BLOG / ENGINEERING
AI engineering articles, written by engineers.
Production-focused writing on RAG, AI agents, MCP, LLM evaluation and shipping AI systems to production. No news roundups, no link dumps, just engineering substance.
Start here
- New to always-on agentsDots vs Muse vs Grok Bot vs Gemini Spark: Best Always-On AI Agent in 2026OpenAI Dots, Meta Muse, xAI Grok Bot and Google Gemini Spark compared on runtime, security, approvals, integrations and price, with a pick for every use case. 11 min read
- New to MCP and tool useWhat is MCP? Model Context Protocol Explained for AI EngineersMCP is the open standard that lets any AI agent use any tool server. How it works, its three primitives, transports, security and when plain function calls win. 11 min read
- Building a RAG systemProduction RAG Architecture: A Complete GuideThe full production RAG architecture: ingestion, chunking, embeddings, hybrid search, reranking, generation, evaluation and observability, with code. 14 min read
- Your AI bill is growingHow to Reduce LLM Costs by 90% Without Hurting QualitySemantic caching, model routing and context compression compound. How each lever works, how to apply it safely, the maths behind 90% and where teams go wrong. 11 min read
Free and open source
Free AI skills
Open-source skills for Claude Code, Codex, Cursor and GitHub Copilot, each with a guide and a real example run.
- Claude Code
- Codex
- Cursor
- GitHub Copilot
agent-approval-policy-reviewReview which agent tool calls should run alone and which need a human. 9 min guide with a real example runbusiness-case-builderBuild a funding-ready AI business case, with no invented numbers. 10 min guide with a real example runfde-interview-coachPractise FDE interviews with timed drills and rubric scoring. 9 min guide with a real example run
FeaturedProduction AIIntermediate
The 90-Day Enterprise AI Adoption Playbook
Days 0-30 foundation, 31-60 controlled scale, 61-90 production adoption, plus a maturity model to tell you honestly which level you're actually at.
By Bhavin Mistry· Sep 11, 2026 15 min readRead article
- Featured · Production AIInside the AI-Augmented Developer WorkflowOne ticket, start to finish, through an AI-augmented development workflow: where the agent helps, where it stalls and where a human has to step in.
- Featured · Production AIThe AI-Native SDLC: What AI Does and What Humans DecideEvery stage of the software lifecycle rebuilt around AI: what AI now does, what changes and what stays human-controlled at each of twelve stages.
Browse by topic
Every article on a topic in one place, with the chapter, lab or roadmap to go further.
Agent skills · 3 articles
Free skills for Claude Code, Codex, Cursor and Copilot, with real example runs.
Always-on agents · 4 articles
OpenAI Dots, Meta Muse, xAI Grok Bot and how to build or govern your own.
- Dots vs Muse vs Grok Bot vs Gemini Spark: Best Always-On AI Agent in 2026
- Forward Deployed Engineer in 2026: Skills, Role and Why Always-On Agents Need FDEs
- What Is Grok Bot? xAI's Always-On AI Teammates Explained (Pricing, Routines, Risks)
- What Are OpenAI Dots? Always-On ChatGPT Agents Explained (2026)
AI agents in production · 12 articles
Loops, approvals, identity and the guardrails that keep agents safe.
- Review an AI Agent's Approval Policy in 10 Minutes with an AI SkillFree skill
- Prompt Injection, Explained: Attacks and Defences That Hold
- When Should an AI Agent Ask Permission? Human-in-the-Loop Approval Design
- AI Agent Identity: Why Every Agent Needs Its Own Credentials (Not Your Password)
- All 12 articles on AI agents in production
RAG and retrieval · 5 articles
From chunking to reranking, and why good retrieval can still give bad answers.
MCP and tool use · 5 articles
The open protocol for connecting models to tools, data and prompts.
Evaluation · 2 articles
Golden sets, faithfulness, LLM-as-judge and regression gates.
Cost and production · 4 articles
Caching, routing and reliability, and what to measure before you scale.
