AIGC BOT WEEKLY / 2026.08.09

Agent observability digest: accept one safe, connected trace first

OpenTelemetry Demo 3.0 puts an agent, model calls, and MCP tools on one telemetry path. The source cards below describe only what the official material makes checkable. Our editorial call is narrower: spend 45 minutes accepting one redacted, connected, versioned trace before investing in a polished dashboard.

Issue window: 2026-07-27 - 2026-08-02
Published: 2026-08-09
Ranking scope: AIGC Bot research entry point, not an official GitHub ranking

SOURCE-LINKED SIGNALS

Separate source facts before writing a judgment.

OBSERVABILITY / End-to-end path

Demo 3.0 routes agent-to-tool telemetry through one Collector path

Checkable fact: OpenTelemetry's July 28, 2026 Demo 3.0 article describes a LangGraph ReAct agent using native and MCP tools. Telemetry from the chatbot, agent, MCP layer, and downstream services flows through the OpenTelemetry Collector.
Why it belongs on the validation card: This belongs on the card because seeing a model request is not the same as following the parent-child path from an agent decision through tool execution and back. The latter is the boundary this lab accepts.

OpenTelemetry Demo 3.0 article

AI / Reproducible fixture

The versioned 3.0.0 Agent docs replace customer prompts with fixed fixtures

Checkable fact: The versioned OpenTelemetry Demo 3.0.0 Agent documentation describes its FastAPI endpoint, built-in and MCP tools, OTLP export configuration, canned requests that need no live LLM access, and VCR replay for deterministic tests.
Why it belongs on the validation card: This belongs on the card because telemetry acceptance can begin with a fixed, non-sensitive fixture instead of customer prompts or paid inference. That makes the service and field contract easier to reproduce.

OpenTelemetry Demo 3.0 Agent docs

DATA / Field contract

The GenAI normalizer exposes both standardized and source-native fields

Checkable fact: Merged Demo PR #3604 maps OpenLLMetry/Traceloop attributes to `gen_ai.*` while retaining the original attributes, and records Alpha stability for the normalizer processor used at that time. OpenTelemetry has also moved GenAI conventions into a dedicated repository.
Why it belongs on the validation card: This belongs on the card because a familiar field prefix does not guarantee stable meaning, versioning, or safe content. Record the source format, target convention, and version so an upgrade cannot silently invalidate a query.

OpenTelemetry Demo gen_ai_normalizer PR

ACTION LAB

Use 45 minutes to earn a continue-or-stop decision.

Goal: Decide whether current telemetry can safely connect one agent, model, and tool invocation, then leave a repeatable field contract.
Time boundary: 45 minutes / no full Demo deployment; use a redacted fixture, existing trace JSON, or a field inventory

01
5 min: draw five stages
Write only user request, agent, model, tool, and result. Give each stage an expected span and parent; do not paste a real prompt.
02
10 min: select minimum join fields
Mark trace/span parentage, service name, operation name, request/response model, and tool name for each stage. Write “missing” instead of inventing a value.
03
10 min: map the fields
Place current source fields beside target `gen_ai.*` fields. Record the source format, convention repository or version, and whether a normalizer participates.
04
10 min: run a redaction review
Remove API keys, identities, complete prompts/responses, and tool argument values. Keep only the enums, identifiers, and status needed for correlation and failure review.
05
5 min: draft one acceptance query
In your current backend's query language, draft a query that starts from the Agent span, finds model and tool spans in the same trace, and displays failure status.
06
5 min: sign continue or stop
Apply the conditions below. If passing requires raw content or guessed parentage, record the blocker and stop expanding the dashboard.

Expected output: One versioned acceptance card containing a five-stage path, source-to-`gen_ai.*` mapping, redaction checklist, one acceptance query, and a continue-or-stop decision.
Pass condition: One redacted fixture connects the agent, model, and at least one tool invocation through trace/span relationships; service, operation, model, tool, and failure state have explicit fields; the card names its convention source/version and contains no key, identity, or full conversation.
Stop condition: The path cannot be joined without collecting secrets, full prompts/responses, or high-risk tool arguments, or the fields cannot distinguish agent, model, and tool boundaries. Stop dashboard work and repair instrumentation, redaction, or the version contract first.

DOMAIN BOARDS / RESEARCH ENTRY

Three AIGC Bot research entry points

The notes below are AIGC Bot editorial guidance, not an official GitHub ranking or a performance claim about every agent framework. This issue asks one shared question: can you review an agent-to-tool call without collecting secrets or full conversations?

Models, workflows, and applications

AI Open Source

For AI, choose one fixed task and one side-effect-free tool. Capture the model and tool boundary without treating full prompts as default telemetry.

01
n8nConnect AI, product, and operations systems through reusable workflows.
Read analysis
02
FlowisePrototype LLM applications and orchestration flows before custom coding.
View source
03
ComfyUINode-based image-generation pipelines for controllable creative production.
View source

View the full board

Events, compute, and analytics

Big Data

For data, keep source fields, standardized fields, convention version, and redaction ownership in one mapping table instead of governing names in isolation.

01
Apache KafkaA durable event backbone for connecting services and data systems.
View source
02
Apache SparkDistributed processing for batch analysis and lakehouse workloads.
View source
03
TrinoQuery data where it lives across lakes, warehouses, and operational sources.
View source

View the full board

Quality, cost, and production governance

Observability

For observability, accept parentage, failure state, and one minimum query before adding content capture, metrics, or dashboards.

01
OpenTelemetry CollectorReceive, process, and route traces, metrics, and logs between systems.
Read analysis
02
PrometheusA foundational time-series monitoring system with a broad exporter ecosystem.
View source
03
GrafanaVisualize and correlate operational signals across a wide source ecosystem.
View source

View the full board

SOURCES & EDITORIAL SCOPE

Sources and editorial scope

This issue cites public primary project material and distinguishes AIGC Bot editorial judgment from project facts. The links support review; they are not an official ranking or an income promise.

PUBLISHING CADENCE

Every Sunday covers the previous week.

The digest follows a natural week: Monday through Sunday, then a Sunday review. Before email subscriptions open, this public page acts as SEO content and a sample library for the future newsletter.

NEXT ACTION

After the digest, keep one validation card.

Use the checklist to break one project into deployment, license, data, cost, and first-user questions before deciding whether it deserves a deep article, sponsor slot, or digital product package.

Get the checklist Read the Doris observability guide