Kategorie: AI & Automation

Veröffentlicht am

AI Code Review Needs a Platform Contract: Governing Instructions, Network Access, and Runners

AI-assisted code review is moving beyond a simple product toggle. Once a reviewer can read repository instructions, execute setup steps, access networks, and run on different compute environments, it becomes part of the software delivery platform. That changes the governance question from “Should we enable AI review?” to “Under which versioned and auditable contract may it operate?”

GitHub’s latest Copilot code review improvements make that shift visible. Copilot code review can now read instructions from the pull request’s head branch, consume additional instruction formats such as REVIEW.md, GEMINI.md, and CLAUDE.md, execute custom setup steps, run behind a configurable firewall, and use runner settings that are independent from the Copilot cloud agent.

These capabilities can improve relevance and reduce false positives. They also expand the execution surface around every review. Enterprises should therefore treat the configuration as platform policy rather than scattered repository preferences.

The reviewer now has an execution environment

Traditional static analysis usually operates under a relatively fixed contract: a known rule set evaluates a defined artifact and emits deterministic results. An AI reviewer behaves differently. Its output depends on context, instructions, available tools, network access, and the environment in which it runs.

AI review can work in an enterprise, but model selection is only one part of the operating model.

Four control planes now shape the result:

  • Instruction policy: which repository files influence the review and who may change them.
  • Setup policy: which commands and dependencies prepare the review environment.
  • Network policy: which external destinations the reviewer may reach.
  • Runner policy: where the review executes and which identities, secrets, and internal resources are available there.

If these controls evolve independently, two repositories using the same AI reviewer can have very different security and quality properties. A platform contract makes those differences explicit.

Instructions are executable review policy

Reading custom instructions from the head branch is useful because teams can test review guidance before merging it. It also creates a trust-boundary question: a pull request can propose code changes and changes to the instructions used to review that code.

This is not inherently unsafe. Pull requests already modify workflows, build scripts, and policy files. The important point is that instruction files deserve comparable governance.

A practical enterprise baseline should:

  • define which instruction formats are supported and where they may live;
  • assign CODEOWNERS to organization-critical instruction and agent-skill paths;
  • require human review when a pull request changes both application code and reviewer policy;
  • separate mandatory organization guidance from repository-specific context;
  • test instruction changes against representative pull requests before broad rollout;
  • retain the exact instruction revision with review evidence where audit requirements justify it.

Product teams still need room for domain-specific guidance. The platform team should own the contract, validation, and protected baseline, while repository owners maintain local context within those boundaries.

Setup steps belong to the software supply chain

Custom setup steps can make reviews substantially better. A reviewer may need generated code, dependency metadata, project-specific linters, or a compiled schema before it can understand a change.

However, setup steps are code execution. They can download dependencies, run scripts from the branch, and interact with credentials exposed to the runner. Treating them as harmless customization would recreate risks that enterprises already learned to manage in CI/CD.

The platform contract should require:

  • minimal and explicit setup steps;
  • pinned versions or immutable references for downloaded tools and reusable actions;
  • dependency caching that does not weaken provenance checks;
  • timeouts and resource limits;
  • no production credentials in the review environment;
  • clear separation between untrusted pull-request code and trusted setup logic;
  • logs that show what ran without exposing secrets.

Where setup behavior is common across many repositories, publish a maintained golden path instead of copying shell fragments. Teams can then inherit a reviewed baseline and add only the repository-specific preparation they need.

Network access needs an exception model

GitHub states that Copilot code review now runs behind a firewall by default and that its network rules can be configured independently from the Copilot cloud agent. This is an important default because review tasks often need less network access than general-purpose coding agents.

The right enterprise pattern is deny by default with narrow, observable exceptions. Typical allowed destinations may include an internal package repository, a schema registry, or an approved documentation endpoint. General internet access should require a documented reason.

