Build logs, guides, and field notes

What shipped, what broke, and what it really cost — from running production AI systems.

Latest articlesWhat shipped, what broke, and what it really cost — from running production AI systems.

Qwen 4: Qwen4-Max Explained

Qwen 4: Qwen4-Max Explained

Qwen4-Max explained: production versus preview access, context and output limits, open-model promises, and when to move from Qwen3.8-Max.

Nov 17, 2026AI

The lanes.Pick the one you came for.

996–1019 / 1174
Claude Opus 5 and the Effort Dial: Spend More Compute Only When the Work Deserves It

Claude Opus 5 and the Effort Dial: Spend More Compute Only When the Work Deserves It

Claude Opus 5 costs half as much as Fable 5. Learn when its five effort levels save money, when to escalate, and what breaks during migration.Jul 26, 2026AI
ElevenLabs Music Finetunes API Turns Your Sound Into a Reusable Product Layer

ElevenLabs Music Finetunes API Turns Your Sound Into a Reusable Product Layer

Train Eleven Music on owned audio, call that sonic identity by API, and build consistent soundtracks for brands, creators, and games.Jul 26, 2026Design
How to Pick an Open-Source AI Coding Agent: 9 Tools, Four Different Jobs

How to Pick an Open-Source AI Coding Agent: 9 Tools, Four Different Jobs

Nine open-source AI coding agents compared by workflow, license, privacy, setup, and the point each breaks, with current costs and replacements.Jul 25, 2026AI
Meta AI's New Email and Calendar Agent: Set the Approval Boundary First

Meta AI's New Email and Calendar Agent: Set the Approval Boundary First

Meta AI can now plan from email and calendar context. See what shipped, what is undisclosed, and the approval boundary to set first.Jul 25, 2026AI
Employee Onboarding Software Is Five Jobs: 11 Tools Ranked

Employee Onboarding Software Is Five Jobs: 11 Tools Ranked

Eleven employee onboarding tools ranked by the job they actually solve, with current pricing, free-trial terms, limits, and a decision guide.Jul 24, 2026Business
Gemini 3.6 Flash Review: Adopt It for Agents, Not Every Task

Gemini 3.6 Flash Review: Adopt It for Agents, Not Every Task

Gemini 3.6 Flash cuts output price to $7.50/M and output use by 17%. See the benchmarks, migration risks, cost math, and adoption verdict.Jul 23, 2026AI
ChatGPT Ads Custom Audiences: The 25,000-User Floor Changes the Strategy

ChatGPT Ads Custom Audiences: The 25,000-User Floor Changes the Strategy

ChatGPT Ads Custom Audiences require 25,000 matched users. Learn when to include, exclude, or adjust bids, plus the setup that avoids wasted spend.Jul 23, 2026Business
OpenAI Presence Is a Managed Agent Deployment, Not a Self-Serve Tool

OpenAI Presence Is a Managed Agent Deployment, Not a Self-Serve Tool

OpenAI Presence bundles policies, simulations, approved actions, human escalation, and a Codex improvement loop. Who should buy it, and who should build?Jul 22, 2026AI
GPT-5.6 vs Claude Sonnet 5: Sonnet Wins on Value, Sol Wins the Ceiling

GPT-5.6 vs Claude Sonnet 5: Sonnet Wins on Value, Sol Wins the Ceiling

GPT-5.6 vs Claude Sonnet 5 on price, coding, agent benchmarks, long context, and the routing rule that decides which model fits.Jul 22, 2026AI
Grok 4.5 Review: The $2/$6 Agent Model Wins on Cost, Not Raw Coding

Grok 4.5 Review: The $2/$6 Agent Model Wins on Cost, Not Raw Coding

Grok 4.5 costs $2/$6 per million tokens. See current benchmarks, the 200K pricing cliff, Cursor math, and exactly who should adopt it.Jul 22, 2026AI
Google Pics Rolls Out August 18: Why It Won't Replace Canva Yet

Google Pics Rolls Out August 18: Why It Won't Replace Canva Yet

Google Pics arrives for eligible Workspace accounts on August 18. See who gets it, what it edits, the plan cost, and why Canva still stays in the stack.Jul 21, 2026Design
GLM-5.2 Review (2026): The Open-Weight Model That Runs Claude Code at Half the Cost of Opus

GLM-5.2 Review (2026): The Open-Weight Model That Runs Claude Code at Half the Cost of Opus

