I Spend $600 a Month on AI — And It Is the Best Investment I Have Made

I pay for eight AI services every month. The total comes to approximately $600 USD. When I tell people this, the reaction splits cleanly: either they think I am foolish with money, or they understand immediately why.

At my laptop. Most of the $600/month is invisible — sub-agents running in the background while I work on something else.

Here is why.

What $600 a Month on AI Actually Buys

The services span different categories. There are the large language models — ChatGPT, Claude, Gemini — which form the backbone of most of what I do. There are specialised tools for research, for legal document analysis, for code generation. There are agent orchestration platforms that let the tools talk to each other and operate autonomously on tasks I define.

The result is not just faster work. It is a different category of work. Things I could not do before — not because of skill, but because of time — are now possible.

60 Email Agents

My email system is built from 60 AI agents. This is not an exaggeration for effect. Sixty agents, each with a specific role: one reads incoming messages and categorises them by urgency and type. Others draft responses based on context and relationship history. Others follow up on unanswered threads. Others summarise long email chains before I read them.

One screen for the work, one screen for the system that does the other work. Most days, the second screen has more activity than the first.

I do not spend my mornings in email. My mornings are for thinking, writing, and the work that requires my actual attention. The agents handle what agents are good at. I handle what humans are still necessary for.

This was one of the reasons I left Bangkok. I needed time to build this system properly. The city’s social gravity — dinners, events, the constant friction of urban life — made focused building difficult. I removed myself, built the system, and then could re-enter the city with the system working.

225,000 Searchable Embeddings

I built a custom database of 225,000 vectorised embeddings covering Thai law, legal cases, court decisions, and legal doctrine. This database can be queried in natural language, cross-referenced, and filtered in seconds.

A Python terminal showing pip install errors while setting up notebooklm-py for AI legal research — Where the law firm’s AI infrastructure actually lives — a terminal, a few Python packages, and a lot of patience with externally-managed environments.

The practical application: I used it to analyse 50 years of Thai Supreme Court decisions involving foreigners — cases spanning from the 1970s to the present, covering property law, family law, criminal matters, commercial disputes. A human lawyer working by hand — reading, categorising, cross-referencing — would need months to do this analysis. The AI system did it in approximately two minutes.

The value is not in replacing the lawyer’s judgment. The value is in giving the lawyer complete information before the judgment is made. I can now answer questions about Thai legal precedent that would previously have required weeks of research, if the research was possible at all.

The AI That Analyses My Travel Photos

This one surprised me most when I first used it. I asked an AI to summarise my travel history based on my photo library — geotagged images from 15 years of trips across 30+ countries. It mapped the journey, identified the patterns, told me things about my own travel history that I had half-forgotten.

“I asked AI to summarise my trips abroad based on my pool of pictures. Even me, I cannot remember the dates.”

This is not a trick. This is what happens when you combine a large, well-organised personal data set with an AI that can reason about it. The output is a kind of structured memory — an external version of recall that does not fade the way human memory does.

What AI Cannot Do (Yet)

I spend $600 a month on AI tools. I am under no illusion that they are magic.

AI is excellent at processing, summarising, categorising, drafting, and generating. It is not yet excellent at judgment — the assessment of context, nuance, relationship, and ethics that makes a lawyer or advisor useful rather than merely informative. A system that can analyse 50 years of Supreme Court decisions in two minutes still requires a human to interpret what those decisions mean for a specific client’s specific situation.

The combination — AI speed and scale plus human judgment — is where the current value is. The people who will be left behind are not those who refuse to use AI. They are those who use AI as a replacement for thinking rather than as a tool for more thinking.

Why $600 and Not Zero

Free AI tools exist. They are good. They are not the same as paid professional-grade access.

The difference between the free and paid versions of the major AI systems is measurable in context window size, response quality, speed, and especially in the API access that allows you to build automations. The 60-agent email system does not run on a free ChatGPT account. The embedding database requires API access at scale. The research tools that surface legal cases require premium subscriptions.

If you are using AI professionally and relying on free tools, you are using a hammer when the job needs a crane. The investment is proportionate to the return.