Build logs, guides, and field notes
What shipped, what broke, and what it really cost — from running production AI systems.
Prefer this site in Google
Mark omidsaffari.com as preferred and Google lifts it in Top Stories, AI Overviews and AI Mode for you.
Latest articlesWhat shipped, what broke, and what it really cost — from running production AI systems.
Antigravity Local Jobs Need an October 5 Migration
Keep Antigravity jobs running after October 5. Learn which integrations need new tool adapters and which only need the new agent ID.Sep 18, 2026Explained
Best Software for Paper Quilling Patterns
Compare software for paper quilling patterns by editable outlines, reusable shapes, size controls, and printable templates you can build from.Sep 18, 2026Design
Knit Visualizer vs Stitchmastery
Compare Knit Visualizer and Stitchmastery on computer compatibility, chart editing, written instructions, exports, price, and migration work.
Sep 18, 2026Design
How to Project Sewing Patterns With Inkscape
Project a sewing pattern with Inkscape: import the PDF, preserve scale, set calibration, and check the projected pattern before cutting.Sep 18, 2026Design
KnitBird Knitting Software Cost
KnitBird is free, but support has ended. Check activation, Adobe AIR, old paid-license wording, and compatibility before installing.Sep 18, 2026Design
PDFStitcher Review
Review PDFStitcher for joining tiled sewing PDFs, preserving size layers, fixing overlaps, and preparing a correctly scaled pattern.Sep 18, 2026Design
Cloudflare Shows Which Worker Slowed a Customer Request
Follow a slow request across Cloudflare Workers and Durable Objects, find the slow call, and check tracing costs before rollout.Sep 17, 2026Explained
WeavePoint 9 Review for Loom Control
Review WeavePoint 9 for manual pick tracking and electronic loom control, including driver compatibility, Windows requirements and upgrade checks.Sep 17, 2026Design
Best Software for Punch Needle Patterns
Compare punch needle pattern software for clean outlines, photo conversion, color control, mirrored transfers and full-size printing.Sep 17, 2026Design
Glass Eye 2000 Review
Review Glass Eye 2000 for stained glass drawings, piece labels and full-size patterns, with edition limits, trial access and compatibility checks.Sep 17, 2026DesignThe lanes.Pick the one you came for.
143 pieces
AI
Models, agents, coding assistants, image, video, voice136 piecesBuild
Ship with AI when you're not an engineer: vibe-coding, app/SaaS builders, automation, MCP, just-enough infra135 piecesDesign
For designers, brand owners, creative directors138 piecesGrowth
AI for marketing, SEO/GEO, content engines, and the operator-business work of turning audience into revenue6 piecesPlaybooks
AI for your niche: the workflows that actually move, tool by tool, priced45 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 run564–587 / 603
Canva AI 2.0 Brand Intelligence vs a billable brand refresh
I ran Canva AI 2.0's Brand Intelligence against the monthly brand-refresh I bill studios for. What held in the research preview, what I redid in Figma.May 17, 2026Design
Krea 2 Style Transfer: 40-Frame Campaign, Real Cost Ladder
Krea 2's style-transfer tested on a real 40-frame client campaign: the compute-cost ladder vs a Midjourney pipeline and where art direction still carried.May 17, 2026Design
LLM SEO Tracking: Why One Tool Can't See All Engines
Only ~11% of domains ChatGPT cites also appear in Perplexity, why one AI-visibility tracker misses most of it, and the cross-engine stack that fixes it.May 17, 2026Growth
Newsletter CAC: $6,400 Across Boosts, SparkLoop & Referral
A 90-day, $6,400 test of three paid newsletter channels: the cost-per-retained-subscriber math, the deliverability tax, and which channel is wrong.May 17, 2026Growth
Next.js May 2026: the CVE that hurts cache-fronted RSC
Operator triage of Next.js's 13 May-2026 security advisories on a Vercel plus edge-cache stack: which can actually reach you, and the safe upgrade path.May 17, 2026Build
Codex folded into the ChatGPT org: my roadmap-risk test
OpenAI folded Codex, ChatGPT and the API into one org under Brockman. Codex is now a super-app wedge, not a standalone tool. My roadmap-risk test.May 17, 2026AI
AI Content Engine: $1.40/article, 4 rules, no penalty
Running a daily AI content engine: the real per-article cost, the keyword spine, and the four rules that keep it clear of Google's Scaled Content Abuse.May 17, 2026Growth
Falcon-H1 Arabic vs Intercom Fin: AED 11,500/mo rebuild
Running a sovereign UAE Arabic model as a retailer's bilingual support layer vs Intercom Fin: cost, deflection, PDPL, and when the SaaS bot still wins.May 16, 2026Growth
Notion Custom Agents: $340/month shock — cap or rebuild?
