Solo UnicornFrom Solo Consulting to Three AI Businesses — My Full Tool Stack Revealed
From Solo Consulting to Three AI Businesses — My Full Tool Stack Revealed
In early 2024, I could only serve one client at a time. I had to finish one project before I could take the next — time was always the bottleneck. By early 2026, I was running three businesses simultaneously: ArkTop AI (AI solutions for luxury retail), JewelFlow (a SaaS platform for the jewelry industry), and the Solo Unicorn Club (an entrepreneurship community with 700+ members).
It's not because I started working longer hours. It's because I switched to a different tool stack.
This article lays out every tool I use, what it costs, and why I chose it. Feel free to copy the whole thing.
Background: Why I Built This System
In 2024, I was doing AI consulting — helping enterprises implement generative AI projects. The technology wasn't the problem. The problem was that my time was finite: proposals, client calls, delivery, follow-ups — these four things took turns filling up my calendar.
That's when I started asking myself a question: of all the things eating my time, how many actually require my judgment, and how many are just "someone has to do it"?
The answer surprised me: roughly 60% of my work didn't need my judgment at all — it just needed someone to execute. And that 60% happened to be exactly what AI Agents excel at.
I began systematically handing off the "someone has to do it" work, keeping only the "it has to be me" part for myself. There's a saying in the Solo Unicorn Club: Humans decide, AI executes. That phrase became the core principle behind how I built my tool stack.
Three Principles: How I Choose Tools
Principle 1: Ask "Does this task require my judgment?"
I sort every daily task into two categories:
- Decision tasks: Setting product direction, negotiating key partnerships, handling escalated user complaints. These must be me.
- Execution tasks: Content scheduling, data organization, email follow-ups, community Q&A. These can be delegated.
Before introducing any new tool, I ask: which category does it address? If it's a decision task, the tool can only assist — never replace me. If it's an execution task, the goal is full automation, with me handling only the exceptions.
ArkTop AI's client reports used to take me 4 hours every week: pulling data, formatting, writing summaries, sending emails. Pure execution. Now a Python script plus the Claude API runs it automatically each week, and I spend 15 minutes reviewing the output.
Principle 2: Shorter tool chains are more stable
Early on, I made a common mistake: I used too many tools, each handling a small piece, stringing together seven or eight nodes in a single workflow. Then one day a Zapier webhook went down for half a day, my entire automation pipeline froze — and it was a client who told me.
My principle now: one workflow, three tool nodes maximum. If one tool can do the job, I absolutely don't bring in a second. If I can connect via API directly, I skip the middleware platform.
JewelFlow's user feedback pipeline went from Typeform → Zapier → Notion → Slack (four hops) to Tally → n8n → Notion (three hops). Reliability went up immediately, and the monthly cost dropped from $89 to $31.
Principle 3: Get it running first, optimize later
I've seen plenty of people spend too much time on tool selection — comparing ten options and ending up with none of them actually running.
My approach: pick something good enough, get it running, and evaluate whether to switch after three months. JewelFlow originally used Bubble for a low-code frontend; we only migrated to Next.js after usage grew. If I had agonized over the tech stack from the start, the product would have launched at least two months late.
My Complete Tool Stack
Here's everything I'm using as of March 2026, organized by business function.
Content & Writing
| Use Case | Tool | Monthly Cost | Why I Chose It |
|---|---|---|---|
| Long-form writing, code, complex reasoning | Claude (claude.ai Pro) | $20 | Best at long-context handling, most controllable writing style |
| Research, web lookups | ChatGPT (GPT-4o) | $20 | Great search integration, convenient chart generation |
| Quick drafts, content variants | Gemini Advanced | $19.99 | Google ecosystem integration, works directly in Docs |
| Social media scheduling | Buffer | $15 | Simple, gets the job done, easy API integration |
How I actually use them: My writing workflow is Claude drafts → I edit → Buffer schedules. I mainly use ChatGPT for competitive research and pulling public data, and Gemini for batch tasks in Google Sheets.
Development & Automation
| Use Case | Tool | Monthly Cost | Why I Chose It |
|---|---|---|---|
| AI Agent development | Python + LangChain | Pay-per-use API | Flexible, most stable in production |
| Workflow automation | n8n (self-hosted) | $5 (server cost) | Data stays on-premise, cheaper than Zapier |
| Database | Supabase | $25 | Postgres + real-time subscriptions, great developer experience |
| Deployment | Railway | $20 | Cheaper than Heroku, simpler than AWS |
| Code assistant | Cursor | $20 | Agent mode makes writing boilerplate code much faster |
How I actually use them: ArkTop AI's core system is 8 Agents written in Python + LangChain, running on Railway, with data stored in Supabase. n8n handles data flow between systems.