GLM-5.2 is a 753B open-weight model that matches Claude Opus on coding at under half the cost. Real pricing, benchmarks, and when to actually use it.Jul 19, 2026AI
Meta Muse Image Review: Social-Ready, Not Studio-Ready

Meta Muse Image Review: Social-Ready, Not Studio-Ready

Muse Image is Meta's free, agentic image generator. See where it beats simple tools, where it falls short, and who should use it.Jul 18, 2026Design
Kimi K3 Review: Frontier Performance at Half the API Cost, but the Weights Are Not Here Yet

Kimi K3 Review: Frontier Performance at Half the API Cost, but the Weights Are Not Here Yet

Kimi K3 reviewed with live API pricing, independent benchmarks, production limits, and the decision rule versus GPT-5.6 Sol and Claude Fable 5.Jul 17, 2026AI
Gemini 3.5 Pro Is Delayed: No Release Date, No Reason to Wait

Gemini 3.5 Pro Is Delayed: No Release Date, No Reason to Wait

Gemini 3.5 Pro missed its June window and is reportedly months behind. See what to use now, the real cost, and the proof to wait for.Jul 17, 2026AI
Meta Muse Spark 1.1 Review (2026): The 75%-Cheaper Frontier API, and When It Beats GPT and Claude

Meta Muse Spark 1.1 Review (2026): The 75%-Cheaper Frontier API, and When It Beats GPT and Claude

Meta's first paid frontier API: $1.25/$4.25 per M tokens, ~75% under GPT and Claude. Where Muse Spark 1.1 wins, where it trails, and when to switch.Jul 13, 2026AI
Claude Sonnet 5 Review (2026): What It Really Costs, Where It Beats Opus, and the Tokenizer Catch

Claude Sonnet 5 Review (2026): What It Really Costs, Where It Beats Opus, and the Tokenizer Catch

Claude Sonnet 5 lands at $2/$10 intro pricing with near-Opus agentic skill, but a new tokenizer adds ~30% more tokens. The honest switch verdict.Jul 13, 2026AI
Seedream 5.0 Pro Review (2026): The Update That Turns It Into Design Software

Seedream 5.0 Pro Review (2026): The Update That Turns It Into Design Software

Seedream 5.0 Pro adds layer separation and coordinate editing, making ByteDance's cheap image model behave like design software. What it does and costs.Jul 13, 2026Design
Codex Remote GA: Your Phone Is an Engineering Control Plane

Codex Remote GA: Your Phone Is an Engineering Control Plane

Codex Remote now lets you start, steer, review, and approve work on Mac or Windows hosts from the ChatGPT mobile app.Jul 11, 2026Build
ChatGPT Work Review: The $20/Seat Agent That Turns Goals Into Deliverables

ChatGPT Work Review: The $20/Seat Agent That Turns Goals Into Deliverables

ChatGPT Work turns goals into reviewable deliverables. Here is what it does, what Business costs, and the workflow that makes it useful.Jul 11, 2026AI
GPT-5.6 Review: Sol, Terra, Luna, and the Routing Rule

GPT-5.6 Review: Sol, Terra, Luna, and the Routing Rule

GPT-5.6 Sol, Terra, and Luna reviewed with live API prices, benchmarks, context limits, and the routing rule that decides which tier to use.Jul 10, 2026AI
Gemini Managed Agents Can Now Finish the Job After You Disconnect

Gemini Managed Agents Can Now Finish the Job After You Disconnect

Gemini Managed Agents now run in the background, connect to remote tools, call your functions, and refresh credentials without losing their workspace.Jul 10, 2026AI
Codex Pricing in 2026: What Free, Plus, Pro, Business, and API Really Cost

Codex Pricing in 2026: What Free, Plus, Pro, Business, and API Really Cost

OpenAI Codex pricing in 2026: Free, Go, Plus, Pro, Business, Enterprise, API rates, five-hour limits, and which plan builders should buy.Jul 8, 2026Build
Codex vs Claude Code After the Microsoft CLI Study: The 24% PR Lift Is Not the Whole Decision

Codex vs Claude Code After the Microsoft CLI Study: The 24% PR Lift Is Not the Whole Decision

Microsoft's CLI-agent study found 24% more merged PRs. Here's what that changes, and what it does not, for Codex vs Claude Code in 2026.Jul 5, 2026AI

Prefer this site in Google

Add as a preferred sourceon Google

Mark 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.