Every exception should answer four questions:

  1. Which destination is required?
  2. Which review capability depends on it?
  3. What data may leave the runner?
  4. Who owns and periodically revalidates the exception?

The current GitHub notice also matters operationally: self-hosted runners do not support the code-review firewall. Organizations using self-hosted runners therefore need compensating controls at the network, workload, or runner-pool level. “Self-hosted” should not be interpreted as “automatically safer.”

Runner selection is a data-boundary decision

Separating runner configuration for Copilot code review from the Copilot cloud agent allows enterprises to match infrastructure to the task. It also prevents an overly broad agent configuration from becoming the default for review workloads.

Runner policy should be based on data classification and required connectivity:

  • use isolated hosted runners when repositories need no private network access;
  • use hardened self-hosted pools only when internal dependencies make them necessary;
  • keep runner images minimal and reproducible;
  • use short-lived credentials with repository-scoped permissions;
  • prevent persistence between review jobs;
  • separate runner pools for different trust zones.

Runner selection affects far more than performance. It determines what source code, internal services, credentials, and telemetry the review process can reach.

A platform contract for AI code review

A useful contract should be small enough for teams to understand and strict enough for governance teams to trust. It can be implemented as policy-as-code, reusable repository configuration, protected settings, and automated conformance checks.

The contract should define at least:

  1. Purpose: AI review supplements human review and deterministic controls; it does not become an unmeasured release gate.
  2. Instruction ownership: approved formats, protected paths, and required reviewers.
  3. Execution policy: allowed setup operations, dependency provenance, resource limits, and secret handling.
  4. Egress policy: default network posture, approved destinations, and exception ownership.
  5. Runner policy: permitted runner classes by repository classification and connectivity need.
  6. Evidence: the configuration version, runner class, policy result, and operational outcome retained for each review where needed.
  7. Escalation: how developers report incorrect findings, missing context, or policy constraints that reduce review quality.

This contract should be versioned. A change to review instructions, setup behavior, firewall rules, or runner selection can alter both quality and risk. It deserves the same change-management discipline as a CI template or deployment policy.

Measure the operating model, not just adoption

Counting how many repositories enabled AI review says little about whether it works safely. Platform and engineering leaders need evidence across quality, reliability, and governance.

Useful measures include:

  • acceptance and dismissal rates for review findings;
  • time from finding to resolution;
  • setup-step failure and timeout rates;
  • repositories compliant with the approved runner and firewall baseline;
  • number and age of network exceptions;
  • frequency of instruction changes and their effect on finding quality;
  • cases where deterministic checks or human reviewers contradicted the AI reviewer.

These measures create a feedback loop. Teams can improve instructions and setup context while the platform team identifies systemic policy or reliability problems.

A pragmatic rollout sequence

Enterprises do not need to design the final governance model before gaining value. A controlled rollout can proceed in four steps:

  1. Inventory: identify existing instruction files, setup workflows, runner types, and network dependencies.
  2. Baseline: define one default contract with protected instructions, minimal setup, restricted egress, and an approved runner class.
  3. Pilot: onboard a representative set of repositories and measure finding quality, setup reliability, and exceptions.
  4. Scale: publish golden-path configuration, automate conformance checks, and review exceptions periodically.

This approach lets the organization learn without turning every repository into a separate governance experiment.

From feature configuration to delivery policy

GitHub’s new controls are useful because they expose the real shape of AI-assisted review: it is a configurable workload operating inside the delivery system. Instructions influence judgment. Setup steps execute code. Network rules define egress. Runners define the data boundary.

Enterprises now have to make this customization reviewable, reusable, and measurable.

Platform teams should provide the paved road: a versioned contract, secure defaults, approved extension points, and evidence that the reviewer operated within policy. Product teams should own the domain context that makes reviews useful. Security and engineering leadership should define the exceptions and success measures.

That division of responsibility allows AI code review to improve delivery quality without creating an invisible parallel control plane.

