Ship software with AI, whether or not you write code.
Setups, guides and tool verdicts for shipping with AI: Claude Code, Codex and Cursor, vibe-coding and app builders, automation, MCP servers and just enough infra on Vercel, Cloudflare and Supabase. Priced and tested on the stack behind my own live products.
Latest articlesWhat shipped, what broke, and what it really cost โ from running production AI systems.
LoRA Fine Tuning: QLoRA, Adapters and Evaluation
Learn how LoRA and QLoRA work, when to fine-tune, which hosted and GPU routes train adapters, and how to evaluate and serve them.Oct 26, 2026Build
LangChain Alternatives in 2026: LangGraph, LlamaIndex, Pydantic AI and More (Compared)
Compare LangGraph, LlamaIndex, Pydantic AI and six more LangChain alternatives by language, models, licence and the problem each solves.Oct 26, 2026Build
LLM vs SLM: Which Model Fits Your Feature?
Choose large or small language models by task. Compare current model families, deployment limits and a routing example for production features.
Oct 26, 2026Build
Model Distillation: Teacher to Student for Production LLMs
When to distill a narrow LLM task, which managed routes are available, what output-training terms allow, and how to evaluate before switching traffic.Oct 26, 2026Build
Vision Language Models: What Builders Can Trust
What vision language models can read, where they fail, and how to choose API or self-hosted models with a test plan for your own images.Oct 26, 2026Build
Multi Agent Systems: When One Agent Beats a Team
When should you split one LLM agent into several? Compare supervisors, hierarchies, handoffs and group chat, with a task-based decision table.Oct 26, 2026Build
Prompt Injection: Direct and Indirect Attacks in AI Agents
Understand direct and indirect prompt injection, how agents expose business data, and the safeguards to use before connecting an inbox or CRM.Oct 26, 2026Build
GraphRAG: How It Works and When to Use It
How Microsoft GraphRAG answers questions across documents, what drives indexing work, and when to use it instead of vector, hybrid or agentic RAG.Oct 26, 2026Build
AI Agent Architecture: The Production Map
Map the model, tools, memory, orchestration, approvals and evaluation behind a production AI agent, with a support workflow and buy-versus-build decisions.Oct 25, 2026Build
Cursor Models: What You Can Use After the October Change
Which Cursor models you can use after October, how Auto routes tasks, and how to choose across Grok, Composer, Claude, GPT and Gemini.Oct 25, 2026BuildThe other lanes.Pick the one you came for.
271 pieces
AI
Models, agents and coding assistants, image, video and voice tools: reviewed, priced and compared.125 piecesDesign
AI as a craft tool for designers, brand owners and creative directors: image, video, type, and the prompts behind the work.211 piecesBusiness
Voice and phone agents, customer service, sales and marketing, operations and AI search visibility, for the people who run a business.8 piecesPlaybooks
AI for your niche: the workflows that actually move, tool by tool, priced52 piecesExplained
What shipped, explained: every real release, decoded for the people who use it0 piecesLab
The agent journey, entry by entry: what shipped, what broke, what was decided, in the ventures I run12โ35 / 321
LLM Parameters: Model Weights and Generation Settings
Understand model size, active MoE weights and API controls, with provider-specific rules and starting settings for extraction, drafting and code.Oct 25, 2026Build
Continuous Batching: How It Improves LLM Throughput
How continuous batching improves shared LLM throughput, changes latency, and works with KV cache in vLLM, SGLang, TensorRT-LLM and TGI.Oct 25, 2026Build
Tree of Thoughts Prompting: When to Use It
Learn how Tree of Thoughts branches, evaluates and prunes, how it differs from chain-of-thought, and when extra model calls earn their place.Oct 24, 2026Build
Agent Loop: Model Calls, Tools and Stop Conditions
Follow the agent loop from model call to tools and observation. Learn the stop conditions, framework controls and fixes for stalled runs.Oct 24, 2026Build
Agentic Coding: How It Works and How Much to Delegate
Understand agentic coding, how Claude Code, Codex, Cursor and Copilot work, and where specs, permissions, tests and review limit delegation.Oct 24, 2026Build
