Ahmed Fareed
Blog

Notes on shipping AI that works.

Short, opinionated writing on AI application engineering — agents, reliability, and the unglamorous work that makes AI dependable.

"Evals for RAG: test retrieval and generation separately"
RAG
  • by Ahmed
  • 18 Sept

"Evals for RAG: test retrieval and generation separately"

"The agent loop, dissected: plan, act, observe — and when to stop"
Agents
  • by Ahmed
  • 17 Sept

"The agent loop, dissected: plan, act, observe — and when to stop"

"Shadow mode: running your AI feature in production without letting it act"
AI engineering
  • by Ahmed
  • 16 Sept

"Shadow mode: running your AI feature in production without letting it act"

"What \"AI agent\" actually means — and what it doesn't"
AI agents
  • by Ahmed
  • 15 Sept

"What \"AI agent\" actually means — and what it doesn't"

"MCP servers: giving models safe access to your systems"
MCP
  • by Ahmed
  • 14 Sept

"MCP servers: giving models safe access to your systems"

"What running AI systems for my own business taught me about client work"
AI engineering
  • by Ahmed
  • 11 Sept

"What running AI systems for my own business taught me about client work"

"Build vs buy for AI features: a decision framework"
build-vs-buy
  • by Ahmed
  • 10 Sept

"Build vs buy for AI features: a decision framework"

"Webhooks, queues and LLMs: wiring AI into systems that already exist"
Integration
  • by Ahmed
  • 9 Sept

"Webhooks, queues and LLMs: wiring AI into systems that already exist"

"Keeping the knowledge base fresh: the update pipeline nobody plans for"
RAG
  • by Ahmed
  • 8 Sept

"Keeping the knowledge base fresh: the update pipeline nobody plans for"

"Prompt caching: the cost lever most teams ignore"
llm
  • by Ahmed
  • 7 Sept

"Prompt caching: the cost lever most teams ignore"

"Measuring quality drift in production: the metrics that move first"
Monitoring
  • by Ahmed
  • 4 Sept

"Measuring quality drift in production: the metrics that move first"

"Multi-agent systems: when two agents are worse than one"
Agents
  • by Ahmed
  • 3 Sept

"Multi-agent systems: when two agents are worse than one"

"Deploying AI features safely: feature flags, shadow mode and staged rollout"
AI engineering
  • by Ahmed
  • 2 Sept

"Deploying AI features safely: feature flags, shadow mode and staged rollout"

"Prompting, RAG or fine-tuning? A plain-English guide"
AI basics
  • by Ahmed
  • 1 Sept

"Prompting, RAG or fine-tuning? A plain-English guide"

"Structured outputs and function calling: the reliable core of LLM integration"
AI engineering
  • by Ahmed
  • 31 Aug

"Structured outputs and function calling: the reliable core of LLM integration"

"The confident wrong answer that cost trust, not money"
AI engineering
  • by Ahmed
  • 28 Aug

"The confident wrong answer that cost trust, not money"

"Hiring an AI engineer: questions that separate builders from demo-makers"
hiring
  • by Ahmed
  • 27 Aug

"Hiring an AI engineer: questions that separate builders from demo-makers"

"CRM enrichment with AI: cheap, useful, and easy to get wrong"
CRM
  • by Ahmed
  • 26 Aug

"CRM enrichment with AI: cheap, useful, and easy to get wrong"

"Your retrieval is the problem, not your model"
RAG
  • by Ahmed
  • 25 Aug

"Your retrieval is the problem, not your model"

"Small models for big workloads: when cheap wins"
llm
  • by Ahmed
  • 24 Aug

"Small models for big workloads: when cheap wins"

"Regression testing prompts: catching the fix that breaks everything else"
Evals
  • by Ahmed
  • 21 Aug

"Regression testing prompts: catching the fix that breaks everything else"

"Guardrails for agents that touch real systems"
Agents
  • by Ahmed
  • 20 Aug

"Guardrails for agents that touch real systems"

"Human handoff done right: design the escalation path before you need it"
AI engineering
  • by Ahmed
  • 19 Aug

"Human handoff done right: design the escalation path before you need it"

"Hallucination isn't a bug you can fix — it's a property you design around"
AI basics
  • by Ahmed
  • 18 Aug

"Hallucination isn't a bug you can fix — it's a property you design around"

"A production AI stack in Python: FastAPI and LangGraph patterns that hold up"
AI engineering
  • by Ahmed
  • 17 Aug

"A production AI stack in Python: FastAPI and LangGraph patterns that hold up"

"The model update that changed everything"
AI engineering
  • by Ahmed
  • 14 Aug

"The model update that changed everything"

"What an AI feature really costs: a budgeting guide"
budgeting
  • by Ahmed
  • 13 Aug

"What an AI feature really costs: a budgeting guide"

"Email automation with LLMs: classification, drafting and the trust ladder"
Email automation
  • by Ahmed
  • 12 Aug

"Email automation with LLMs: classification, drafting and the trust ladder"

"Chunking is a product decision, not a preprocessing step"
RAG
  • by Ahmed
  • 11 Aug

"Chunking is a product decision, not a preprocessing step"

"Streaming, caching and other ways to make AI feel fast"
llm
  • by Ahmed
  • 10 Aug

"Streaming, caching and other ways to make AI feel fast"

"LLM-as-judge: useful, noisy, and how to keep it honest"
Evals
  • by Ahmed
  • 7 Aug

"LLM-as-judge: useful, noisy, and how to keep it honest"

"How much memory does an agent actually need?"
Agents
  • by Ahmed
  • 6 Aug

"How much memory does an agent actually need?"