GitHub’s implementation details are described in the Copilot code review customization announcement, the code review environment documentation, the repository custom instructions guidance, and the firewall configuration guidance.

Veröffentlicht am

Pods Are Workers, Not Agents: Designing the Runtime Boundary for Enterprise Agent Platforms

Kubernetes Pods are excellent execution units. They provide scheduling, resource controls, networking, workload identity integration, and a natural boundary for security and observability.

That does not automatically make a Pod the right representation of an AI agent.

Enterprise agent platforms need to distinguish two concepts that are easy to collapse during early implementations: the logical agent and the runtime worker executing its current task. Treating them as the same object can work for prototypes and continuously running agents. At scale, it creates idle infrastructure, slow burst handling, fragmented identity, and weak lifecycle semantics.

The durable pattern is to let Kubernetes manage execution workers while an agent control plane manages agent identity, state, policy, placement, and lifecycle. Pods remain essential. They become workers rather than the agent itself.

Why one Pod per agent is an attractive first design

The one-agent-per-Pod model solves several real problems quickly.

  • A Pod provides a process and container isolation boundary.
  • A ServiceAccount gives the workload a Kubernetes identity.
  • NetworkPolicy and admission policy can constrain its environment.
  • CPU and memory requests make resource consumption schedulable.
  • Logs, metrics, and traces can be attributed to a workload instance.
  • Existing GitOps, deployment, and incident-response practices remain usable.

For a small number of high-value agents, those benefits may outweigh the overhead. The model is understandable and conservative. It uses boundaries that platform and security teams already know how to operate.

The problem appears when the organization assumes that the execution container is also the durable identity and lifecycle of the agent.

Agents do not behave like ordinary services

A typical service is expected to remain available and handle a continuing stream of requests. An agent may wake up for a task, run for seconds or minutes, wait for a human decision, delegate work to subagents, and then remain idle for hours.

These characteristics create a different workload shape:

  • Bursty demand: a single business event can fan out into many parallel agent tasks.
  • Long idle periods: logical agents may exist without needing compute.
  • External waiting: execution may pause for approval, data, or another system.
  • Variable duration: tasks range from short tool calls to extended research or coding sessions.
  • Delegated authority: an agent often acts on behalf of a user or workflow rather than only as itself.
  • Stateful continuation: a later execution may need to resume the same logical conversation or plan on a different worker.

Keeping one Pod alive for every logical agent reserves capacity for identities that are not doing work. Creating a fresh Pod for every short task can introduce startup latency and control-plane churn. Encoding state inside the Pod makes rescheduling and recovery harder.

The architectural question is therefore not whether Kubernetes should run agents. It is which responsibilities belong to Kubernetes and which belong to an agent-specific control plane.

The runtime boundary: agents, actors, and workers

A recent CNCF article describing kagent’s agent-substrate architecture illustrates this separation. Kubernetes continues to manage Pods, networking, storage, and compute. A higher-level control plane manages logical actors and places them onto a pool of execution workers.

In that model:

  • The logical agent has durable identity, ownership, policy, configuration, and state.
  • An agent task or actor instance represents a unit of active execution.
  • A worker is a sandboxed runtime capable of executing one or more assigned actors.
  • A worker pool defines capacity, runtime profile, isolation class, and placement characteristics.

Agent-substrate is one implementation, not a universal enterprise standard. Its value for platform design is the principle it demonstrates: logical lifecycle can be decoupled from Pod lifecycle without removing Kubernetes from the architecture.

Six contracts the control plane must preserve

Decoupling an agent from a Pod improves efficiency only if the platform preserves the controls that dedicated Pods made easy.

1. Durable agent identity

An agent needs an identity that survives worker replacement. That identity should identify the agent definition, tenant, owner, environment, risk tier, and approved capabilities.

