8 min · intermediate · DevOps & CI/CD

Event fields from the OpenTelemetry log data model, message templates instead of interpolation, trace ids on every line, what never to log, and retention that matches the bill.

Topics covered

• logging
• structured-logging
• OpenTelemetry
• KQL
• observability

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/devops-cicd/structured-logging

8 min · advanced · AI Engineering

Prompt chaining, routing, orchestrator-workers, ReAct, plan-and-execute, reflection and tree search: what each costs, what the research shows, and when to climb to the next one.

Topics covered

• agent-patterns
• ReAct
• reflection
• workflows
• agents

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-engineering/agent-design-patterns

8 min · advanced · AI Engineering

KV cache arithmetic, PagedAttention, continuous batching, prefix caching, FP8 and speculative decoding, the vLLM flags that matter, and the five metrics that show which limit you hit.

Topics covered

• model-serving
• vLLM
• KV-cache
• quantization
• speculative-decoding

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-engineering/model-serving-with-vllm

5 min · intermediate · ai-coding

Agent mistakes are fluent, and their tests often confirm their bugs. A reading order that puts risk first, invented dependencies, fresh review sessions and review by risk.

Topics covered

• code-review
• coding-agents
• slopsquatting
• testing

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-coding/reviewing-agent-written-code

7 min · foundational · ai-coding

What belongs in your repo instructions, what to cut, how to share one file across Claude Code, Cursor and Codex, and when a rule needs a real check behind it.

Topics covered

• AGENTS.md
• CLAUDE.md
• coding-agents
• monorepo

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-coding/agents-md-and-repo-instructions

6 min · foundational · ai-coding

Getting code from an agent is easy. Getting code you can trust takes a way of working: explore, plan, implement, verify and review, with small diffs and a spec first.

Topics covered

• coding-agents
• Claude-Code
• Codex
• spec-driven-development

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-coding/working-with-coding-agents

6 min · intermediate · Agentic Systems

Stop growing your system prompt. How SKILL.md folders and progressive disclosure work, how to write a description that fires, and when a skill beats a tool or an MCP server.

Topics covered

• agent-skills
• SKILL.md
• agents
• progressive-disclosure

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/agentic-systems/agent-skills

5 min · advanced · Agentic Systems

Checkpoints do not make a side effect safe. Idempotency keys, the interrupt replay trap, choosing a runtime, and the isolation and egress rules that contain generated code.

Topics covered

• durable-execution
• sandboxing
• LangGraph
• idempotency

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/agentic-systems/durable-agents-and-sandboxing

6 min · advanced · Agentic Systems

Long-running agents fail in the code around the model. Clean sessions, progress on disk, a task list that starts failing, and checks the agent cannot talk its way past.

Topics covered

• harness-engineering
• agents
• long-running-agents
• verification

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/agentic-systems/harness-engineering

6 min · intermediate · Prompt Engineering

Why agents get worse as they get more context, and what to do about it: budgets by source, cleared tool results, a cache-friendly layout and compaction that keeps your decisions.

Topics covered

• context-engineering
• agents
• prompt-caching
• compaction

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/prompt-engineering/context-engineering

5 min · advanced · AI Engineering

Speech-to-speech or a pipeline, a latency budget for every stage, interruptions that really work, prompts written for the ear, and what a call actually costs.

Topics covered

• voice-agents
• realtime
• latency
• streaming

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-engineering/voice-agents

5 min · advanced · Agentic Systems

Agent Cards, long-running tasks and artifacts, how A2A differs from MCP, and the honest test for when a plain function call is the better answer.

Topics covered

• A2A
• agent-protocols
• multi-agent
• interoperability

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/agentic-systems/a2a-protocol

5 min · advanced · Security Engineering

ASI01 to ASI10 explained for engineers: least agency, the lethal trifecta, per-agent identity, and policy gates in code that injected text cannot argue with.

