Shingo Irie

KingCoding - Your Claude Code & Codex dev, now in your pocket

A Mac app to run Claude Code and Codex on your desk β€” drive them from your phone or a second PC. Dispatch tasks, review progress, approve work from anywhere, all synced to one session. No more git-pulling between machines. King Mode turns a goal into a full plan and ships it, auto-reviewing with verification screenshots. Built by a solo dev tired of losing AI agents across terminal tabs. Your job: set direction, grant permissions, receive results. πŸš€

Add a comment

Replies

Best
Shingo Irie
Hey Product Hunt! πŸ‘‹ I'm Shingo, a solo developer from Japan. I built KING CODING because I had a very specific problem: I wanted to run multiple AI coding agents at once, but keeping track of them was harder than the coding itself. Every time I fired up Claude Code for a few tasks in parallel, I'd lose track of which task finished, which one was waiting for input, and which one silently failed. It was death by terminal tabs. So I built a control tower. Now I register my projects, create tasks, pick an AI agent (Claude Code, Codex), and let them run. The dashboard shows me everything β€” what's running, what's done, what needs my attention. When a task finishes, another AI automatically reviews the result and even takes verification screenshots for UI changes. And the best part? I don't have to be at my desk. KING CODING connects to Discord, so I can send instructions, check progress, and approve plans right from my phone. I've shipped features while waiting for coffee, from the train, even from bed. My Mac keeps coding while I'm out living life. The "King Mode" is my favorite feature β€” I describe a goal like "add dark mode to the settings page," and the AI creates a plan, breaks it into tasks, executes them one by one, and adapts if something goes wrong. Combined with Discord, it means I can kick off a big task from my phone, go for a walk, and come back to a finished PR. This whole app was built using KING CODING itself. It's the most meta thing I've ever made. Would love to hear your thoughts! πŸ™
Jennifer Griffiths

@iritec_jpΒ Love this "death by terminal tabs” is painfully relatable πŸ˜…

Also really like the idea of having a control layer rather than just more tools, that feels like where things are heading.

I have actually launched today as well (Global Sponsor Hub, completely different space, international hiring), so I know how intense this day is!

Just checked yours out, really interesting concept

Ryan W. McClellan, MS

Congrats on the launch, Shingo.

"Built using itself" is always a good sign. If it couldn't survive its own development process, that would be telling.

The terminal tab problem is real, and the Discord integration for remote approval is a genuinely practical solution rather than a gimmick.

My skepticism is around King Mode:

"AI creates a plan, executes, and adapts if something goes wrong" is where most autonomous coding pipelines quietly go off the rails on anything non-trivial.

What does "adapts" actually mean when it hits a real blocker? Does it stop and ask, or keep going?

Shingo Irie

@ryanwmcc1Β Thanks β€” right question to push on.

Short version: King Mode doesn't try to be clever about blockers. It stops and asks.

What matters more than "adapts" is what happens before execution: at project start, King Mode confirms purpose, target, and non-goals with you, then builds the plan against that fixed spec. Most pipelines derail because they're executing against a vague ask β€” we front-load that. PM behavior is also configurable per project (non-goals, decision principles, approval policy), so the "decide myself vs. ask" line shifts per context.

Then "adapts" means: fixes its own mistakes, re-plans on scope drift, and on real blockers (ambiguous spec, a call only you can make, same error repeating) it stops, summarizes what's stuck, and pings you on Discord if wired up. It also tracks questions already asked so it can't retry the same dead end.

Where it's weak: judging "blocker vs. solvable hiccup" is an LLM call and not perfect. We bias toward asking and tune from there.