The worker also needs its own workload identity. The two must not be confused. A worker identity proves which runtime is communicating with the platform. The agent identity determines which business permissions and policies apply to the assigned execution.

When an agent acts for a person, the authorization decision should include delegated user context with explicit scope and expiry. Copying a user’s full credentials into a worker is not delegation.

2. Execution leases

Placement should create a time-bound execution lease binding an agent task to a specific worker. The lease should include the agent identity, policy revision, tool permissions, state reference, deadline, and expected resource profile.

Leases make reassignment and failure handling explicit. If a worker disappears, the control plane can determine whether the task is safe to retry, must resume from a checkpoint, or requires human review.

3. Isolation classes

Sharing workers does not mean sharing trust. The platform needs multiple runtime profiles based on risk.

  • Low-risk, read-only tasks may use a warm multi-tenant worker pool.
  • Tasks handling confidential data may require stronger sandboxing and tenant-dedicated workers.
  • Agents with write access to production systems may require a dedicated Pod or ephemeral sandbox per execution.
  • Untrusted code execution may require gVisor, microVMs, or another hardened isolation boundary.

The scheduling decision should derive from policy. Developers should request a workload class rather than select a weaker runtime to reduce latency.

4. Policy attribution

Kubernetes policy usually sees the Pod, namespace, and ServiceAccount. A shared worker introduces another logical principal inside that boundary. The platform must propagate agent, tenant, task, and delegated-user context to every policy enforcement point.

Tool gateways, model gateways, data APIs, and egress proxies should authorize the logical execution, not merely trust the worker’s network location. Audit events should record both worker identity and agent identity so investigators can reconstruct who did what and where it ran.

5. Externalized state and checkpoints

Agent state should not depend on the continued existence of a worker Pod. Conversation state, plans, artifacts, approval state, and checkpoints need durable storage with tenant-aware encryption and retention controls.

Externalizing state allows the platform to release compute while an agent is idle and rehydrate it when work resumes. It also creates a controlled recovery point instead of treating the worker filesystem as an accidental system of record.

6. End-to-end observability

Pod-level telemetry remains necessary but is no longer sufficient. Operators need to follow a logical agent across workers and over time.

Every execution should carry stable correlation fields such as:

  • agent, tenant, task, session, and parent-task identifiers;
  • worker and worker-pool identity;
  • policy, prompt, model, and tool versions;
  • delegated user and approval references where permitted;
  • token, latency, tool-call, cost, and outcome signals;
  • checkpoint, retry, reassignment, and termination reasons.

This creates observability for the business execution rather than only for the container currently hosting it.

A reference enterprise architecture

A practical runtime separates responsibilities across four layers.

Agent control plane

The control plane stores agent definitions, ownership, policy, lifecycle, state references, and desired runtime class. It accepts tasks, decides placement, issues leases, tracks execution, and coordinates retries or resumptions.

Worker pools

Kubernetes Deployments or other controllers maintain warm capacity for defined execution profiles. Pools may differ by tenant, geography, accelerator, sandbox technology, network access, or data classification.

Shared platform gateways

Model, tool, MCP, data, and egress gateways enforce logical identity and policy. They keep privileged credentials out of agent code and provide consistent rate limits, approval checks, observability, and revocation.

Durable state and evidence

State services store checkpoints and artifacts. An evidence plane records immutable links between the agent definition, execution lease, policy decision, worker, model interaction, tool call, and outcome.

Kubernetes remains the infrastructure substrate. The agent control plane provides semantics Kubernetes was not designed to infer.

Multi-tenancy must shape worker placement

Worker utilization can improve dramatically when idle logical agents do not retain Pods. That benefit should not override tenant boundaries.

Platform teams should define placement rules covering:

  • whether tenants may share a worker process, Pod, node, or cluster;
  • which data classifications require dedicated runtime capacity;
  • how memory, filesystems, caches, and credentials are cleared between assignments;
  • whether agent-generated code can execute and under which sandbox;
  • which tools and destinations each pool can reach;
  • how noisy-neighbor behavior is detected and constrained;
  • where state and inference traffic may be processed geographically.

