Timothy Wong

Topic dashboard

AI Analytics

Last refreshed May 10, 2026 · 17 concepts

AI Analytics

Analytics is becoming a streaming, agent-driven discipline, and the human-opening-a-dashboard workflow is no longer the centre of the system.

My take

Most data teams are still optimising for a human opening a dashboard, asking a question, and waiting. That workflow doesn’t go away, but it stops being primary. When the consumer is an agent running in a loop, the contract changes - agents want structured answers with provenance, latency, and cost characteristics that look nothing like BI as we built it.

The unit of work in this new world isn’t the dashboard, and it isn’t the data contract - contracts are a 2022 idea. It’s the context layer that makes data legible to agents: business context, org context, interaction history, governance, managed as a first-class platform. Whoever builds this gets compounding returns. Whoever skips it ends up with a graveyard of one-off RAG pipelines that each rediscover the same business logic.

This is a harness problem. Models keep getting better but stay noisy. What compounds is the system around them - context, observability, evals, human review. “Add an LLM to the BI tool” loses to platform rebuild.

What I’m watching:

  • Evals and harness design becoming the core craft - the teams that win will be the ones who can measure agent analytics quality the way we used to measure model accuracy, and iterate on the system around the model rather than the model itself.
  • Context-building paired with observability as a single discipline - you can’t curate context you can’t observe in production, and you can’t trust observability without the context layer underneath it.
  • Whether the dashboard layer goes AI-native fast enough to stay relevant, or gets routed around entirely.

Everything above the divider is mine. Everything below is auto-assembled daily from my knowledge base — individual links and summaries may be stale or off-target. Last refreshed: 2026-05-10.

What’s shifted recently

  • Agentic Analytics Bi Evolution (updated 2026-05-09)
    Agentic analytics is the structural shift in business intelligence from static dashboards and human-triggered queries to AI agents that autonomously plan, retrieve, reason across,… — source · source · source

  • Agentic Edge Deployment (updated 2026-05-09)
    Agentic edge deployment is the pattern of running autonomous AI agents at or near the data source — on-device, in a local data centre rack, or at a network edge node — rather than… — source · source · source

  • Enterprise Agent Implementation Labor (updated 2026-05-09)
    Enterprise agent implementation labor is the emerging category of work — and workers — created specifically by the gap between agent capabilities and production-ready enterprise d… — source · source · source

  • Enterprise Agent Observability OTEL Tracing (updated 2026-05-09)
    Enterprise agent observability is the discipline of instrumenting every step of an autonomous agent’s execution — tool calls, retrieval operations, subagent invocations, and inter… — source · source · source

  • Enterprise Context Layer For Agents (updated 2026-05-09)
    An enterprise context layer for agents is the governed infrastructure that sits between an organization’s data systems and any AI agent querying them, translating raw metadata int… — source · source · source

  • Enterprise Agent Integration Layer (updated 2026-05-08)
    The enterprise agent integration layer refers to the set of CLI tools, APIs, protocol adapters, and platform extensions that established enterprise software vendors are building t… — source · source · source

  • Production Agent Platform Layer (updated 2026-05-08)
    The production agent platform layer is the set of shared infrastructure that sits between model APIs and business applications in deployed multi-agent systems — covering memory an… — source · source · source

  • Streaming Multi Agent Observability (updated 2026-05-08)
    Streaming multi-agent observability is the discipline of instrumenting continuously-running, event-driven AI agents — agents that operate on persistent data streams rather than di… — source · source · source

  • Agent CI CD Trust Boundary Expansion (updated 2026-05-07)
    AI coding agents deployed in CI/CD pipelines inherit the trust model of interactive developer tools — where a human is present to validate actions — but operate in headless, autom… — source · source · source

  • Agent Control Plane Execution Layer (updated 2026-05-07)
    The agent control plane is the infrastructure layer that sits between an organization’s business systems and its AI execution layer, enforcing cross-cutting operational concerns —… — source · source · source

  • Agent Roi Decision Intelligence Pricing (updated 2026-05-07)
    Agent ROI and decision-intelligence pricing are two sides of the same procurement problem: enterprises must justify agentic AI spend against metrics that automation-era frameworks… — source · source · source

  • Agent Subagent Decomposition Production Pattern (updated 2026-05-07)
    Agent-subagent decomposition is the architectural pattern of splitting a production AI workflow into a parent orchestrator and one or more specialized child agents, each scoped to… — source · source · source

