top of page
Irina Wong agency logo
THE TELESCOPE

How might users understand, trust, and get real work done with something that can think for itself, when the user can't see where it keeps its brain?

Map of the relationship of AI tools to analog tools to non-AI digital tools.

"These documents are too long. After reading all this, I'm left with no time to do my real work." The Telescope aims to mitigate this frustration by leveraging large language models and multiple AI agents to do the tedious technical reading for our users, so they can focus their time on winning government bids instead.

SERVICES

  • UX Leadership

  • UX Research

  • UI Design

CLIENTS

  • Internal employees

Impact

My UX leadership in owning the end-to-end user experience of three new products resulted in:

More than 9x time savings in document processing steps, with 6+ hours savings per person per task

95% leadership agreement that they strongly recommend their teams use  tools

What are our users trying to do?

Through first-hand user interviews, we built a durable understanding of who we serve and what they're trying to do: their workflows, their goals, and where The Telescope helps or gets in the way:

Illustration of a user at a laptop with an unhappy facial expression.

I have a Very Complicated Thing I need to do.

Illustration of a user with an upset facial expression, with piles of documents on either side of them, holding two pages in his hands.

Normally, I'm reading a load of documents and data just to be able to start. It takes forever to even be ready to begin my task.

Illustration of a user talking to an open laptop.

Computer, read all the documents and data for me...

Illustration of a user with a happy facial expression, looking at an open laptop.

... and tell me what I need to know so I can just start doing the Very Complicated Thing I need to do!

Demystifying The Telescope:

A simple user experience overlays the complex multi-agent workflows

Demystifying.png

Meet our users' concerns

Because users can't see where The Telescope is keeping its brain, we prioritized designing for trust (which is built interaction-by-interaction) as a first-class outcome of our products. The following are still the realities of AI that are part of the experience, not edge cases to ignore:

TRANSPARENCY AND COMPREHENSION

"How did the agent arrive at its answer... and can I understand it or believe it?"

Feelings: uncertainty, mistrust

The Telescope must present rich output legibly and trustworthily, and make it clear what it knows, where its answers come from, and when a human should stay in the loop.

Illustration of a user at a laptop with a pensive facial expression and a thought bubble about being confused and uncertain about pages.
Transparency 1.png
  • In-line interactive citation badges indicate the sources that The Telescope used to generate its content.

Transparency 2.png
  • When selected, the badges link to a new browser tab, so users can read the complete source, without interrupting their workflow.

LATENCY

"This is taking too long."
Feelings: impatience, mistrust

For the graceful handling of long-running responses, how can we reduce user frustration when the user needs something done quickly, but the computer can't work that fast?

Illustration of a user at a laptop with a clock overhead, with a frustrated facial expression.
Latency 1.png
  • User can understand the time commitment for task before choosing to proceed. Also, they can control their future experience by customizing their visibility preferences.

Latency 2.png
  • Although the Telescope can't give time estimates on task progress because of its probablistic nature, the user can still understand the major milestones towards completion.

CONTROL OVER UNDESIRABLE RESULTS

"I don't like what the computer gave me. What do I do now?"
Feelings: helplessness, frustration, mistrust

AI patterns deliver more probabilistic results, subjective error, and potential failure points in the user experience. How can users identify and recover from things that were not their fault?

Illustration of a user at a laptop with a confused and frustrated facial expression.
Control 1.png
  • User has full override control over agentic results after they review the agent's output.

Control 2.png
  • User can also control the process by giving the agent more specific instructions before it begins.

AI as design process

AI tools -- like Cursor, Claude, Figma Design agents, and Figma Make -- supplement our design-build processes, not replace it. While AI strongly supports research synthesis, ideation, and specific design workflow efficiencies quite admirably, human conversation and literal pencils are still in the middle of our design process. In other words, we are clear on what AI is for, and what it is not:

A map that links together the use cases for talking (the center of the map) with design tools and AI tools.

Reflections

In this era of the youth of AI (and it really is the youth), I'm proud we've approached AI with healthy dose of prudence and temperance, keeping humans closely in the loop while leveraging AI's speed at synthesizing information. Our focus was on presenting AI's rich output legibly and trustworthily, communicating in a way that feels human -- ultimately, in designing for trust as a first-class outcome. 

bottom of page