Context Rot: Why Long Chats Degrade and How to Fix It
Why agents lose accuracy before the context window fills, what Chroma's study shows, and how to use compaction, retrieval and clean contexts.Oct 24, 2026Build
Hybrid Search: BM25, Vectors and a Recall Test
Combine BM25 and vector search with RRF or weighted scores. Add hybrid retrieval in your existing store and measure recall on labelled queries.Oct 24, 2026Build
Mixture of Experts Explained: Active vs Total Parameters
Understand MoE routers, experts and active vs total parameters, with maker-sourced examples and practical rules for API use and self-hosting.Oct 24, 2026Build
Eval Driven Development: A Practical Loop for LLM Changes
Learn eval driven development with a 20-case example: define checks, record a baseline, change one thing, and gate the release on the rerun.Oct 24, 2026Build
Tokens per Second: How to Measure LLM Output Speed
Understand LLM tokens per second, first-token delay and end-to-end latency, then measure your own streamed response.Oct 24, 2026Build
LLMOps: What It Is and What to Put in Place First
What LLMOps covers, how it differs from MLOps, and the release, evaluation, tracing and incident controls a production team needs first.Oct 23, 2026Build
Structured Outputs: OpenAI, Claude and Gemini Explained
Understand structured outputs across OpenAI, Claude and Gemini, with schema limits, invoice extraction, validation, retries and streaming.Oct 23, 2026Build
Railway Alternatives in 2026: Render, Fly.io, Northflank, Heroku and More
Compare Railway alternatives for services, Postgres, workers and cron, with current starting prices, secrets handling and a staged migration checklist.Oct 22, 2026Build
Gumloop Review
Gumloop reviewed for lead research, scraping, documents and email: a step-by-step setup, current credits, approval controls and alternatives.Oct 22, 2026Build
Synthetic Data Generation: LLM Training and Test Data
Generate LLM evals, fine-tuning examples and test fixtures with a worked recipe, filtering, deduplication, human checks and provider terms.Oct 22, 2026Build
RLHF Explained: Human Feedback, DPO and Fine-Tuning
How RLHF turns human rankings into model behavior, how DPO and verifiable rewards differ, and which preference fine-tuning APIs a small company can use.Oct 22, 2026Build
Model Router: How LLM Routing Works and How to Test It
Learn how LLM model routing works, compare documented options, and test quality with a support-app example, cost formula, and held-out eval set.Oct 21, 2026Build
Semantic Caching: How It Works and When to Reuse Answers
Understand semantic caching, its four-step lookup, and how thresholds, expiry and user scope prevent similar questions from receiving the wrong answer.Oct 21, 2026Build
Pydantic AI vs LangChain
Choose a Python agent framework with matching ticket-triage code, validated outputs, LangGraph workflows, tracing, evals and migration guidance.Oct 21, 2026Build
Zed vs Cursor 2026: Editing Feel, Agents and Teamwork
Zed vs Cursor for developers: compare editing feel, agents, model choice, collaboration and the limits of switching or using both.Oct 21, 2026Build
LLM Inference Explained: Prefill, Decode and Serving Choices
Understand LLM inference, from prefill and decode to caching and batching, then choose between a model API, an inference provider and your own GPUs.Oct 21, 2026Build
Semantic Search: How It Works and When to Use It
Learn how semantic search uses embeddings, when keywords still win, and how hybrid retrieval and reranking fit a production search system.Oct 21, 2026Build
Factory AI
Factory Droids for engineering teams: verified capabilities, workflow examples, limitations, and when to choose it beside Claude Code, Codex or Cursor.Oct 21, 2026Build
Speech to Text API 2026: Mistral, Deepgram, AssemblyAI, OpenAI, Google, ElevenLabs and Speechmatics Compared
Compare Mistral Voxtral 3, Deepgram, AssemblyAI, OpenAI, Google, ElevenLabs and Speechmatics for files, live agents, languages and speaker labels.Oct 21, 2026BuildPrefer this site in Google
Add as a preferred sourceon GoogleMark omidsaffari.com as preferred and Google lifts it in Top Stories, AI Overviews and AI Mode for you.
Newsletter
One letter, every Sunday.Working systems, not hot takes.
Weekly. No spam. Unsubscribe anytime.