  • Agentic Rag Iterative Retrieval (updated 2026-05-07)
    Agentic RAG (retrieval-augmented generation) is a retrieval architecture in which the LLM does not retrieve once and generate — it loops: grading retrieved chunks, rewriting queri… — source · source · source

  • AI Coding Acceleration Stack Order (updated 2026-05-07)
    AI coding agents do not accelerate all software work equally. — source · source · source

  • AI Native Banking Infrastructure (updated 2026-05-07)
    AI-native banking infrastructure refers to ground-up financial backend systems — ledger, payments orchestration, compliance, and AML — built from the start to expose AI agents as… — source · source · source

  • Local First Workflow Engine Deterministic Agents (updated 2026-05-03)
    Local-first workflow engines for AI agents are self-hosted, open-source platforms that execute multi-step agentic pipelines entirely within the operator’s own infrastructure — no… — source · source · source

  • Social Agent Queue Backpressure Pattern (updated 2026-05-03)
    Social agents — autonomous processes monitoring feeds on platforms like Farcaster, Moltbook, and Nostr — generate write bursts to downstream research or memory systems faster than… — source · source · source

The ideas I keep coming back to

Currently active (last 30 days):

  • Agentic Analytics Bi Evolution — Agentic analytics is the structural shift in business intelligence from static dashboards and human-triggered queries to AI agents that autonomously plan, retrieve, reason across,…
  • Agentic Edge Deployment — Agentic edge deployment is the pattern of running autonomous AI agents at or near the data source — on-device, in a local data centre rack, or at a network edge node — rather than…
  • Enterprise Agent Implementation Labor — Enterprise agent implementation labor is the emerging category of work — and workers — created specifically by the gap between agent capabilities and production-ready enterprise d…
  • Enterprise Agent Observability OTEL Tracing — Enterprise agent observability is the discipline of instrumenting every step of an autonomous agent’s execution — tool calls, retrieval operations, subagent invocations, and inter…
  • Enterprise Context Layer For Agents — An enterprise context layer for agents is the governed infrastructure that sits between an organization’s data systems and any AI agent querying them, translating raw metadata int…
  • Enterprise Agent Integration Layer — The enterprise agent integration layer refers to the set of CLI tools, APIs, protocol adapters, and platform extensions that established enterprise software vendors are building t…
  • Production Agent Platform Layer — The production agent platform layer is the set of shared infrastructure that sits between model APIs and business applications in deployed multi-agent systems — covering memory an…
  • Streaming Multi Agent Observability — Streaming multi-agent observability is the discipline of instrumenting continuously-running, event-driven AI agents — agents that operate on persistent data streams rather than di…
  • Agent CI CD Trust Boundary Expansion — AI coding agents deployed in CI/CD pipelines inherit the trust model of interactive developer tools — where a human is present to validate actions — but operate in headless, autom…
  • Agent Control Plane Execution Layer — The agent control plane is the infrastructure layer that sits between an organization’s business systems and its AI execution layer, enforcing cross-cutting operational concerns —…
  • Agent Roi Decision Intelligence Pricing — Agent ROI and decision-intelligence pricing are two sides of the same procurement problem: enterprises must justify agentic AI spend against metrics that automation-era frameworks…
  • Agent Subagent Decomposition Production Pattern — Agent-subagent decomposition is the architectural pattern of splitting a production AI workflow into a parent orchestrator and one or more specialized child agents, each scoped to…
  • Agentic Rag Iterative Retrieval — Agentic RAG (retrieval-augmented generation) is a retrieval architecture in which the LLM does not retrieve once and generate — it loops: grading retrieved chunks, rewriting queri…
  • AI Coding Acceleration Stack Order — AI coding agents do not accelerate all software work equally.
  • AI Native Banking Infrastructure — AI-native banking infrastructure refers to ground-up financial backend systems — ledger, payments orchestration, compliance, and AML — built from the start to expose AI agents as…
  • Local First Workflow Engine Deterministic Agents — Local-first workflow engines for AI agents are self-hosted, open-source platforms that execute multi-step agentic pipelines entirely within the operator’s own infrastructure — no…
  • Social Agent Queue Backpressure Pattern — Social agents — autonomous processes monitoring feeds on platforms like Farcaster, Moltbook, and Nostr — generate write bursts to downstream research or memory systems faster than…

Who I’m watching

No people or companies crosslinked to this area yet.

Sources I’ve been drawing on