Enterprise AI engineering · 5 articles
Blueprint, guardrails, SDLC and a 90-day adoption playbook.
Forward deployed engineering · 5 articles
Shipping AI inside customer organisations, and the career behind it.
AI platform engineering · 4 articles
Gateways, serving, observability and the shared plumbing every AI team needs.
AI leadership and strategy · 3 articles
Measuring impact, governing AI and deciding what to fund, scale or stop.
Latest articles
38 publishedFDE· IntermediateFree skill
Build an AI Business Case with a Skill That Won't Invent Numbers
A free AI skill that builds a funding-ready business case with options, an assumptions register and best, base and worst cases, and refuses to make up numbers.
Oct 14, 2026 10m
business caseAI ROIagent skills
FDE· FoundationsFree skill
Practise Forward Deployed Engineer Interviews with an AI Coach Skill
A free AI skill that runs timed FDE drills, client role-plays and deployment-design rounds and scores you against a rubric. Includes a real coaching session.
Oct 14, 2026 9m
forward deployed engineerFDE interviewagent skills
AI Agents· IntermediateFree skill
Review an AI Agent's Approval Policy in 10 Minutes with an AI Skill
Paste your agent's tool list into a free AI skill and get a risk-tiered approval review, ranked findings and tests, with a real example run and install steps.
Oct 14, 2026 9m
human in the loopAI agent approvalsagent safety
Production AI· Intermediate
AI Strategy for Engineering Leaders: Choose, Fund and Stop
How engineering leaders decide which AI projects to back: a portfolio view, stage gates, stop criteria and evidence-based reporting, with reusable scorecards.
Oct 11, 2026 9m
AI strategyengineering leadershipAI portfolio
Production AI· Intermediate
An AI Governance Framework for Engineering Teams
A lightweight AI governance framework for engineering teams: inventory, risk tiers, approvals, evaluation gates, incidents and a review cadence.
Oct 11, 2026 11m
AI governanceengineering leadershiprisk tiers
Production AI· Intermediate
API or Local Models: Where the Cost Break-Even Sits
When does running a model yourself beat paying for an API? A total-cost framework, an illustrative break-even calculation and the non-cost factors that decide.
Oct 11, 2026 9m
API vs localself-hostingopen-weight models
Architecture· Foundations
Context Engineering vs Prompt Engineering: What Changes
What context engineering is and how it differs from prompt engineering: what goes into context, budgets, ordering, compaction and how to evaluate it.
Oct 11, 2026 9m
context engineeringprompt engineeringcontext window
RAG· Foundations
Embeddings and Vector Search, Explained for Engineers
How embeddings and vector search work: similarity, nearest-neighbour indexes, filters, hybrid search and the version trap, explained step by step for engineers.
Oct 11, 2026 10m
embeddingsvector searchsemantic search
FDE· Intermediate
How to Build the Business Case for an AI Project
How to build a credible AI business case: costs, value, adoption ramp and best, base and worst cases, with a clearly illustrative worked example.
Oct 11, 2026 10m
business caseAI ROIFDE
Evaluation· Advanced
How to Evaluate AI Agents: Beyond Final-Answer Accuracy
How to evaluate AI agents that take many steps: outcome, trajectory, tool use, policy adherence and cost, with a task-based golden set and regression checks.
Oct 11, 2026 10m
agent evaluationAI agentstrajectory evaluation
Production AI· Intermediate
How to Measure the Impact of AI Coding Tools on Your Team
How to tell whether AI coding tools are working: measure delivery, quality and developer experience against a baseline, and avoid activity metrics that mislead.
Oct 11, 2026 10m
AI coding toolsengineering productivitydeveloper experience
Architecture· Advanced
LLM Gateways: What They Do and How to Choose One
What an LLM gateway does, when you need one and how to choose: authentication, routing, budgets, caching, logging and policy, with a vendor-neutral scorecard.