There is no single correct sharing boundary. The platform should offer a small set of reviewed isolation classes and make the selected class visible in cost, latency, and risk reporting.

When one Pod per agent is still the right answer

Decoupling should not become an objective by itself. A dedicated Pod remains a strong choice when:

  • the agent is continuously active or exposes a stable service endpoint;
  • startup latency is acceptable and the fleet is small;
  • the workload needs strong tenant or process isolation;
  • it runs untrusted code or privileged tools;
  • its memory and resource profile do not fit a shared pool;
  • existing Kubernetes controls provide sufficient lifecycle semantics;
  • the added agent scheduler would cost more to operate than it saves.

The mature platform supports more than one runtime pattern. It chooses the boundary based on workload behavior and risk rather than forcing every agent into the same optimization.

Measure the runtime as a platform product

Worker density is useful, but cost efficiency alone is an incomplete success measure. Track flow, reliability, isolation, and control together.

  • Task queue time and time to first execution
  • Warm-start and cold-start latency
  • Active versus idle worker utilization
  • Logical agents per worker and per isolation class
  • Checkpoint, resume, retry, and reassignment success rates
  • Policy denials and unauthorized cross-tenant attempts
  • State cleanup and credential revocation failures
  • Cost per successful agent task
  • Trace and audit coverage from task request to external side effect

A cheaper runtime that cannot explain an agent’s actions is not an enterprise improvement.

A staged adoption path

1. Separate identifiers before changing runtime

Introduce stable agent, task, tenant, and worker identifiers in the current platform. Propagate them through logs, traces, policy decisions, and tool calls. This exposes hidden coupling before a scheduler is introduced.

2. Externalize state

Move durable state and artifacts out of the Pod. Define checkpoint, retry, expiry, encryption, and deletion semantics. Test recovery from worker termination.

3. Add one low-risk worker pool

Select bursty, read-only tasks with clear resource limits. Compare queue time, utilization, cost, and operational effort with the dedicated-Pod baseline.

4. Add policy-aware placement

Introduce reviewed isolation classes and execution leases. Integrate logical identity with tool, model, data, and egress gateways. Exercise tenant separation and credential revocation.

5. Expand only with evidence

Move higher-risk agents after proving state hygiene, observability, rollback, and incident response. Keep dedicated Pods as an explicit option rather than treating them as a failed legacy design.

Pods should host work, not define the agent

The Pod remains one of the strongest execution boundaries available to cloud-native platforms. The mistake is asking it to carry semantics it does not own: durable agent identity, delegated authority, conversation lifecycle, human approval, and cross-execution state.

Enterprise agent platforms should model those concerns explicitly. Kubernetes can then do what it does best — schedule and isolate execution — while the agent control plane decides which logical work runs where, under whose authority, with which policy, and with what evidence.

That separation improves utilization, but its greater value is governance. It allows the platform to scale agents without losing the identity and accountability that production systems require.

Sources

Veröffentlicht am

The AI-Native Platform Contract: Expanding Golden Paths Beyond Application Delivery

Platform engineering earned its place by turning application delivery into a repeatable product. Golden paths combined infrastructure, security, deployment, and operational standards into a paved route that developers could use without learning every platform detail.

AI-native workloads do not invalidate that model. They expose where it stops too early.

A conventional golden path typically starts with source code and ends with a running service. An AI-native product depends on a wider chain: governed data, accelerator capacity, models and prompts, evaluation evidence, inference controls, agent identities, external tools, and continuous cost and risk feedback. If each of those capabilities arrives through a separate specialist portal, the organization has not created an AI platform. It has created another integration problem.

The next platform contract should therefore extend the golden path rather than build a parallel AI silo. The goal is not to hide every AI decision behind automation. It is to make safe defaults easy, exceptions explicit, and every promoted artifact traceable.