Topics covered

• OWASP
• agent-security
• prompt-injection
• least-agency

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/security-engineering/owasp-agentic-top-10

5 min · intermediate · AI Observability

Agents fail with healthy infrastructure. One trace per run, OpenTelemetry GenAI spans on a schema that is still moving, cost per task, and evals that catch what dashboards miss.

Topics covered

• agent-observability
• OpenTelemetry
• tracing
• Langfuse

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-observability/agent-observability

11 min · foundational · LLM Model Landscape

The GPT-6 tiers, what reasoning effort replaced, when to use the Responses API over Chat Completions, how the bill is calculated, and what is different on Azure.

Topics covered

• GPT-6
• OpenAI
• reasoning-effort
• Responses-API
• model-selection

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/llm-landscape/gpt-model-family

11 min · foundational · LLM Model Landscape

The four Claude tiers, adaptive thinking and effort, the tool-use rules that differ from other providers, prompt caching, and where you can run it.

Topics covered

• Claude
• Anthropic
• adaptive-thinking
• prompt-caching
• tool-use

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/llm-landscape/claude

11 min · intermediate · LLM Model Landscape

The Llama lineup and licence, serving with Ollama and vLLM, GPU memory sizing, and an honest look at when a hosted API is still the better choice.

Topics covered

• Llama
• open-weight
• self-hosting
• vLLM
• quantization

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/llm-landscape/llama

11 min · intermediate · AI Observability

Promptfoo, DeepEval, RAGAS, LangSmith and Phoenix compared by what each is good at, with working examples and a way to choose.

Topics covered

• evaluation
• Promptfoo
• DeepEval
• RAGAS
• LangSmith

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-observability/evaluation-frameworks

11 min · intermediate · AI Observability

Build an evaluation dataset from real use, pick the cheapest grader that works, measure the noise floor, and turn the suite into a release gate.

Topics covered

• offline-evals
• regression-testing
• datasets
• LLM-as-judge
• CI

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-observability/offline-evals

10 min · intermediate · LLM Model Landscape

Why public benchmarks do not transfer to your workload, how to build a model-selection eval, and why cost per successful task beats price per token.

Topics covered

• evals
• model-selection
• benchmarks
• cost-per-task
• LLM

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/llm-landscape/evals

11 min · intermediate · LLM Model Landscape

The three kinds of hallucination, a test set that includes questions with no answer, claim-by-claim faithfulness scoring, and what lowers the rate.

Topics covered

• hallucination
• faithfulness
• grounding
• RAG
• evaluation

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/llm-landscape/hallucination-testing

8 min · intermediate · LLM Model Landscape

What fits in device memory, why bandwidth sets generation speed, which runtime to use per platform, and how to ship quantised weights with a cloud fallback that keeps the privacy promise.

Topics covered

• edge AI
• on-device
• SLM
• quantization
• llama.cpp
• WebLLM

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/llm-landscape/edge-models

9 min · advanced · LLM Model Landscape

Break-even maths, GPU sizing from weights and KV cache, choosing between vLLM, SGLang and llama.cpp, and what to scale and alert on once open-weight models carry production traffic.

Topics covered

• self-hosting
• vLLM
• inference
• GPU
• KV cache
• Kubernetes

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/llm-landscape/self-hosted-models

9 min · advanced · Prompt Engineering

From a one-off rewrite to refinement loops on failures, candidate search and compiled DSPy programs. How to optimise prompts against a metric without overfitting, and how to review what the optimiser produces.

Topics covered

• meta-prompting
• prompt optimisation
• DSPy
• APE
• OPRO
• evaluation

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/prompt-engineering/meta-prompting

9 min · advanced · Prompt Engineering

Threat model first, then manual sessions, then automated scans with garak, promptfoo and PyRIT. How to find jailbreaks, data leaks and unsafe tool calls, fix them in the architecture, and keep them fixed in CI.

