not the first.

Ongoing product design as an extension of your team. In your Slack, at your rituals, working to a plan developed with your team. This is how most of the studio’s clients work with Halaska.

Prototype in code

A working prototype, not a click-through

We build the core flow as a working prototype, not a Figma click-through. One agentic operations prototype shipped as an HTML build the client’s own customers used in a workshop. A creative tool’s new interaction models were tested as live concepts. Your engineers get something real to build from.

Ship and stay close

Design system, handover, Halaska OS

Design system, handover, and a system your team and your AI tools keep building from after we step back.

Halaska OS

Who it is for

Funded seed to Series B teams with an engineering team in place and a raise, launch or major release inside three months. Usually no product designer in-house yet, or one who is overloaded. If you are pre-seed with no engineers, Dash is the better starting point.

Work

Operations platform (NDA)

Agentic-first data-centre operations

Agents act, operators stay in the loop, dashboards reconfigure by context. A working prototype for a customer demo in six weeks.

Creative tool (NDA)

New interaction models for AI creative tools

A HUD, concentric navigation and nodal workflows that make a powerful tool usable by creators who don’t want to learn software.

Research platform (NDA)

AI knowledge verification

From product concept and personas to a deck that raised, then the MVP.

Pascal

A trading product with a social layer

Brand to Bloomberg-style terminal, $15M+ raised.

Chris-led partnerships from $20,000 a month, with a designer alongside him and two engagements running at a time. MVP design and prototype as a scoped project from $15,000 to $30,000. Full pricing on the pricing page.

Bring us the product that doesn’t have a pattern yet

Tell us what the AI does and what the current interface hides. You’ll know within a call whether this is the kind of problem we take on.

The exact shape and price of a partnership comes out of a conversation, because it depends on what your product needs, not on a menu. What the studio can tell you now: the studio doesn’t discount, the studio doesn’t take equity as payment, and the studio caps the number of clients the studio works with so the work stays senior. If the floor above is outside your budget, Dash might be the right door, and the studio will tell you honestly if it is.

Get in touch

Frequently asked questions

Can't find your answer?

Get in touch

Our AI product looks like every other dashboard. Who can redesign it to feel AI-native?

Halaska. We start by rethinking the interaction model, not the visual layer: how users express intent, see what the AI did, and stay in control. Then we prototype it in code so you can feel the difference before committing to a rebuild.

What is an AI product design studio?

A studio that designs the product around what the AI can do, rather than adding AI features to a conventional interface. The work covers interaction models, trust and recovery states, agent oversight and the visual system, and ends in a working prototype.

Which design agency should I hire for my AI startup?

One that can show you a shipped AI product and explain how it handles wrong answers, user corrections and agent permissions. Halaska’s AI work includes Retell AI and Pascal, plus AI-native products for clients under NDA. Other studios doing credible AI-native work include Late Checkout, The Gradient and Wonder Lab.

Is chat the right interface for our AI product?

Sometimes, rarely as the whole thing. Chat is the right pattern when the task is open-ended and conversational. For most products, the stronger pattern is intent plus an adaptive workspace, with chat as one input among several.

Do you build as well as design?

We prototype the core flows in code as standard, and we build the product with our engineering partners when you need it. Most clients take the prototype and design system to their own team.

How long does it take?

A foundation sprint and first prototype in four to six weeks. A full product in three to four months, in sprints.