Intuit · Platform Design · Expert Workspace

Service Studio Assembled, not built.

A workspace where a new expert service gets fully defined before anyone builds it — so the quality of the definition is what makes the build fast.

Role

Product Designer · Concept, vision & prototype

Team

Platform Design · Expert Workspace

Stage

Design discovery · Leadership pitch

Platform

Web · Agent-assisted service design

Expert Workspace is about to go from two services to many

Expert Workspace is the platform that routes tasks to experts and serves the customer. Two services run on it today — CSM and TTL — and each was added by hand, learned while we built it.

The backlog doesn't wait. One business group alone launches six or more services a year. Hand-assembling every one of them doesn't scale, and every shortcut taken on one service becomes the template for the next.

The service gets defined. How it runs doesn't.

A new service gets defined in business and operational terms, and stops there. What design draws, what engineering carries, what legal and operations have to sign — all of it gets settled downstream, by whoever hits the gap first. Not carelessness: there was never one artifact that asked for the rest of it.

  • No source of truth Every craft holds a different version, so the definition gets re-agreed in a room — then drifts before the next one.
  • Design redraws A use case lands that the approved flow can't hold, so the design moves under everyone.
  • Development rebuilds Architecture was committed to the old shape; absorbing the change means tearing up work.
  • Launch ships a v0 Scope gets cut to hit the date. Customers get the compromised version — and we rebuild it anyway.
The insight

Defining longer doesn't launch later. It launches once. Define grows by exactly what rework gives back — the date doesn't move, what ships on it does.

Prototype walkthrough · library → intake → blueprint → preview → validate → build → publish

From a bespoke build to an assembled service

Service Studio moves the work upstream. Design, standards and engineering happen once, at the platform level — not once per service. Each new service is the same parts, combined a different way.

From

Bespoke build

Every service a different shape, drawn from scratch. A deck, a Jira epic and a meeting recording stand in for the definition.

To

Assembled service

One structured definition that people can't misread — and that an agent can act on. Same artifact, both audiences.

01

It already knows the platform

Existing services, platform capabilities, journey structure, standards, agent skills and shipped bundles are loaded in. Drafting a service instead of discovering one.

02

Definition and output settle together

Generating is how you find out the definition is wrong. Nothing publishes until it holds — and every gap closed is inherited by the next service.

03

The Studio preps, people decide

A builder agent drafts, generates and asks. Product, Design, Development, Operations and Legal review and approve — instead of meeting to align.

Discuss, draft, preview, validate, build, publish

I prototyped the end-to-end flow to make the idea concrete for leadership — following one service, TurboTax Full Service FY27, from a rough brief to a governed publish.

Service library listing live and draft services with their stage
01 · Library — every service on the platform, live or in draft
New service dialog with a written brief and attached blueprint and PRD sources
02 · New service — describe it in whatever form it exists, attach sources
Conversational intake where the builder agent raises gaps and pushes back on decisions
03 · Discuss — the agent surfaces gaps and pushes back before drafting
Catalog check flagging overlapping services before the draft is generated
04 · Catalog check — overlaps named as conflicts, not silent duplicates
AI-drafted service blueprint with personas down the side and milestones across the top
05 · Blueprint canvas — personas down the side, milestones across, the line of interaction in between. The spine every other piece assumes.
Step specification panel defining owner, completion criteria, outcomes and escalation
06 · Step spec — who owns it, what done means, what escalates
Expert Workspace preview rendering the drafted service as an expert would see it
07 · Preview — the service rendered in Expert Workspace before it exists
Validation results with passed checks and open gaps assigned to owners
08 · Validate — every open gap carries the name of who closes it
Service bundle graph mapping steps to agents, skills and tools
09 · Build — the bundle graph: steps → agents → skills → tools
Test run of the bundle showing execution progress and audit log
10 · Test run — the bundle executes against a live harness
Governed publish screen showing readiness and outstanding approvals
11 · Publish — gated on readiness and named approvals

It removes the catch-up, not the judgment

The readiness view goes out before the meeting, so the meeting starts at the decisions. Each craft answers one question:

  • ProductIs this the service?
  • DesignDoes this flow hold end to end?
  • DevelopmentDo the tools it asks for exist?
  • OperationsCan experts actually run it — at peak?
  • Legal & RiskAre we allowed to — and can we roll it back?

Six pieces, one spine

I broke the Studio into six shippable pieces so leadership could make the sequencing call. The blueprint canvas is the spine the other five assume.

A

Conversational intake

Start from a brief, a deck or a PRD — the agent asks for the rest.

B

Service catalog

Every service already running, checked before anything new is drafted.

C

Blueprint canvas

The structured definition — personas, milestones, steps and specs.

D

Workspace preview

See the service as experts and customers will, before it's built.

E

Readiness & validation

Covered, open or blocked — per craft, with an owner on every gap.

F

Governed publish

Nothing ships until the definition holds and approvals are in.

2→n

Services on the platform

6+

New services a year, one group alone

4+

Review rounds per design cycle today

6

Shippable pieces, one spine

The next AI leverage isn't inside the crafts. It's upstream.

Agents accelerate execution against a clear plan. They can't decide what the plan is — and that's the part of our own lifecycle nobody had designed. A definition an agent can act on is one people can't misread.

The measurement gap is the same story: every lifecycle metric starts the clock at "In Progress," and everything the Studio touches happens before that. So part of the proposal is instrumenting time-to-launch, design churn, post-launch rework and quality variance across experts — you can't hold a team to a span nobody times.

Make it the way the platform adds a service

Clarity up front is the whole proposal: fewer meetings to align, a build nothing moves under, and a service the next one inherits. The ask to leadership — start with the next service.

Working material from design discovery, not a committed roadmap. The prototype follows a worked example rather than a live service.