Community Operations (Solo Unicorn Club)
| Use Case | Tool | Monthly Cost | Why I Chose It |
|---|---|---|---|
| Community platform | Discord | $0 (community tier) | Members are already on it |
| Auto-welcome + onboarding | Custom Agent (Claude API) | ~$12 | Highly customizable, consistent voice |
| Event management | Luma | $0 | Great user experience, free tier sufficient |
| Newsletter | Kit (formerly ConvertKit) | $29 | Segmented sends, clean data |
| Member data | Airtable | $20 | Flexible table views, convenient API |
How I actually use them: When a new member joins Discord, the custom Agent automatically sends a welcome message, runs an interest survey, and adds them to the relevant channels. These three steps used to take me 15 minutes each time — now they're fully automated, and I only step in when a member has a specific question.
Real Numbers
I'll lay out the numbers as clearly as I can, so you can make your own judgment:
Monthly tool cost: About $185 (excluding pay-per-use APIs — those run separately; ArkTop AI is roughly $50/month, the community Agent about $12).
Time saved: Roughly 18–22 hours per week of pure execution work. Specifically: client report automation saves 4 hours, community onboarding saves 6–8 hours (varies with new member volume), JewelFlow user data processing saves 4 hours, content scheduling saves 2–4 hours.
Revenue impact: In 2024, my single highest month was about $12K from one consulting project. By the second half of 2025, the combined monthly revenue from three businesses stabilized in the $18K–$22K range. The growth didn't come from doing more consulting — it came from recurring product revenue (JewelFlow subscriptions + ArkTop service contracts).
Mistakes I've Made
Mistake 1: Over-reliance on a single platform
In April 2025, I bet part of the club's automation on a small tool. That tool suddenly tripled its pricing, then shut down two weeks later. The migration ate an entire weekend.
Lesson: Use self-hostable tools or tools with data export for core workflows. Spending time migrating beats being held hostage.
Mistake 2: Automating too early
In JewelFlow's early days, I spent three weeks building an elaborate user onboarding automation flow, only to discover that users didn't follow the path I'd designed at all. The whole thing had to be torn down and rebuilt.
Lesson: Manually run through 10–20 users first to understand real behavior before automating. Get the process right before you automate it.
Mistake 3: Forgetting to account for the learning curve
When I adopted Cursor, I underestimated how much time it would take to truly master its Agent mode. For the first two weeks, I was actually slower than with a regular editor.
Lesson: Factor in three weeks of friction cost when calculating a new tool's ROI. If there's no clear speed improvement after three weeks, drop it.
For Those Looking to Get Started
I wouldn't recommend copying my entire tool stack out of the gate. Tools are solutions to problems, and your problems are different from mine — so your tools should be too.
But three steps are universal:
Step one: Track your work time for one week. List everything you do and label each item "only I can do this" or "anyone could do this." Many people skip this step and jump straight to shopping for tools. They find the tools, but they're not solving the real bottleneck.
Step two: Pick one execution task and run a three-week automation experiment. Don't start by overhauling your core workflow. Find a low-risk, repetitive task — like compiling a weekly report, posting scheduled content, or organizing form submissions. Get one working, then expand from there.
Step three: Track what you do with the time you save. A lot of people skip this, and the result is that they automate plenty of things but can't figure out where the time went. Saved time only counts as a real win if you reinvest it into higher-value work.
What I'm Thinking About Next
This tool stack can reliably support three businesses right now, but I'm clear about its limits. If ArkTop AI's client count doubles, the current Agent architecture needs to be rewritten. JewelFlow's data model is accumulating tech debt as features grow.
A tool stack isn't a one-time setup — it's a system you iterate on continuously. I expect to do a fairly major overhaul in the second half of this year, focused mainly on Agent reliability and observability.
Quite a few members in the Solo Unicorn Club are also building their own AI tool stacks, and there's a lot of overlap in the mistakes we've all made. If you're building yours — or just starting to think about it — come join the club and share notes. Among 700 people, someone has almost certainly run into the same problem you're facing, and already solved it.
What's your biggest time bottleneck right now? Is it content, operations, development, or something else?