The real burn math after Notion metered Custom Agents at $10 per 1,000 credits: a client OS's three-way decision and the accounting recaps skip.May 16, 2026Growth
Figma Make Custom Skills: 3-Skill Pack for Design Systems
Packaging a design system as a reusable Figma Make skill: the skill files, the Notion-connector wiring for live tokens, and the two real limits.May 16, 2026Design
AI Bandit vs A/B Testing: 30-Day CVR Data & the Trade
Replacing A/B testing with AI bandits over 30 days: tool costs, the rollout, and the statistical-significance problem the CRO listicles skip.May 16, 2026Growth
Production AI-agent guardrails: blast-radius architecture
The production blast-radius architecture for AI agents: one cost chokepoint, a no-shell tool allowlist, idempotent steps, and pre-execution denial.May 16, 2026Build
Opus 4.7 +27%, GPT-5.5 2x: The Harness That Held
Three vendors quietly raised AI cost in one week. Which mechanism hits your stack, and the cache, router and spend-cap architecture that absorbs it.May 16, 2026AI
Grok Build: cheapest model, $300/mo door — the operator math
Grok Build's cheap model is gated behind a $300/mo tier: the subscription-vs-API math for an operator already on Claude Max, and why it waits.May 16, 2026AI
Dubai Legal AI Stack: 68% Less Review, AED 9,500/Month
A workflow-mode playbook for the managing partner of a 10-25 lawyer UAE boutique who has watched Harvey AI, Vincent AI, and Lexis+ Protégé pricing drift past USD 1,000 per seat per month and concluded — correctly — that the enterprise…May 15, 2026Growth
Mercury OCC Approval: My 5-Venture Banking Decision
Mercury received OCC conditional approval for Mercury Bank, N.A. on April 28, 2026, two weeks after acquiring Central, an AI-native payroll platform. Senior founders running multiple entities now have to decide whether to consolidate…May 15, 2026Growth
Runway Agent vs 4-Tool Pipeline: 9 min vs 4.5 hrs
Runway Agent (launched May 13, 2026) compresses what used to be a four-tool brand-video pipeline — Midjourney + Runway Gen-4.5 + ElevenLabs + DaVinci — into a single conversation, with voiceover, dialogue, and music assembled in minutes.…May 15, 2026Design
AdCreative.ai vs Pencil Pro vs Control: $4,800 CPA Data
A 6-week, $4,800 Meta ads test pitting AdCreative.ai Professional ($249/mo) against Pencil Pro ($249/mo) against hand-designed control creative on a single DTC product, with matched audiences, identical offers, and clean 7-day click…May 15, 2026Growth
Meta Andromeda + Advantage+ ad-creative stack for DTC, 2026
Meta defaulted Advantage+ Creative on and dropped the Sales conversion threshold to 25/week. Creative variety is the new lever. Stack I'd run now.May 15, 2026Growth
Anthropic SDK 0.100–0.102: What to Delete on Cloudflare
A migration playbook for senior devs running production AI on a Durable-Object-plus-Workflows stack. Maps every piece of the Managed Agents API that landed in SDK 0.100–0.102 to the Cloudflare primitive it replaces, surfaces the three…May 15, 2026Build
Claude Agent SDK on a Separate Meter June 15: The Token Math
On May 13 Anthropic announced that starting June 15, 2026, programmatic Claude usage (Agent SDK, `claude -p`, Claude Code GitHub Actions, third-party apps) draws from a separate credit pool instead of subscription limits. Pro gets $20,…May 15, 2026AI
PwC Certifies 30,000 on Claude: 18-Month Arbitrage Window
PwC and Anthropic announced a multi-year alliance expansion on May 14, 2026 that puts Claude Code and Cowork into hundreds of thousands of PwC professionals' hands, with 30,000 US staff certified inside the year. The release publishes…May 15, 2026AI
Codex Mobile Preview: Async Coding as the Solo-Op Default
On May 14, 2026, OpenAI shipped a preview of Codex inside the ChatGPT mobile app on iOS, iPad, and Android, available on all plans including Free and Go. The phone becomes a thin controller over a Codex desktop session running on a paired…May 15, 2026AI
Codex in ChatGPT mobile: async coding for solo operators
OpenAI shipped Codex inside ChatGPT mobile on May 14, free tier included. The phone is now the agent control plane for solo Mac operators.May 15, 2026AINewsletter
One letter, every Sunday.Working systems, not hot takes.
Weekly. No spam. Unsubscribe anytime.