The application delivery contract is no longer enough

Platform Engineering 1.0 concentrated on a familiar delivery unit: an application packaged as a container, deployed through a pipeline, and operated with standard observability and security controls. That remains valuable, but AI changes both the workload and its consumers.

ML engineers need experiment tracking, model registries, feature and data access, and specialized compute. Application teams need stable inference endpoints and predictable latency. Security teams need controls for model provenance, prompt injection, data leakage, and non-human identities. FinOps teams need to attribute expensive training and inference usage. AI agents themselves become platform consumers that request tools, credentials, and runtime actions.

The CNCF discussion of evolving platform engineering for AI-native workloads captures this expansion through capabilities such as GPU and TPU allocation, model serving, MCP gateways, agentic guardrails, embedded FinOps, and policy-driven governance. The important organizational point is that these should not become an isolated platform owned by a small AI team. They should become extensions of the same product model, interfaces, and control philosophy used by the enterprise platform.

Define a platform contract, not a catalog of tools

A platform contract describes what a product team can request, what evidence it must provide, what the platform guarantees, and which controls are automatically applied. It is stronger than a service catalog entry and more flexible than a single mandatory implementation.

For an AI-native workload, that contract should cover at least six dimensions.

1. Governed data access

The path should make data classification, residency, retention, and permitted use visible before a workload reaches production. A request for a dataset should resolve to an approved identity, purpose, environment, and audit trail. The platform can automate access, but the product team remains accountable for whether the data is appropriate for the use case.

2. Compute and accelerator intent

Teams should request capabilities rather than hard-code a particular GPU model into every manifest. The contract can express workload class, memory, performance objective, duration, geographic constraints, and cost ceiling. Kubernetes mechanisms such as Dynamic Resource Allocation can support more structured resource claims, but the platform still needs policy for quotas, scarcity, preemption, and approved hardware profiles.

3. Model, prompt, and artifact provenance

Container images are not the only production artifacts. The platform must track model version, source, license, evaluation result, prompt bundle, retrieval configuration, tool definitions, and deployment policy. Promotion should be based on an immutable set of linked artifacts, not a model name copied into an environment variable.

4. Evaluation as a release gate

AI quality is probabilistic and context-dependent. A successful build does not prove production fitness. Golden paths should provide standard evaluation suites for task quality, safety, latency, robustness, and cost. Teams can add domain-specific tests, while the platform supplies the execution environment, evidence format, thresholds, and promotion workflow.

5. Runtime identity and guardrails

An inference service or autonomous agent needs a workload identity, scoped data access, approved tools, network boundaries, and observable policy decisions. The contract should distinguish a human user’s authority from an agent’s delegated authority. It should also define what happens when a model, tool, or policy is unavailable rather than allowing silent fallback to an uncontrolled path.

6. Cost and operational accountability

AI infrastructure introduces different cost behavior from ordinary stateless services. Training jobs can consume scarce capacity in bursts. Inference cost depends on model choice, token volume, batching, cache efficiency, and service-level objectives. Cost attribution and budgets should therefore be part of provisioning and release decisions, not a dashboard reviewed after the invoice arrives.

What an AI-native golden path looks like

A useful golden path follows the product lifecycle rather than exposing a collection of disconnected infrastructure forms.

  1. Declare the workload. The team selects an archetype such as batch training, online inference, retrieval-augmented generation, or tool-using agent. It declares data class, expected scale, latency objective, risk tier, and ownership.
  2. Provision an isolated workspace. The platform creates namespaces, identities, network boundaries, secrets references, storage, accelerator claims, quotas, and standard telemetry.
  3. Develop with approved building blocks. Teams consume versioned model endpoints, registries, feature services, MCP or tool gateways, and evaluation templates through stable APIs.
  4. Produce evidence. CI records model and data lineage, software dependencies, evaluation results, policy decisions, security findings, and predicted operating cost.
  5. Promote as a release set. GitOps promotes the linked application, model, prompt, policy, and tool configuration together. A rollback restores the complete known-good set.
  6. Operate with continuous feedback. Runtime telemetry covers service health, model quality indicators, policy denials, data drift, tool calls, accelerator utilization, and unit economics.