Topics covered

• red teaming
• security
• jailbreak
• prompt injection
• promptfoo
• garak

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/prompt-engineering/red-teaming

10 min · advanced · Prompt Engineering

Propose several next steps, judge the partial states, keep the best and expand them. How to build a Tree-of-Thought search, why the evaluator decides everything, and when a reasoning model makes it unnecessary.

Topics covered

• tree-of-thought
• ToT
• reasoning
• search
• beam search
• prompt-engineering

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/prompt-engineering/tree-of-thought

13 min · foundational · TypeScript Deep Dive

Compiler options, strict mode, target vs lib, project references, and a pragmatic strategy for adopting strict mode on a legacy codebase.

Topics covered

• tsconfig
• strict-mode
• compilerOptions
• paths
• project-references

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/typescript/tsconfig

12 min · advanced · TypeScript Deep Dive

String manipulation at the type level — pattern matching, type-safe routes, and event name validation.

Topics covered

• template-literals
• string-types
• pattern-matching
• Capitalize
• type-safe-routes

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/typescript/template-literals

12 min · intermediate · TypeScript Deep Dive

Node, Node16, Bundler — understand how TypeScript actually finds and resolves your import paths, and why it breaks at runtime anyway.

Topics covered

• module-resolution
• Node16
• Bundler
• package-exports
• paths

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/typescript/module-resolution

13 min · advanced · TypeScript Deep Dive

TC39 Stage 3 decorators vs legacy experimentalDecorators — class, method, and field decorators for metadata and cross-cutting concerns.

Topics covered

• decorators
• TC39
• metadata
• class-decorator
• experimentalDecorators

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/typescript/decorators

13 min · intermediate · TypeScript Deep Dive

.d.ts files and @types — type definitions for JavaScript libraries, ambient declarations, and module augmentation.

Topics covered

• declaration-files
• .d.ts
• types
• DefinitelyTyped
• module-augmentation

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/typescript/declaration-files

16 min · intermediate · TypeScript Deep Dive

asserts condition — functions that narrow types by throwing, for fail-fast validation patterns at function boundaries.

Topics covered

• assertion-functions
• asserts
• invariant
• fail-fast
• type-narrowing

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/typescript/assertion-functions

13 min · advanced · Prompt Engineering

Enforce output structure at the token level with grammars and token masks — how it differs from native JSON mode, and when the strictness costs you reasoning quality.

Topics covered

• constrained-generation
• grammar
• token-masking
• Outlines
• structured-output

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/prompt-engineering/constrained-generation

11 min · foundational · LLM Model Landscape

Google's natively multimodal model family — the Pro/Flash trade-off, Search grounding, built-in code execution, and Vertex AI for enterprise deployment.

Topics covered

• Gemini
• Google-DeepMind
• multimodal
• grounding
• Vertex-AI

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/llm-landscape/gemini

13 min · intermediate · LLM Model Landscape

Frontier vs mini-tier trade-offs, model routing and cascading architectures, batch pricing, and a worked cost model for a real production feature.

Topics covered

• cost-performance
• pricing
• model-routing
• cascading
• batch-API

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/llm-landscape/cost-performance

14 min · foundational · LLM Model Landscape

Context length limits, the 'lost in the middle' effect, prompt caching, and when a huge context window is the wrong fix for a retrieval problem.

Topics covered

• context-window
• token-limit
• lost-in-the-middle
• context-compression
• chunking

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/llm-landscape/context-windows

16 min · foundational · LLM Model Landscape

OpenAI, Anthropic, Google, Azure, AWS Bedrock, Together, Groq — compare direct, cloud-hosted, and inference-as-a-service API providers on pricing, latency, and enterprise fit.

Topics covered

• API-providers
• OpenAI-API
• Anthropic-API
• Bedrock
• Groq

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/llm-landscape/api-providers

12 min · advanced · DevOps & CI/CD

Sidecar proxies for traffic management, mTLS, observability, and resilience — what a service mesh actually buys you, and when the operational cost is worth it.

