Agentic Surfaces
Surfaces
T7

Representation & Equity

9 surfaces
7.1
Service design

Agency-provided agents (a public option)

Every service entry point offers a free government agent, presented alongside whatever agent the user already has.

City Housing Office

Apply for a housing transfer

An assistant can prepare this application with you. Use the free public one, keep the one you already have, or do it without an assistant. The outcome is assessed the same way.

You can switch assistants at any point without starting over.

The public agent sits in the same list as the commercial ones, at the same weight, rather than behind a link to an alternatives page. Using a government service never requires a commercial account.
7.2
Interaction design

Preserving the person's own voice

Edits that change what the user meant are shown before submit, so the user can keep their own wording.

Your letter, with the assistant’s changes

4 of 4 changes applied

Dear City Housing Office, assistant's wording, Register change: My grandmother and I have been waiting six months for the housing transfer. assistant's wording, Voice / cultural marker change: This delay is unacceptable. She assistant's wording, Register change: is no longer able to manage the stairs, and the office assistant's wording, Substance change: has repeatedly failed to respond substantively.

Underlined text shows where the assistant changed your words.
7.3
Service design

Non-agent fallback channels, kept first-class

The user can choose the non-agent route, and its published processing times hold the agency to its promise that the route is equivalent.

User choosing channel

Apply for a housing transfer

Every route leads to the same assessment. The times below are what each route delivered over the last quarter.

You can switch routes at any point without losing anything.

Choosing a route without an assistant never changes how the application is assessed; the published per-route numbers are what hold the agency to that.

System monitors equivalence
Housing transfer · channel equivalence, Q2
Decision time gap
Non-agent routes +1.6 days vs assistant route
within the 3-day tolerance
Approval gap
No significant difference by channel
within tolerance
Completion rate
Phone route down 6% this quarter
review opened 28 Jun

When the two channels diverge past the set tolerance, a review opens automatically. The numbers shown above come from this monitor.
7.4
Interaction design

Accessibility baseline: WCAG and beyond

The user sets the pace of the conversation and keeps a transcript they can go back over.

City Assistant
What’s happening with my transfer application?
Your transfer application needs one more thing: a letter from your doctor about the stairs. I can help you ask for one. Nothing is sent until you say so.
7.5
Interaction design

Plain language and multilingual intake

The agent states how well it handles the user's language and offers a human interpreter where its command of that language runs out.

Which language should we use?

Your browser is set to Vietnamese, so it’s selected. Change it if that’s wrong. Plain language is the default in every language.

A human interpreter is available in any language, at any time. Request one at any point.

The support level is set by accredited translators and republished as it changes, so it is an audited rating rather than the system’s own estimate of how well it is doing.
7.6
Service design

Co-design with affected communities

An affected-community panel keeps oversight of how the agent behaves once it is live, informed by what users say it got wrong.

User in conversation
Tôi muốn khiếu nại về việc chuyển nhà bị trì hoãn.
I’ll record this as a formal complaint about the delayed housing transfer and send it to the complaints team.
Did this say what you meant?

The check appears whenever City Assistant restates the user’s intent across a language or a change in tone. Those are the moments when a mistranslation does the most damage.

Community panel
Housing assistant · panel findingsJune session
Vietnamese-language sessions: "khiếu nại" (complaint) softened to "feedback" in 14 flagged exchanges
Recommendation: Keep the user’s word “complaint”: it triggers the formal process
Adopted 12 May
Sessions that mention guardianship were sent to the standard flow, which assumes the applicant acts alone
Recommendation: Add a supported-decision route, checked by the disability advocacy group
In review
Elders flagged City Assistant’s follow-up pace as rushed in 9 of 31 reviewed sessions
Recommendation: Default to slower pace when the user sets larger text
Declined 30 Jun· reasons published, REC-2026-014

Panel members come from the affected communities and see logs with identities removed. Every recommendation gets a published answer: adopted, or declined with reasons.
7.7
Interaction design

Whose interests the platform serves

When a third-party agent acts for the user, the surface names who it answers to and offers a switch to the public option.

HomeHelper wants to act for you here

This assistant will prepare your housing transfer application. First, here is who it answers to.

Interest disclosure
Operated byFinlio Pty Ltd
Acting forYou, with contractual obligations to Finlio
How it makes moneyCommission when you take up partnered removalists, lenders, or insurers it suggests
Your dataYour interaction history is used to market Finlio’s other products to you, unless you opt out
Declared to City Digital Services · DEC-2026-0417

Both routes complete the same application. The disclosure is standardized, so any assistant acting here must answer the same four lines.
7.8
Service design

Government-provided agents for high-stakes interactions

For an appeal or dispute, the user is routed to a stronger public agent, with a human escalation available throughout.

What do you need to do?

City AdvocateEnhanced

Challenging a debt notice has serious consequences, so the enhanced assistant is used. It works only for you on this matter, and it will:

  • Retrieve the decision record and the calculation behind the debt
  • Check the figures against your income records, with every source cited
  • Draft the challenge in your words, for your review before anything is lodged

A case officer from the Debt Review Team is assigned when you start. You can hand any step to them, and the assistant hands over automatically if your challenge is refused.

The enhanced level is set by what the task puts at risk. It is never set by who is asking, or by what they could pay for privately.
7.9
Policy design

Digital divide and exclusion risk

The populations an agent channel is likely to fail are named in advance, each with a committed alternative that later uptake reporting checks.

Before launch
Housing assistant channel · exclusion-risk assessment
Published 3 Feb · record ERA-2026-011
Households without reliable connectivity
Risk: Cannot sustain an agent conversation; drop out mid-application
CommittedPhone and in-person routes with identical processing; offline form kept
People without a smartphone or personal device
Risk: Agent channel assumes a private device and persistent identity
CommittedLibrary and service-office kiosks with staff help; no device required to finish
Speakers of languages with limited assistant support
Risk: Confident-but-wrong translation of housing and legal terms
CommittedPublished per-language support levels; funded interpreter route
People under guardianship or with fluctuating capacity
Risk: Flows assume the applicant acts alone and consents in one sitting
CommittedSupported-decision route co-designed with advocacy groups

Naming who the channel will fail is the assessment’s job. A population it doesn’t name is a population nobody is answerable for.

After launch
Quarterly check against ERA-2026-011Q2
No-connectivity households
Phone route holding at 96% of pre-launch completion
No-device applicants
Kiosk completions up 12% since launch
Limited-support languages
Completions down 18% vs pre-launch
Action: Interpreter funding doubled; outreach restarted 20 Jun
Guardianship and supported decisions
Route live, volumes too small to read yet

Uptake is reported for each named cohort, not as a single average.