This lifecycle gives specialists room to innovate without forcing every product team to assemble the control plane themselves.

Avoid the separate AI platform trap

A dedicated AI enablement team may be necessary, but a separate delivery system should not be the default. Parallel identity models, pipelines, policy engines, and observability stacks increase cost and weaken governance. They also create a handoff between application engineers and AI specialists exactly where the product needs shared accountability.

A better operating model separates platform ownership by capability while preserving one product contract:

  • The core platform team owns common interfaces, workload identity, delivery workflows, policy integration, and the developer experience.
  • The AI platform capability team owns model-serving patterns, evaluation services, accelerator profiles, registries, and AI-specific runtime controls.
  • Data teams own governed data products and access semantics.
  • Security and risk teams define control objectives and approval boundaries as policy and evidence requirements.
  • Product teams own business fitness, domain evaluations, production outcomes, and accepted residual risk.

The teams collaborate through APIs, schemas, policy bundles, and service-level objectives rather than tickets and undocumented exceptions.

Measure whether the contract creates value

An AI-native platform should not be measured by the number of services in its catalog. Measure whether teams can deliver trustworthy outcomes faster.

  • Time from approved use case to first governed experiment
  • Time from candidate model to production release
  • Percentage of releases with complete model, data, prompt, and policy provenance
  • Evaluation failure escape rate
  • Percentage of agent tool calls using approved identities and gateways
  • Accelerator utilization and queue time by workload class
  • Inference cost per business transaction
  • Rollback time for a complete AI release set
  • Adoption and exception rates for each golden path

These metrics reveal whether the platform improves flow and control together. High adoption with slow delivery signals an overloaded path. Fast delivery with weak evidence signals unmanaged risk.

A practical 90-day starting point

Do not begin by designing a universal AI platform. Choose one real workload and use it to define the minimum viable contract.

Days 1–30: map the lifecycle

  • Select one representative AI product with a committed owner.
  • Map every artifact, identity, environment, approval, and operational dependency.
  • Classify which existing platform capabilities can be reused and where AI-specific gaps exist.
  • Define the workload’s risk tier, evaluation evidence, and cost objectives.

Days 31–60: build one vertical path

  • Create one workload template and governed workspace.
  • Connect model and prompt provenance to the existing GitOps release flow.
  • Add standard telemetry, policy checks, evaluation execution, and cost labels.
  • Document escape hatches with owners, expiry dates, and review requirements.

Days 61–90: prove and productize

  • Run a production-like release and rollback.
  • Measure lead time, evidence completeness, operational quality, and unit cost.
  • Interview the platform consumers and remove unnecessary steps.
  • Publish the contract as versioned schemas, APIs, examples, and service-level expectations.

The platform becomes the organizational control surface

AI-native platform engineering is not a race to add GPUs and model registries to an internal portal. It is the work of extending a proven product contract across a more complex value stream.

The strongest platforms will preserve what already works: product thinking, self-service, golden paths, policy automation, and composable cloud-native interfaces. They will add the missing contracts for data, models, evaluations, agents, specialized compute, and cost. That approach avoids a new silo while giving teams a credible path from experimentation to governed production.

Sources

Veröffentlicht am

AI Observability: Why Your AI Agents Need OpenTelemetry

The Black Box Problem in AI Agents

When you deploy an AI agent in production, you’re essentially running a complex system that makes decisions, calls external APIs, processes data, and interacts with users—all in ways that can be difficult to understand after the fact. Traditional logging tells you that something happened, but not why or how long or at what cost.