Topics covered

• service-mesh
• Istio
• Linkerd
• sidecar-proxy
• mTLS

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/devops-cicd/service-mesh

12 min · intermediate · DevOps & CI/CD

Kubernetes's package manager — chart structure, values.yaml, Go templates, dependencies, and release management (install, upgrade, rollback) for real deployments.

Topics covered

• Helm
• charts
• values
• templates
• release-management

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/devops-cicd/helm

11 min · intermediate · DevOps & CI/CD

Store, scan, and distribute container images — immutable tags, vulnerability scanning policies, geo-replication, and retention rules that keep a registry trustworthy.

Topics covered

• container-registry
• ACR
• Docker-Hub
• GHCR
• image-scanning

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/devops-cicd/container-registry

12 min · intermediate · DevOps & CI/CD

Actionable alerts, on-call rotations, and incident response — alert rules, severity, routing, and the fatigue-prevention discipline that keeps on-call sane.

Topics covered

• alerting
• PagerDuty
• Alertmanager
• on-call
• incident-response

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/devops-cicd/alerting

12 min · intermediate · AI Observability

Monitor token usage and costs — per-request tracking, budgets, cost allocation across teams, and the optimization strategies that shrink the bill.

Topics covered

• token-tracking
• cost-allocation
• token-budget
• usage-metrics
• FinOps

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-observability/token-tracking

14 min · intermediate · AI Observability

Arize's open-source, OTel-native LLM observability stack — traces, eval workflows, and prompt experiments, self-hosted with no data leaving your VPC.

Topics covered

• Phoenix
• Arize
• open-source
• RAG-evaluation
• tracing

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-observability/phoenix

14 min · intermediate · AI Observability

Validate LLM outputs — schema conformance, factuality checks against source documents, toxicity detection, and structured output parsing before serving users.

Topics covered

• output-validation
• schema-validation
• factuality
• toxicity
• hallucination-detection

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-observability/output-validation

14 min · intermediate · AI Observability

Instrument LLM apps with OpenTelemetry — the gen-ai semantic conventions, distributed tracing across RAG pipelines, and vendor-neutral export to any backend.

Topics covered

• OpenTelemetry
• gen-ai-semantic-conventions
• distributed-tracing
• OTEL
• vendor-neutral

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-observability/opentelemetry-ai

12 min · intermediate · AI Observability

Production-time evaluation — implicit user feedback, A/B testing, and real-time LLM-as-judge scoring for continuous quality monitoring after ship.

Topics covered

• online-evals
• A/B-testing
• user-feedback
• production-monitoring
• quality-score

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-observability/online-evals

12 min · intermediate · AI Observability

Production dashboards for LLM apps — latency, error rates, token usage, and quality drift detection, wired into the alerting you already trust.

Topics covered

• LLM-monitoring
• dashboards
• drift-detection
• alerting
• APM

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-observability/llm-monitoring

11 min · intermediate · AI Observability

Measure and optimize LLM response times — Time to First Token, tokens per second, P99 tail latency, and the streaming/caching levers that actually move the needle.

Topics covered

• latency
• TTFT
• tokens-per-second
• P99
• streaming

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-observability/latency-analysis

13 min · intermediate · AI Observability

LangChain's observability platform — distributed tracing, dataset curation, evaluation suites, and the prompt playground, for LangChain and framework-agnostic apps alike.

Topics covered

• LangSmith
• tracing
• LangChain
• observability
• evaluation

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-observability/langsmith

12 min · intermediate · AI Observability

Validate and sanitize LLM inputs — prompt injection detection, PII filtering, and content moderation that run before the model call to prevent misuse and cut cost.

Topics covered

• input-validation
• prompt-injection
• PII-filtering
• content-moderation
• guardrails

👉 Read the full article on AI Wisdom →

Originally published at https://www.aiwisdom.dev/articles/ai-observability/input-validation