smithbydesign.
All selected work
ServiceNow / Project Workspace + AI

Assistance
before autonomy.

Designing AI for project managers without designing the project manager out of the process.

Organization
ServiceNow
Strategic Portfolio Management
Role
Staff Product Designer
Capabilities
AI Strategy · Product Design
Enterprise UX · Generative AI
Product approach
Context-aware assistance.
Human judgment and control.
Inside the project. Alongside the PM.Project Workspace / AI Insights
Office mockup based on the recorded UI. Select to inspect the original capture.
Generate.Analyze.Recommend.You decide.
01 / The reality

Managing the project
had become a project itself.

Every project brings a moving set of tasks, people, costs, risks, and commitments. The PM has to turn all of it into a clear view of what needs attention.

The project, in contextMany signals. One working picture.
Tasks
Dependencies
Resources
Financials
Risks / RIDAC
Milestones
Status
Project WorkspaceOne clear view.Understand. Prioritize. Decide.

The opportunity wasn’t more information.
It was less interpretation.

02 / The human tension

People wanted AI to help with their jobs.

They weren’t asking AI to take their jobs.

Help meSummarizeAnalyzeGenerateRecommend
Keep me in controlReviewEditDecideApprove

We didn’t need to maximize autonomy.
We needed to earn trust.

03 / My AI approach

Autonomy should be earned,
not assumed.

Project Workspace / Assistive AIHuman review throughout
  1. 01UnderstandRead the context
  2. 02AssistPrepare the work
  3. 03RecommendPropose a direction
  4. 04DelegateOwn a bounded task
  5. 05ActExecute within limits

The first three stages describe this work. Delegation and autonomous action are the next design question.

Start with trust

Prove usefulness before asking people to hand over control.

Reduce cognitive load

Remove interpretation work without creating another interface to manage.

Keep judgment human

The more consequential the decision, the stronger the human checkpoint.

Match autonomy to risk

Reading a project is different from changing its budget, resources, or scope.

04 / AI as a system

AI belongs in the workflow.
Not beside it.

The design starts with the project’s context. Intelligence helps interpret it. Human judgment determines what to do with it.

01 / Project context
TasksDependenciesResourcesFinancialsRisks / RIDACMilestonesStatus
02 / Project intelligenceFrom signals to understanding.
GenerateAnalyzeSurfaceRecommend
03 / Human judgmentYou make the call.
ReviewEditDecideAct
05 / The product experiences

Less groundwork.
More judgment.

Four expressions of the same idea: help the PM move from information to a considered decision.

01 / AI project plan generation

Generate the
starting point.

A charter, statement of work, or project description becomes an initial plan. Tasks, dates, durations, and work breakdown structure give the PM something concrete to review.

From source context to a generated plan ready for review.
Detail / The review step before project creation.

AI removes the blank page.
The PM determines the plan.

  1. Bring project context
  2. Generate initial structure
  3. Review and refine
Office mockup based on the recorded UI. Select to inspect the original capture.
02 / AI-identified risks

Surface what
humans may miss.

A proposed risk includes impact, probability, rationale, and mitigation. The AI Draft state makes the boundary explicit: identifying a risk is not the same as accepting it.

Project WorkspaceRIDAC / AI-identified risks
AI-generated · Check for accuracy

AI-identified risks

Regenerate
Risks start in AI Draft. Accept to move to Pending, or reject to close and hide them.
AI DraftRSK001010

Resource overallocation may delay delivery

Several key resources are significantly overallocated across multiple months, creating a risk of delayed tasks, reduced quality, and missed handoffs.

Impact
1 · High
Probability
High

AI rationale

  • RA0001352 · Software Engineer shows 300% utilization across four months.
  • RA0001354 · Project Manager and RA0001349 · Project Manager show 265.1% average utilization.

Mitigation plan

Rebalance assignments and reduce concurrent work for the most overallocated resources. Shift noncritical tasks to available team members.

Review the evidence before changing the project.
Recreated review detail · Content condensed from the recording.
01 / ExplainShow the evidence.

Rationale links the proposed risk to project records.

02 / ReviewKeep it a draft.

