BLOG · AI AGENTS
AI Agents
How production AI agents are actually built — architecture, evals, observability.
AI Agents
4 min read
AI agent observability: tracing every step before it costs you
Agents fail silently — the API call succeeds while the agent does the wrong thing. Here is the lightweight, self-hosted tracing layer I run in production: an in-memory ring buffer that captures every step for zero added cost.
AI Agents
4 min read
AI automation services in UAE: what to expect and what it should include (2026)
A buyer's due-diligence guide from the builder's side: what a legitimate AI automation engagement in the UAE actually includes, what the stack looks like, and the questions that expose a weak vendor.
AI Agents
5 min read
How to connect an AI agent to Google Sheets and Calendar (service accounts done right)
The service-account setup I use in production agents: keys, scopes, the sharing step everyone misses, the auth code pattern, and what Google's 403 and 404 errors actually mean.
AI Agents
4 min read
How to build a lead scoring agent that sales teams actually trust
Five steps to a lead scoring agent reps actually use: outcome-derived weights, a visible reason for every score, and CRM write-back into the views reps already work in.
AI Agents
5 min read
How to build an AI voice agent that books real appointments
Build notes from shipping a production voice agent: the state machine, the latency budget, and why the booking step is never left to the LLM.
AI Agents
4 min read
How to show up in ChatGPT and Perplexity answers (GEO guide)
AI engines cite by retrieval, not PageRank. The verifiable mechanics — answer-first structure, schema markup, crawler access, topical depth — without the agency hype.
AI Agents
4 min read
LLM evaluation framework for AI agents: deterministic, semantic, behavioral
The three-tier eval harness I run on my own agent engine: cheap deterministic checks first, an LLM judge for meaning, and behavioral checks on the tool-call path.
AI Agents
4 min read
n8n vs custom code for business automation: how I decide
A decision framework from using both in production: where n8n's canvas wins, where it starts to hurt, and the hybrid pattern that gets you both.
AI Agents
4 min read
Prompt versioning for production AI: the system nobody talks about
Every prompt on my agent engine is a git-tracked, semantically versioned markdown file. Here's the full system: diffs, reviews, trace logging, and minutes-fast rollbacks.
AI Agents
5 min read
Rate limiting public AI demos: surviving free-tier quotas gracefully
How I rate limit the 73 live agent demos on this site so one bot or one viral link can't drain the API quota — KV-backed limits, token-aware caps, and honest fallback UX.
AI Agents
5 min read
What production-grade AI actually means (the 5 pillars I build on)
The five criteria that separate an AI system that survives real usage from a demo: auditable reasoning, designed fallbacks, outcome measurement, real integration, and ownership after handover.
AI Agents
5 min read
WhatsApp AI agents for UAE businesses: what actually works
A blunt map of WhatsApp automation for UAE businesses: what an AI agent reliably handles, where Meta's rules and off-script conversations break it, and what a working architecture looks like.
AI Agents
4 min read
AI agents for marketing vs marketing automation tools: what's actually different
"AI agent" gets slapped on every marketing tool with a chatbot now. Here's the real technical line between an agent and automation — and how to tell which one you're actually buying.
AI Agents
5 min read
The AI engineer roadmap most people share — and what I'd change after shipping 73 production systems
A 3-month AI engineer roadmap is making the rounds. The phase order is mostly right. But two things are missing that will burn you the moment you touch a real client system — and one phase is too light for what the job actually demands.
AI Agents
4 min read
How I built a Google Ads autonomous agent — full walkthrough
A full video walkthrough of the Google Ads autonomous agent I run in production: how it pulls live data from Google Ads and GA4, makes budget and bid decisions, and reports — every day, without me touching it.
AI Agents
4 min read
Claude Duet explained: Anthropic's proactive AI agent that messages you first
Claude Duet is Anthropic's research preview of a proactive agent: a morning brief from the apps you connect, its own cloud computer, no separate price on Claude Pro and Max, and a US-only waitlist for now.
AI Agents
4 min read
Kimi K3 on 8GB RAM with no GPU: how kimi-k3-in-c runs a 2.78T model locally
A 2.78-trillion-parameter model ran at an 8.24 GB RAM peak with no GPU, because kimi-k3-in-c streams its experts from a fast SSD. The catch is your disk: about 1.7 TB free and about 30 seconds per token.
AI Agents
4 min read
Reverse Engineer Anything with Claude Code: REA setup and the 3 prompts I use
REA (Reverse Engineer Anything) is a free, MIT-licensed tool that lets Claude Code trace how a feature in an app works, from the click to the final output, with evidence. Here is the setup, the limits and my three prompts.
AI Agents
4 min read
NVIDIA SkillSpector: scan Claude Code skills before you install them
About 1 in 4 marketplace agent skills had a security flaw in a January 2026 study. NVIDIA SkillSpector is a free, open-source scanner that gives a skill a 0-100 risk score before it reaches Claude Code. Here is how to install it, run it offline and read the score.
AI Agents
5 min read
AI lead generation on a schedule: ScrapeGraphAI, GPT-6 Astra and a 7am task
ScrapeGraphAI is a free, MIT-licensed Python library with over 31,000 GitHub stars that turns company websites into clean rows. Pair it with a scheduled task in ChatGPT or Claude Code and you get new UK company leads in one sheet every morning. Here is the setup, what it costs and the UK rules.
AI Agents
5 min read
Claude skills for YouTube: the free YouTube Agent Skill, installed and tested
The YouTube Agent Skill gives Claude Code 11 free, MIT-licensed skills for scripts, titles, edit lists, Shorts and SEO. What each one does, how to install it, and what 6 of them produced when I ran them on my own footage.
AI Agents
4 min read
AI website design: fix a generic page with a better brief for your AI coding assistant
AI website design looks generic when the brief is generic. Give the page one goal, a specific offer, a defined type and colour system, then check mobile and contrast. The full brief and review checklist are free.
AI Agents
4 min read
Claude skills for LinkedIn: 27 free installs, sorted by the job they do
Out of the box Claude knows nothing about your voice or your LinkedIn. These 27 free skills, plugins and MCP servers fix that, sorted into five jobs: voice, ideas, writing, visuals and tools.
AI Agents
4 min read
AI motion graphics prompts: start from a Prompt Motion example, then write a Remotion brief
AI motion graphics prompts work better as a production brief than a one-liner. Study a Prompt Motion example made with Claude, then write your own Remotion brief with product, palette, timing and final frame.
AI Agents
4 min read
Opus 5.5 vs Sonnet 5.5: I told Claude Code to "make it better" 10 times
Same ugly coffee shop site, same prompt, a fresh Claude Code session every round. Sonnet 5.5 made careful upgrades and ended at 231 lines. Opus 5.5 kept inventing, reached 1,089 lines and made up customer reviews. Neither searched the web.
AI Agents
3 min read
How to edit videos with Claude Code: the 10-step setup behind my reels
Claude Code can run a whole video edit once ffmpeg, Node, Whisper, OpenCV and Remotion are on your machine. Here is the 10-step setup I use: one take in, a finished reel out, and a skill file that remembers every note.
AI Agents
4 min read
The Claude Code video edit prompt: 7 installs and one prompt that edits a reel
Ask Claude Code to edit a video and you get a plan, not an MP4. Seven free installs and one structured prompt (goal, folder, references, workflow, checks, notes, export) turn it into a repeatable video editor.
AI Agents
1 min read
How to build a Google Ads agent with n8n + MCP
The architecture behind an autonomous Google Ads agent: n8n scheduling, MCP tool wiring, and a Claude media-buyer prompt with guardrails.