Oct 11, 2026 9m
LLM gatewayAI platformmodel routing
Production AI· Intermediate
LLM Observability: What to Trace and Why
What to trace and log in a production LLM app: prompts, retrieved context, tool calls, tokens, cost and latency, with sampling, privacy and alerting guidance.
Oct 11, 2026 9m
LLM observabilitytracingAI platform
AI Agents· Intermediate
Prompt Injection, Explained: Attacks and Defences That Hold
How prompt injection works and which defences hold: direct and indirect attacks, why filters fail, least privilege, approvals and red-team tests.
Oct 11, 2026 10m
prompt injectionLLM securityAI agents
Architecture· Foundations
Prompting, RAG or Fine-Tuning: A Decision Guide
Which technique first? A decision guide for prompting, RAG and fine-tuning: what each changes, what it costs to maintain and how to decide with an evaluation.
Oct 11, 2026 10m
fine-tuningRAGprompt engineering
FDE· Foundations
Scoring AI Use Cases Before You Build Anything
How to choose a first AI use case: score value, data readiness, feasibility, risk and sponsor on one page, then test it with a short proof of concept.
Oct 11, 2026 9m
use case scoringFDEAI discovery
Architecture· Intermediate
What an AI Platform Engineer Actually Owns
What an AI platform team builds and runs: gateway, serving, evaluation pipelines, observability, cost and security, and where product teams take over.
Oct 11, 2026 9m
AI platformplatform engineeringAI architecture
AI Agents· Intermediate
When Should an AI Agent Ask Permission? Human-in-the-Loop Approval Design
When should an autonomous agent stop and ask? A risk-tiered approval framework drawn from Muse, Dots and Grok Bot, with policy-as-code you can reuse.
Oct 7, 2026 10m
human in the loopAI agent approvalsagent safety
AI Agents· Advanced
AI Agent Identity: Why Every Agent Needs Its Own Credentials (Not Your Password)
Always-on agents need their own identities and scoped credentials, not your password. How Dots, Muse and Grok Bot handle secrets, and a pattern to copy.
Oct 7, 2026 10m
AI agent identitynon-human identityleast privilege
AI Agents· Foundations
Dots vs Muse vs Grok Bot vs Gemini Spark: Best Always-On AI Agent in 2026
OpenAI Dots, Meta Muse, xAI Grok Bot and Google Gemini Spark compared on runtime, security, approvals, integrations and price, with a pick for every use case.
Oct 7, 2026 11m
always-on agentsOpenAI DotsMeta Muse
AI Agents· Intermediate
How to Build Your Own 24/7 Personal AI Agent in 2026 (Hosted or DIY)
Build a personal AI agent that runs 24/7: choose hosted (Dots, Muse, Grok Bot) or DIY, then wire up scheduling, memory, MCP tools, approvals and an audit log.
Oct 7, 2026 13m
build AI agentautonomous agent24/7 AI agent
AI Agents· Advanced
How to Deploy AI Agents in the Enterprise: A 2026 Governance Playbook
How enterprises roll out always-on AI agents safely: first use cases, an identity per agent, approval tiers, SIEM audit logs, evals and a 90-day pilot plan.
Oct 7, 2026 12m
enterprise AI agentsAI agent governanceAI agent deployment
FDE· Intermediate
Forward Deployed Engineer in 2026: Skills, Role and Why Always-On Agents Need FDEs
What a Forward Deployed Engineer does in 2026, the FDE skill stack, why always-on agents like Dots and Grok Bot need FDEs, and how to train your team.
Oct 7, 2026 10m
Forward Deployed EngineerFDEFDE skills
AI Agents· Foundations
What Is Grok Bot? xAI's Always-On AI Teammates Explained (Pricing, Routines, Risks)
Grok Bot gives xAI's agents a cloud computer, skills learned from screen recordings and scheduled routines. How it works, what it costs, and the shared-VM risk.
Oct 7, 2026 9m
Grok BotxAIAI teammates
AI Agents· Intermediate
Meta Muse AI Agent Explained: How It Works, Pricing and the Sentinel Safety Model
Meta's Muse runs in a per-user Secure VM while a separate Sentinel agent approves every action. How it works, pricing, and how to copy the guard-agent pattern.
