CURRENTLY BUILDING AN AI OBSERVABILITY DASHBOARD

Beck Kanno

Ledger • Active prototype • Ongoing 2026

Role
Product DesignerFrontend
Timeline
2026Ongoing
Team
Independent prototype
Skills
Product DesignUX ArchitectureTypeScript

Ledger is not just an AI dashboard. It is an operational intelligence layer for AI-native work.

AI work now moves through providers, IDEs, browser tools, APIs, agents, and product surfaces. Ledger explores how teams make that activity observable without pretending the prototype is a finished governance product.

Built for

  • Product teams tracking AI-assisted work
  • Engineering leads monitoring cost, retries, and provider behavior
  • Founders and operators comparing usage across projects
  • Future AI operations stakeholders responsible for governance paths
The overview connects usage, provider spend, sessions, outputs, and summary signals before deeper inspection.

The shift is from prompt analytics to workflow intelligence.

Observation

What AI Usage Looks Like Today

Provider consoles expose useful slices of activity. They rarely connect the behavior around the work.

Provider consoles expose slices of usage, but not cross-tool workflow behavior.

The missing questions are operational: which sessions became retry-heavy, where model switching happened, which projects accumulated cost, and where provider data became stale.

Workflow Intelligence

The System Ledger Had to Define

The product problem was object definition: what should hold the behavior, which states matter, and what a stakeholder can do next.

Providers
Sessions
Projects
Signals
Governance
Interventions

Workflow model

From prompt event to operational record

  1. 1

    Request

    A bounded unit of AI-assisted work starts in a provider, IDE, browser tool, API, or agent.

  2. 2

    Retry

    The same intent is attempted again after a weak output, failed generation, or unresolved path.

  3. 3

    Model Switch

    The work moves between models or providers as the user searches for a better result.

  4. 4

    Signal

    Cost, latency, stale sync, or repeated attempts make the session operationally important.

  5. 5

    Resolution

    Provider, project, status, cost, output, and timing resolve into one session record.

Workflow behavior becomes readable when it resolves into a session record.

Upcoming artifact

Design pending

Product model diagram

Why it matters
The system model needs one artifact that explains Ledger as more than a set of pages.
What it should prove
Providers feed sessions, sessions attach to projects, and signals create governance or intervention paths.
Expected content
Providers -> Sessions -> Projects -> Signals -> Governance / Interventions.
Product model diagram

Providers

Providers as Health Surfaces

A provider is treated as infrastructure: connection status, permissions, sync coverage, ingestion state, and recent events.

Provider health is treated as infrastructure, not settings.

OpenAI is connected in the active prototype. Other providers are in-progress integrations, so the case study keeps coverage honest.

Core Product Decision

Sessions as the Atomic Unit

Sessions became the atomic unit of the system.

Prompt

Too narrow to explain retries, switching, and cost accumulation.

Project

Too broad to show one bounded unit of AI-assisted work.

Session

The observable unit where provider, model, project, cost, retry, status, and time meet.

Sessions group provider, model, project, cost, retry, status, and time into one observable unit.
SuccessRetry HeavyExpensiveOptimizedFailedStalledUnclassified

Upcoming artifact

Design pending

Session detail screen

Why it matters
The table proves the object exists. A detail screen will show how one unit of AI-assisted work unfolds.
What it should prove
Retries, model switching, cost accumulation, output status, project context, and signal tags belong to one session.
Expected content
One session timeline with provider/model, retry events, cost changes, status, linked project, and resolution state.
Session detail screen

Projects

Projects as Context

Project context turns activity into work. A cost spike means something different in auth, onboarding, pricing, research, or support.

Upcoming artifact

Design pending

Projects screen

Why it matters
Project context is visible in the session table, but it needs a dedicated artifact.
What it should prove
AI usage can be grouped by product, feature, team effort, or workstream.
Expected content
Project rows with sessions, provider mix, spend, retry-heavy sessions, failed sessions, recent activity, and active signals.
Projects screen

Signals

Signals as Operational Decisions

Signals are interpreted patterns. They exist to point a person toward inspection, comparison, reconnection, or review.

Retry-heavy sessionsFailed generationsModel switchingCost accumulationMissing dataStale sync

Upcoming artifact

Design pending

Signals screen

Why it matters
The case study needs evidence for how Ledger turns session data into operational attention.
What it should prove
Signals are not raw metrics; they are decision prompts tied to sessions, projects, and providers.
Expected content
Retry Heavy, Expensive, Failed, Optimized, Missing Data, and Stale Sync groups with linked objects and suggested next actions.
Signals screen

Governance / Trust

Governance Starts With Coverage

The trust model is intentionally narrow: read-only permissions, API key state, provider health, data coverage, expired credentials, and missing data.

Team visibility is product direction, not a claim of enterprise compliance automation. The current story is about making governance gaps visible before teams act on the data.

Upcoming artifact

Design pending

Governance/trust screen

Why it matters
Governance needs to stay grounded in concrete trust states.
What it should prove
Teams can see whether the operational picture is complete enough to trust.
Expected content
Provider permissions, read-only access, API key state, coverage, expired credentials, disconnected providers, and team visibility boundaries.
Governance/trust screen

Intervention Paths

From Signal to Next Action

The product direction is not just surfacing anomalies. It is helping a responsible person decide what to inspect, reconnect, compare, investigate, or resolve.

Upcoming artifact

Design pending

Intervention path screen

Why it matters
A signal should not be a dead end.
What it should prove
Operational signals can lead to clear next actions without claiming completed automation.
Expected content
Inspect session, reconnect provider, compare models, investigate failed generation, review cost spike, assign owner, and resolve.
Intervention path screen

Prototype Honesty

What Is Still Ongoing

Implemented prototype behavior

Active React, Next.js, Tailwind CSS, and TypeScript prototype. OpenAI can be connected and read in the testing environment.

Designed product model

Providers, sessions, projects, signals, governance, and intervention paths define the information architecture.

In-progress provider work

Additional providers are planned or being built. The case study does not claim complete coverage.

Future product direction

Deeper signal logic, team visibility, governance views, intervention workflows, and beta testing remain active work.

Reflection / Next

The Interface Around the AI Tool

Ledger clarified that the interface around AI work matters as much as the AI tool itself.

Provider health, session behavior, cost accumulation, project context, governance gaps, and intervention paths are the layer where teams understand AI-assisted work.

Next, the product needs stronger screen evidence for session detail, project context, signal triage, governance states, and intervention workflows.