Kubai Kevin

Older Posts — Page 9 of 19

Buy vs build AI: 3 hard rules for 2026

A framework for deciding which AI components to build versus buy, with failure modes, measurement methods, and code for caching and retries.

15 min read · 11 August 2026

Agent SLOs: measure what agents actually do

Latency and 5xx SLOs miss stuck loops. Define outcome SLOs on completion rate, with schema, PromQL, and failure handling.

13 min read · 11 August 2026

Orchestration tax kills multi-agent systems

Why coordination layers dominate multi-agent latency and memory, with patterns that remove the scheduler and how to measure your own orchestration tax.

13 min read · 11 August 2026

Observability Agents as Application Code

Why telemetry agents break at scale, how to budget their resources, and when the standard install-and-forget playbook is actually fine.

13 min read · 11 August 2026

Agent followed instructions too literally: the cron job trap

Why agents that follow retry instructions literally hang forever, and how to fix it with per-request timeouts, async retries, and DNS limits.

12 min read · 11 August 2026

AWS CloudWatch Synthetics for AI pipelines

Build CloudWatch Synthetics canaries that exercise a full AI pipeline journey, not just an endpoint ping, and alarm on whole-flow failure.

13 min read · 10 August 2026

Agent frameworks: stop rewriting the same glue code

Agent glue code — retries, credentials, state — is mostly boilerplate. Here are three abstractions, worked examples, and how to measure the savings.

13 min read · 10 August 2026

Designing APIs for High-Latency Satellite Links

How to keep an API reliable over high-latency satellite links: circuit breakers, retry budgets, failover gateways, and the metrics that show what broke.

14 min read · 10 August 2026

Least privilege for agents: the config line nobody writes

Why permissive agent tool defaults break in production, and how to add runtime, context-aware policy checks before every tool call.

14 min read · 10 August 2026

eBPF observability in 2026: what actually works

A practical guide to production eBPF observability and runtime security: kernel requirements, verifier failures, ring buffers, pinning, and how to measure overhead.

14 min read · 9 August 2026

Why your multi-agent system degraded slowly

Multi-agent systems can lose p99 responsiveness while every agent stays healthy. Here is how payload creep, queueing and retries compound, and how to measure it.

14 min read · 9 August 2026

Offline-first money apps with DynamoDB Streams

Design an offline-first mobile-money sync pipeline on DynamoDB Streams: idempotency keys, conditional balance writes, DLQ handling, and how to measure it.

15 min read · 8 August 2026

AI cost attribution without the spreadsheet hell

How to attribute AI inference costs to users, features and models: instrumentation, cardinality limits, sampling error, and a 30-minute first step.

15 min read · 7 August 2026

Agents create incidents at 2am

Why autonomous agents that can open, mutate, or page on incidents cause 2am page storms, and how write boundaries and idempotent alerting fix it.

12 min read · 6 August 2026

Cost-Effective AI Infrastructure for Micro-SaaS

How to architect an AI micro-SaaS on a tight budget: caching, batching, serverless scaling, cost measurement, and the failure modes that break naive designs.

15 min read · 5 August 2026

Mobile money agent fees: why your bot keeps failing

USSD session timeouts, not payment errors, break mobile money bots in production. Diagnose the session contract mismatch and fix it with per-request clients.

13 min read · 5 August 2026

Stopping a runaway polling agent from flooding your API

Diagnose and fix runaway agent API calls: bounded retries, capped backoff, idempotency keys, and volume alerts that catch spikes before the bill does.

15 min read · 5 August 2026

WASM servers beat containers here

A practical guide to choosing WebAssembly over containers for short-lived, CPU-bound server tasks, with code, failure modes, and a measurement plan.

12 min read · 5 August 2026

Senior role projects: 3 real paths

Build three portfolio projects that demonstrate tenant isolation, retry-storm handling, and event sourcing, with concrete ways to measure each claim.

16 min read · 5 August 2026

MCP servers: the 3 hidden costs nobody quotes

MCP servers are stateful async processes, not stateless APIs. Here are the real cost drivers, failure modes, and how to measure them yourself.

16 min read · 5 August 2026

RAG rots if you don’t refresh indexes

RAG indexes rot when source documents change. Learn change detection, chunk versioning, and incremental refresh patterns, with code and a cost model.

15 min read · 4 August 2026

Kubernetes or Nomad: the real experiment cost split

A practical comparison of Kubernetes and Nomad for GPU training jobs, covering scheduler latency, cost modelling, GPU sharing, and a decision checklist.

11 min read · 3 August 2026

Supervisor vs swarm vs debate vs pipeline

Compare four multi-agent orchestration patterns: supervisor, swarm, debate, and pipeline. Covers failure modes, backpressure, retries, and how to measure each.

19 min read · 3 August 2026

Prompt Injection Is Not Your Biggest LLM Data Risk

Prompt injection dominates LLM security checklists, but poisoned training data and label drift often cause more damage. A pipeline-stage model for triage.

17 min read · 3 August 2026