INTERSECTIONS · Service Design ConferenceHuman in the Loop · V2
The proposition

Everyone gets
a Jarvis.

Customers still get a person.

14-minute conference narrative01
ContextProduct · service · organisation · software
JL
Jensen Loke · System builder

We design the work around the software.

Consumer and corporate digital products, service workflows, teams and software—with the same goal: solve real business problems.

Earlier chaptersDBS · Temasek · Ant Financial · Amazon
NowCEO, Success IT
Operating context20+ years of B2B software and support
CommunityCo-founder · 1,000+ Agentic Builders
Two promises: easy, configurable software · human, accountable support.
linkedin.com/in/jensenloke · agenticbuilders.sg · Success IT company context02
The organisational problemDifferent starting points · different jobs
The uncomfortable reality

AI adoption will never be synchronised.

A few people race ahead.

They experiment, combine tools and invent new workflows faster than formal training can spread.

Most people have a job to do.

They cannot continually learn new models, prompts and specialist coding interfaces.

Waiting for universal AI fluency stalls the organisation.
The distinction is job context and starting point—not ability or ambition.03
The strategyDiscovery → packaging → use → capacity
Change the sequence

Do not make AI fluency
the price of AI leverage.

01 · PIONEERS

Discover

A useful AI workflow inside real work.

02 · BUILDERS

Package

Turn it into a bounded, repeatable agent.

03 · COLLEAGUES

Use + supervise

Apply it inside familiar conversations.

04 · ORGANISATION

Return capacity

For customers, learning and the next improvement.

Using a well-designed agent can be the beginning of reskilling—not the reward for completing it.

Pioneers discover; the organisation productises; colleagues gain usable capability.04
The castPeople are the protagonists · agents are tools
Learn this visual grammar once

Two humans carry the expertise.
Two agents extend their reach.

Frontstage · humans
Human · customer judgment

Iris

Understands customer meaning and owns the response.

Human · product judgment

Jianzhi

Carries product history and business rules.

Backstage · AI agents
AI agent · support structure

Zach

Turns conversation into reviewable support work.

AI agent · technical reach

Insa

Investigates and prepares proposed changes.

HumanAI agentSystemHuman decision
The category, geometry and colour remain consistent throughout V2.05
The service blueprintFrontstage relationship · backstage capability
Where AI belongs in our B2B service

The AI is backstage—not between us and the customer.

Frontstage
Human
Customer
talks with⇄
Iris / Jianzhi
owns→
Outcome
Backstage
AI agents
Zach
hands off→
Insa
prepares→
Next action
Infrastructure
Systems
Customer context
retrieves⋯
Product software
records⋯
Audit trail

Agents power the service. People remain the service relationship.

Customers describe symptoms; people interpret stakes; agents prepare work behind them.06
Case study 01Customer meaning → structured work
Human expertise stays central

Iris stays with the customer.

Human

Iris

Meaning · urgency · relationship

AI agent

Zach

Structure · missing facts · draft

1Iris describes the customer’s symptom naturally.
2Zach prepares customer, product, urgency and evidence.
3Iris corrects and approves before creation.
4Insa can investigate backstage; Iris owns what happens next.
81–115 tickets closed per month across the three-month range supplied by Jensen.
Public-safe reconstruction of Iris asking Zach to prepare a ticket, then reviewing and approving the draft
Public-safe reconstruction · orange callouts are human actions; blue responses are agent work.
Real Success IT workflow · customer and ticket identifiers redacted07
Case study 02Product judgment → technical reach
Business knowledge sets the boundary

Jianzhi says: “This report only.”