The PM can inspect the impact, probability, and mitigation.

03 / DecideAccept or reject.

Accept moves the risk to Pending. Reject closes and hides it.

Inspect the proposed risks and their supporting project context.

AI can identify the signal.
Judgment determines whether it matters.

03 / Project insights

Don’t make me
find the signal.

Bring schedule, resources, financials, and risk into one explanation of project health. Reduce the work of assembling the picture before making a decision.

TasksResourcesFinancialsRIDAC
Show me what changed.
Project context becomes an overview and recommendations inside AI Insights.

Compress the information.
Keep the context.

04 / AI status reports

Turn understanding
into communication.

01 / Project context
ProgressMilestonesRisksCost
AI synthesizes
the working picture.
02 / Generated draftAI

Project status report

Overall healthRed
AI-generated summary

Resource capacity needs attention. Cost, schedule, and scope remain stable.

Executive summary
03 / Human judgmentThe PM owns
the message.
  • Check the assessment
  • Edit the narrative
  • Decide what to communicate
Ready for human review
Workflow illustration based on the recorded status-report experience. The product clip below shows the original interaction.
Status-report generation, review, and editing in the project workflow.

AI creates the first draft.
The PM owns the message.

06 / The interaction pattern

Different capabilities.
One relationship.

  1. 01ContextThe project supplies the evidence.
  2. 02AIGenerate, analyze, or recommend.
  3. 03RationaleMake the result inspectable.
  4. 04ReviewLet the PM change the proposal.
  5. 05DecisionKeep responsibility visible.
AI / Action

A shorter path can make sense when permission and consequence are understood. Here, the review step was intentional.

That was intentional.
A consistent relationship, across different capabilities.

Users could inspect, modify, accept, reject, and decide through the relevant workflow. The interaction model kept assistance and responsibility together.

07 / What I learned

The biggest lesson wasn’t capability.
It was permission.

  1. 01Competence
  2. 02Transparency
  3. 03Predictability
  4. 04Trust
  5. 05Delegation

Before people delegate meaningful work, AI has to show that it understands the context, behaves predictably, explains itself, and keeps them in control.

08 / My POV · Future-state thinkingDesign philosophy / Not shipped ServiceNow functionality

So where do
agents fit?

Assistance helps someone do the work. Delegation gives a system responsibility for a bounded part of it—with permission, limits, and a return path.

My POV / Future-state thinkingBounded delegation
  1. 01UnderstandRead the context
  2. 02AssistPrepare the work
  3. 03RecommendPropose a direction
  4. 04DelegateOwn a bounded task
  5. 05ActExecute within limits

Delegation needs a defined job, permission to act, and a way to return control.

Help me understand this.Recommend what I should do.Handle this for me.
09 / How I would approach an agent

Define the job.
Then define its limits.

RoleWhat job is it responsible for?

A bounded outcome, with a clear owner and a recognizable finish.

ContextWhat may it understand?

The project records and information relevant to that job.

SkillsWhat can it actually do?

A defined set of actions, with outcomes that can be inspected.

PermissionsWhat may it change?

Explicit access to specific systems and records.

GuardrailsWhere must it stop?

Limits on scope, consequence, and acceptable uncertainty.

EscalationWhen does judgment return?

A clear handoff when the task exceeds its authority or confidence.

10 / Autonomy and consequence

Autonomy should scale
with consequence.

As the consequences grow, so should human control.

Lower consequence

Greater autonomy

  • Gather project updates
  • Summarize activity
  • Prepare meeting briefs
  • Identify overdue work
Medium consequence

Confirmation / review

  • Follow up with task owners
  • Draft mitigation plans
  • Recommend schedule adjustments
Higher consequence

Explicit human approval

  • Move major milestones
  • Change budget assumptions
  • Reallocate major resources
  • Change project scope
11 / Back to the work

AI shouldn’t replace
human judgment.
It should give humans
more time to use it.

Project Workspace / Assistive, context-aware, human-in-the-loop AI.

Project Workspace reinforced my belief that the path toward agentic AI starts with usefulness, transparency, trust, and ultimately permission.

Next story / Human-centered AI

Consumer AI
+ ML