Oct 7, 2026 10m
Meta MuseMuse SparkSentinel agent
AI Agents· Intermediate
OpenAI DevDay 2026 for Agent Builders: Agents API Computer Use, GPT-6.1 Sol, MCP Events
The OpenAI DevDay 2026 announcements that matter for agent builders: Agents API computer use, GPT-6.1 Sol at one-fifth the price, MCP Events and Dots.
Oct 7, 2026 9m
OpenAI DevDay 2026Agents APIcomputer use
AI Agents· Foundations
What Are OpenAI Dots? Always-On ChatGPT Agents Explained (2026)
OpenAI Dots are always-on ChatGPT agents with their own cloud computer. How they work, pricing, approval rules, specialist Dots, and how to set one up.
Oct 7, 2026 9m
OpenAI DotsChatGPT agentGPT-6 Astra
Production AI· Intermediate
Inside the AI-Augmented Developer Workflow
One ticket, start to finish, through an AI-augmented development workflow: where the agent helps, where it stalls and where a human has to step in.
Sep 11, 2026 13m
enterprise-ai-engineering-seriescoding-agentsdeveloper-workflow
Production AI· Advanced
The AI-Native SDLC: What AI Does and What Humans Decide
Every stage of the software lifecycle rebuilt around AI: what AI now does, what changes and what stays human-controlled at each of twelve stages.
Sep 11, 2026 15m
enterprise-ai-engineering-seriessdlcdevsecops
Architecture· Advanced
The Enterprise AI Engineering Blueprint
A reference architecture for enterprise AI: gateway, orchestration, RAG, MCP tools, identity, guardrails, evaluation, observability and governance.
Sep 11, 2026 16m
enterprise-ai-engineering-seriesarchitectureai-gateway
Production AI· Advanced
Enterprise AI Guardrails: A Ten-Layer Defense in Depth
A ten-layer guardrail model for enterprise AI: prompt injection, least-privilege tools, approved models, audit trails, kill switches and MCP security.
Sep 11, 2026 17m
enterprise-ai-engineering-seriesguardrailssecurity
AI Agents· Advanced
AI Agent Loops: Stopping Conditions and Guardrails for Production
Why agents loop forever and the guardrails that stop them: iteration limits, cost budgets, tool-call deduplication, progress checks and clear 'done' criteria.
Sep 4, 2026 12m
agentsguardrailsReAct
Production AI· Intermediate
How to Reduce LLM Costs by 90% Without Hurting Quality
Semantic caching, model routing and context compression compound. How each lever works, how to apply it safely, the maths behind 90% and where teams go wrong.
Sep 4, 2026 11m
costcachingmodel routing
Evaluation· Advanced
LLM Evaluation Fundamentals for Production Systems
How to evaluate a production LLM system: golden sets from real traffic, retrieval and generation metrics, calibrated LLM judges, CI gates and online monitoring.
Sep 4, 2026 12m
evaluationfaithfulnessLLM-as-judge
RAG· Advanced
Production RAG Architecture: A Complete Guide
The full production RAG architecture: ingestion, chunking, embeddings, hybrid search, reranking, generation, evaluation and observability, with code.
Sep 4, 2026 14m
RAGarchitectureproduction
MCP· Intermediate
What is MCP? Model Context Protocol Explained for AI Engineers
MCP is the open standard that lets any AI agent use any tool server. How it works, its three primitives, transports, security and when plain function calls win.
Sep 4, 2026 11m
MCPtoolsagents
RAG· Intermediate
Why Good Retrieval Still Produces Bad Answers
Debug RAG failures with a runnable Python exercise that separates retrieval misses, dropped evidence, unsupported claims and incorrect refusals.
Sep 4, 2026 8m
RAGretrievalreranking
Follow new articles
One useful AI engineering lesson in your inbox every week. No spam, unsubscribe anytime.