Human

Jianzhi

Product history · business rule · verification

AI agent

Insa

Impact check · investigation · proposed change

1Jianzhi defines the behaviour and exact scope.
2Insa opens the configured product workroom and investigates.
3Jensen or an engineer reviews and releases.
4Jianzhi tests whether the business behaviour is correct.
Product expertise gains technical reach—without requiring Jianzhi to navigate a coding interface.
Public-safe reconstruction of Jianzhi defining a narrow report change and Insa confirming the scope
Public-safe reconstruction · scope and legal identifiers redacted.
Real product-workroom pattern · Insa prepares; humans retain release authority08
Early evidenceAdoption—not a productivity leaderboard
The adoption surprise

Our heaviest agent users are not engineers.

Jianzhi · human2,538audited steps performed with InsaBD + support · most active attributable user
Iris · human527audited steps performed with InsaPlus 81–115 tickets closed per month
What it meansBusiness expertisecan gain technical reach before every technical skill is acquired.Engineering adoption remains useful but uneven and generally lower per person.
Audited steps are workflow activity—not completed tasks, features or personal productivity.09
The operating contractCapability and authority are separate
Human in the loop

Agents prepare. Humans decide.

AI agents · blue
Structure incomplete information
Recover customer and product context
Investigate safely
Prepare a reviewable action
Stop when uncertain
CORRECT
APPROVE
RELEASE
Humans · orange
Interpret what the customer means
Correct and approve the record
Decide whether evidence is sufficient
Review, release and test
Own the outcome and relationship

The agent moves the work forward. A person owns the consequence.

Human decisions are visible checkpoints—not hidden assumptions.10
The playbookBuild with pioneers · package for everyone
What I would take back to an organisation

Design the capability—not just the AI interaction.

01

Find the leverage

Observe where early adopters already improve effort or judgment.

02

Package the workflow

Give the agent one role, context, permissions and stop conditions.

03

Meet people at work

Use familiar conversations—not a specialist interface.

04

Keep authority visible

Design correction, approval, testing and escalation into the flow.

05

Reinvest capacity

Return time to customers, learning and the next improvement.

Do not wait for everyone to catch up with AI. Build AI that can meet people where they are—and help carry them forward.

Management’s leverage: build the reusable layer, then improve it when the agent fails.11
End · QuestionsEveryone gets a Jarvis · customers get a person
The closing invitation

Build AI people can use—before they can build it.

Questions?

QR code for the Agentic Builders Collective WhatsApp groupJoin Agentic Buildersagenticbuilders.sg QR code for Jensen Loke's LinkedIn profileConnect with Jensenlinkedin.com/in/jensenloke
Jensen Loke · Success IT · Agentic Builders Collective12
AppendixPlain-language technical decoder
Five terms

What the machinery means in ordinary language.

SuccessGraph

The customer map

Which clients use which products and related companies.
CodeGraph

The software impact map

Which parts of the product a change could affect.
Proposed change

A change awaiting review

Prepared for a human; not yet released to customers.
Optional material following the 12-slide main talk.A1
AppendixSuccessGraph · customer context
System · neutral grey

A map of the customer’s operating context.

  • Customer and related companies
  • Products and subscription state
  • Mapped platform and support route
  • Context can degrade gracefully; Iris still interprets meaning
Public-safe SuccessGraph customer map
Public-safe reconstruction · identifying records redactedA2
AppendixCodeGraph · software impact
System · neutral grey

A map of what a change might affect.

  • Connects screens, services and stored procedures
  • Highlights downstream dependencies
  • Supports investigation; does not grant release authority
  • Humans still review and test the proposed change
Public-safe CodeGraph software impact map
Public-safe reconstruction · internal identifiers redactedA3
AppendixWhat the evidence does and does not show
Evidence standard

We can prove adoption.
We should not overclaim causation.

Supported now

  • Substantial attributable use by Iris and Jianzhi
  • Different workflows for different human expertise
  • Visible human approvals and release controls

Measure before claiming

  • Exact time saved
  • Resolution-time or quality improvement
  • Causal relationship to staffing changes
Audited steps are activity evidence—not productivity or quality scores.A4
1 / 1600:00