I will build your ai automation workflow with claude code, human approval built in


About this gig
I help small businesses and teams turn repetitive, manual processes into AI-powered automations - and I design them with safety in mind, not just speed.
My background: I've built a working retrieval-based (RAG) lookup system used by 10 people that cut a 10-minute task down to about 2 minutes, and rolled out a generative-AI meeting-minutes and knowledge-sharing workflow to a 16-person team, used across roughly 20 meetings per week. I also built 5 Excel VBA automation tools with generative-AI assistance.
What sets my approach apart: I actively design and run multi-agent AI systems using Claude Code for my own business processes - systems where different AI roles (research, drafting, review) work together under a written rule-set, with a required human-approval step before anything irreversible happens (sending a message, publishing, spending money). If you've ever worried that "AI automation" means giving up control, that's exactly the problem I build against.
Tools I work with: Claude Code, for building and orchestrating the workflows above.
Message me with a short description of your current process, and any steps that need your approval, before ordering.
Get to know Yuta
VBA and AI Automation, AI Meeting Minutes, ENtoJP Translation Claude Code
- FromJapan
- Member sinceJul 2026
Languages
Japanese, English
FAQ
Do you actually build this yourself, or does an AI tool do it for you?
I design the workflow, the rules, and the approval checkpoints, and use Claude Code as my primary build tool to implement them quickly and reliably. Human judgment on what should happen and when, AI for fast implementation - that's the core of what I'm selling.
Can the AI agent send emails, post to social media, or make payments on its own?
Not by default, and I recommend against it. Every system I build treats sending, publishing, deleting, and payment as actions that require your explicit approval first. If you want full autonomy for a given action, we can discuss it - but my default always includes a human checkpoint.
Can we hop on a call to discuss the project?
I work primarily asynchronously through Fiverr messages/chat. I'm not set up for complex real-time phone calls in English - written async communication works best and leaves a clear record of requirements.
How fast will you respond, and what are your working hours?
I typically work about 1-2 hours on weekday evenings (JST) and about 3 hours on weekends - roughly 10 hours per week total. I'll confirm delivery timing against this before accepting an order with a tight deadline.
What do you need from me to get started?
A short description of your current manual process (step by step), which tools/systems are involved, and which actions (if any) should always require your approval before happening automatically.