"Fail closed, not silent: what an AI system should do when it isn't sure it worked"
AI engineering
  • by Ahmed
  • 5 Aug

"Fail closed, not silent: what an AI system should do when it isn't sure it worked"

"Tokens, context windows and why your AI forgets things"
AI basics
  • by Ahmed
  • 4 Aug

"Tokens, context windows and why your AI forgets things"

"A production AI stack in TypeScript: what I actually use"
AI engineering
  • by Ahmed
  • 3 Aug

"A production AI stack in TypeScript: what I actually use"

"The automation that \"worked\" for three weeks: a silent failure story"
AI engineering
  • by Ahmed
  • 31 Jul

"The automation that \"worked\" for three weeks: a silent failure story"

How to scope an AI project so it doesn't drift
scoping
  • by Ahmed
  • 30 Jul

How to scope an AI project so it doesn't drift

"LLMs inside business workflows: the pattern that actually works"
AI engineering
  • by Ahmed
  • 29 Jul

"LLMs inside business workflows: the pattern that actually works"

"RAG in production: where it actually breaks"
RAG
  • by Ahmed
  • 28 Jul

"RAG in production: where it actually breaks"

"Latency budgets for LLM features: what users will actually wait for"
llm
  • by Ahmed
  • 27 Jul

"Latency budgets for LLM features: what users will actually wait for"

"Building your first golden set: 30 examples beat zero"
Evals
  • by Ahmed
  • 24 Jul

"Building your first golden set: 30 examples beat zero"

"Tool use without trust: the model requests, your code decides"
Agents
  • by Ahmed
  • 23 Jul

"Tool use without trust: the model requests, your code decides"

"The happy path is the easy part: designing AI features for the inputs you didn't expect"
AI engineering
  • by Ahmed
  • 22 Jul

"The happy path is the easy part: designing AI features for the inputs you didn't expect"

"Do you need an agent at all? Choosing the simplest AI architecture that works"
AI engineering
  • by Ahmed
  • 21 Jul

"Do you need an agent at all? Choosing the simplest AI architecture that works"

"Observability for AI systems: what to log so failures are debuggable"
AI engineering
  • by Ahmed
  • 20 Jul

"Observability for AI systems: what to log so failures are debuggable"

"Retries, timeouts and idempotency: the plumbing that keeps agents alive at 3am"
AI engineering
  • by Ahmed
  • 17 Jul

"Retries, timeouts and idempotency: the plumbing that keeps agents alive at 3am"

"Testing AI features without burning API credits: the provider seam"
AI engineering
  • by Ahmed
  • 16 Jul

"Testing AI features without burning API credits: the provider seam"

"Keeping the LLM bill predictable: token ceilings, attempt caps and cost budgets"
AI engineering
  • by Ahmed
  • 15 Jul

"Keeping the LLM bill predictable: token ceilings, attempt caps and cost budgets"

"When the model should say 'I don't know': confidence gates and human-in-the-loop design"
AI engineering
  • by Ahmed
  • 14 Jul

"When the model should say 'I don't know': confidence gates and human-in-the-loop design"

"Validate, don't trust: handling structured LLM output in TypeScript and Python"
AI engineering
  • by Ahmed
  • 13 Jul

"Validate, don't trust: handling structured LLM output in TypeScript and Python"

"Evals: the missing test suite for AI features"
AI engineering
  • by Ahmed
  • 10 Jul

"Evals: the missing test suite for AI features"

"Systems, not prompts: the architecture of an AI feature that doesn't fall over"
AI engineering
  • by Ahmed
  • 9 Jul

"Systems, not prompts: the architecture of an AI feature that doesn't fall over"

"Running LLM automation unattended for months: what actually breaks"
AI engineering
  • by Ahmed
  • 8 Jul

"Running LLM automation unattended for months: what actually breaks"

Anatomy of a production-grade extraction agent
AI engineering
  • by Ahmed
  • 7 Jul

Anatomy of a production-grade extraction agent

Demo vs system: the line most AI work never crosses
AI engineering
  • by Ahmed
  • 15 Jun

Demo vs system: the line most AI work never crosses

© 2026 Ahmed Fareed. All rights reserved.

LOADING