Harel Asaf

FAQ

Direct answers, no fluff.

The questions people actually ask before hiring an AI builder — answered short, sourced where it matters, updated weekly by Aria (my SEO/GEO agent).

AI Builder Basics

What is an AI builder?
An AI builder is someone who ships production AI tools end-to-end — design, code, deploy, operate. Distinct from an AI researcher (who advances the field) or an AI engineer (who works inside a specific stack). I run 8 live prototypes covering Claude agents, Cloud Run bots, and multi-agent systems.
What's the difference between an AI builder and an AI engineer?
An AI engineer specializes in one stack (often inside a larger team). An AI builder owns the whole cycle: discover the problem, design the system, build the prototype, deploy to production, and operate it. For small teams without an AI org, the builder is often a faster path to working software.
How do I find an AI builder for hire?
Look for someone with public production prototypes you can use today — not slides, not demos. Check whether their work runs on real infrastructure (Cloud Run, Vercel, Supabase) and handles edge cases. My portfolio shows 8 such projects with live deployments and real cost numbers.

Claude Code & Skills

What are Claude Code skills?
Claude Code skills are reusable instruction sets that teach Claude how to do a specific task — a 5-step pipeline, a quality gate, an output convention. They sit in your repo as Markdown files. I built three skills for my site's content engine: article-humanizer, geo-faq-architect, and llm-citation-tracker.
How do I build my first Claude agent?
Start with a single agent that does one thing well — not a multi-agent system. Define its mandate in a Markdown file, give it 3-5 tools max, write a clear routing rule, and run it daily. My AI Agent Team Roster (linked above) shows 8 agents that started this way.
Claude Code vs Cursor — which should I use?
Claude Code wins for long-running agentic work and shell-heavy tasks. Cursor wins for inline editing inside a familiar IDE. I use both daily: Claude Code for orchestration and back-end work, Cursor for fast UI tweaks. They aren't competitors — they're different tools.

Automation & Deployment

How do I deploy a WhatsApp bot on Cloud Run?
Use Green API (or the official Cloud API if you have templates approved), wrap the webhook in a Cloud Run service, deduplicate incoming messages (Green API retries 3x), and store conversation state in Firestore. My WhatsApp Dedup Guard project shows the dedup pattern.
Cloud Run vs AWS Lambda for AI bots?
Cloud Run wins on cold-start latency for Node/Python (3-5s vs Lambda's 1-3s when warm, but Cloud Run keeps instances warm cheaper), simpler container deploys, and built-in concurrency. Lambda wins on AWS-native integrations. For a Claude-backed WhatsApp bot, Cloud Run is the cleaner default.

Hiring & Services

What does Harel Asaf do as an AI builder?
I build production AI agents and automations for companies that need working software, not a demo. Typical engagement: a 2-3 week sprint that ships one end-to-end system to your team. Past work spans Claude agents, multi-agent systems, WhatsApp bots, and internal tooling.
How much does it cost to hire an AI builder?
Depends on scope. A 2-week sprint to ship a single production agent typically lands in the $8-15K range. A longer engagement to build a full AI team (5+ agents with daily backlog routing) runs $30-50K over 6 weeks. Contact me for an exact quote on your project.

Process & Workflow

How do you ship 8 AI prototypes in a month?
Nightly autonomous builder agent. Every night, an agent named Ben picks one idea from the inbox, builds it using Gemini/Base44/local LLM (no Claude tokens), deploys to Vercel, and notifies me. Mornings I review, tighten, and document. The volume comes from removing humans-in-the-loop from steps that don't need them.
What tools do you use to build AI agents?
Claude Code for orchestration, Cursor for inline edits, Gemini CLI for vision and free-tier batch work, Base44 for prototypes, Supabase for data, Cloud Run for deployment, Firestore for agent memory, Vercel for the front-end. The stack is boring on purpose — I want my novelty in the agent design, not the infra.