You need someone to nag you about practicalities, whether it be personal finances, doctor’s appointments or tax declarations. Usually, it’s your father or your partner.

Being nagged about such matters is a luxury. We’re forgetful monkeys, and we need robust systems (to do-lists, calendars, bullet journals) to become civilised. A human to nag you is like an intelligent, interactive to do-list in flesh and blood. Best of all, there’s no subscription cost.

Since I don’t have a partner, and I want to take some of the burden off my parents, I built an AI housewife: an AI system to make me to what I ought to do. It’s like a crossover between OpenClaw and Personal AI, a system which knows everything about me and helps me manage practicalities, as well as other aspects of being a good citizen, as efficiently as possible. I built the system in February¹, and I have iterated on it since then.

To have the AI system get to know me, I created a CONTEXT.md file, much like a SOUL.md, with personal information (age, occupation, values, life goals, communication preferences, etc.) I also gave the system access to my Obsidian vault and Google Calendar. While this sounds complex, the cost of the system is very manageable: the DigitalOcean droplet hosting the system costs 6 dollars per month, and Claude Code Pro is 20 dollars².

The system has three components. First, there’s a WhatsApp bot, powered by Claude Sonnet 4.6, with the relevant context on me. I can interact with the WhatsApp bot, which responds with highly personalised answers. Second, there are cron jobs scheduled at 7am and 9pm every day. The 7am message is the morning briefing, which gives my daily agenda and reminds me of my highest-priority task. The 9pm is my evening briefing, which basically just reminds me to get to bed and read. Third, I have Claude skills /day and /Sunday for daily and weekly check-ins, which get logged to my daily and weekly notes in Obsidian.

The nagging happens during the weekly check-ins. When running /Sunday, the bot walks me through a general life admin session. It also prompts me about my habits, as specified in CONTEXT.md, and flags if some habit is slipping. When running these commands, Claude Sonnet 4.6 pulls context from my WhatsApp convos with the bot, Google Calendar and any relevant files in Obsidian and summarises the week and generates week-specific reflection prompts.

So far, I’ve found the AI housewife surprisingly useful, with use cases beyond reminders about household chores. For instance, my AI bot often calls me out for committing planning fallacy, or for having optimism bias more broadly. It asks why I didn’t do such-and-such – though not as an admonishing teacher, but rather like a psychologist which leaves the pleasure of trivial insight to the subject³. You could certainly use the AI bot to further increase self-awareness (though I won’t, I have other means). Moreover, I’ve successfully used the bot to increase focus. Before entering a work session, I might message the bot, stating that I’ll try reading a given paper in the next hour. Once the hour is out, the bot asks me how it went. The bot thus serves as an accountability buddy, helping me stay on task.

Overall, I’m pretty happy with my AI housewife. When running /Sunday, I also ask the system to suggest system improvements – effectively integrating a kind of recursive self-improvement. Having iterated on this system since February, I’m quite content with the state of the bot right now.

While I could build a full-blown AI housewife-coach-friend-psychologist-tutor, I won’t. I like humans too much.