For LLM-based systems, this opacity becomes a serious operational challenge:

  • Token costs can spiral without visibility into per-request usage
  • Latency issues hide in the pipeline between prompt and response
  • Tool calls (file reads, API requests, code execution) happen invisibly
  • Context window management affects quality but rarely surfaces in logs

The answer? Observability—specifically, distributed tracing designed for AI workloads.

OpenTelemetry: The Standard not only for AI Observability

OpenTelemetry (OTEL) has emerged as the industry standard for collecting telemetry data—traces, metrics, and logs—from distributed systems. What makes it particularly powerful for AI applications:

Traces Show the Full Picture

A single user message to an AI agent might trigger:

  1. Webhook reception from Telegram/Slack
  2. Session state lookup
  3. Context assembly (system prompt + history + tools)
  4. LLM API call to Anthropic/OpenAI
  5. Tool execution (file read, web search, code run)
  6. Response streaming back to user

With OTEL traces, each step becomes a span with timing, attributes, and relationships. You can see exactly where time is spent and where failures occur.

Metrics for Cost Control

OTEL metrics give you counters and histograms for:

  • tokens.input / tokens.output per request
  • cost.usd aggregated by model, channel, or user
  • run.duration_ms to track response latency
  • context.tokens to monitor context window usage

This transforms AI spend from „we used $X this month“ to „user Y’s workflow Z costs $0.12 per run.“

Practical Setup: OpenClaw + Jaeger

At it-stud.io, we tested OpenClaw as our AI agent framework – already supporting OTEL by default – and enabled full observability with a simple configuration change:

{
  "plugins": {
    "allow": ["diagnostics-otel"],
    "entries": {
      "diagnostics-otel": { "enabled": true }
    }
  },
  "diagnostics": {
    "enabled": true,
    "otel": {
      "enabled": true,
      "endpoint": "http://localhost:4318",
      "serviceName": "openclaw-gateway",
      "traces": true,
      "metrics": true,
      "sampleRate": 1.0
    }
  }
}

For the backend, we chose Jaeger—a CNCF-graduated project that provides:

  • OTLP ingestion (HTTP on port 4318)
  • Trace storage and search
  • Clean web UI for exploration
  • Zero external dependencies (all-in-one binary)

What You See: Real Traces from AI Operations

Once enabled, every AI interaction generates rich telemetry:

openclaw.model.usage

  • Provider, model name, channel
  • Input/output/cache tokens
  • Cost in USD
  • Duration in milliseconds
  • Session and run identifiers

openclaw.message.processed

  • Message lifecycle from queue to response
  • Outcome (success/error/timeout)
  • Chat and user context

openclaw.webhook.processed

  • Inbound webhook handling per channel
  • Processing duration
  • Error tracking

From Tracing to AI Governance

Observability isn’t just about debugging—it’s the foundation for:

Cost Allocation

Attribute AI spend to specific projects, users, or workflows. Essential for enterprise deployments where multiple teams share infrastructure.

Compliance & Auditing

Traces provide an immutable record of what the AI did, when, and why. Critical for regulated industries and internal governance.

Performance Optimization

Identify slow tool calls, optimize prompt templates, right-size model selection based on actual latency requirements.

Capacity Planning

Metrics trends inform scaling decisions and budget forecasting.

Getting Started

If you’re running AI agents in production without observability, you’re flying blind. The good news: implementing OTEL is straightforward with modern frameworks.

Our recommended stack:

  • Instrumentation: Framework-native (OpenClaw, LangChain, etc.) or OpenLLMetry
  • Collection: OTEL Collector or direct OTLP export
  • Backend: Jaeger (simple), Grafana Tempo (scalable), or Langfuse (LLM-specific)

The investment is minimal; the visibility is transformative.

At it-stud.io, we help organizations build observable, governable AI systems. Interested in implementing AI observability for your team? Get in touch.