My Monthly AI Stack Review: What I Actually Run (June 2026)

By X.Q. — Founder, One Person Company · Last updated: June 5, 2026

Every month, I audit my AI stack: what's delivering value, what's costing more than it returns, and what needs to be replaced. This is the June 2026 review — my fifth month running One Person Company with this setup. Total monthly cost: $247.

The 4-Layer Stack Architecture

I organize my tools into four layers. Each layer feeds the next:

Layer 1: Foundation (Models) — $240/mo

This is where the actual AI "thinking" happens. I use three models for different jobs:

ChatGPT Pro ($200/mo)

My workhorse. I use it for first-draft content generation, outline creation, and brainstorming. Daily usage: ~1.2 hours. In May, it generated first drafts for 11 articles. I still spend 45-60 minutes editing each one — the model gives me structure and initial research synthesis, but the voice, personal examples, and strategic judgment are mine.

Why Pro and not Plus: I hit Plus rate limits in February when I scaled from 4 articles/month to 8+. The Pro tier ($200 vs $20) is 10x the cost, but the unlimited usage and access to the best model saves me about 45 minutes per article in revision time. At my volume, that's worth it.

Claude API ($20/mo)

I use Claude specifically for second-pass editing and strategic analysis. After ChatGPT produces a draft and I do my edit, I run the final version through Claude with a specific prompt: "Find the weakest argument, the vaguest claim, and the section most likely to make a reader bounce." It catches things I miss. In May, Claude flagged 14 claims across my articles that needed stronger evidence or clearer examples.

Perplexity Pro ($20/mo)

Fact-checking and competitor research. Before any article goes live, I run its key claims through Perplexity. This step caught 3 factual errors in May alone — including a pricing number I had memorized wrong from a competitor's old pricing page.

Layer 2: Workflow (Automation) — ~$6/mo

n8n Self-Hosted ($0 software + ~$6/mo VPS)

n8n is the connective tissue. It moves data between my tools so I don't have to. Current active workflows:

• GSC auto-submit: New or updated pages → Google Indexing API. 1,342 URLs submitted in May.
• Weekly GSC report: Pulls top 50 queries every Monday → Airtable. I review this in 5 minutes instead of logging into GSC and manually exporting.
• Content pipeline: Tracks article stage in Notion, sends me a Slack message when something's been in "draft" for more than 3 days.
• Competitor monitor: Checks 12 competitor URLs daily for content changes.
• Backup workflow: Nightly backup of all Notion pages to a Git repo.

Why self-hosted n8n instead of n8n Cloud: The cloud version starts at $20/mo. For my volume, self-hosting on a $6/mo VPS is cheaper and gives me unlimited workflows. The trade-off: I had to spend about 3 hours setting it up, and I'm responsible for keeping it running. It's gone down twice — once because the VPS ran out of disk space, once because I messed up a config change. Both times I was back up within an hour.

Layer 3: Coding (Site Changes) — $20/mo

Cursor Pro ($20/mo)

I'm not a developer, but with Cursor I can make site changes that work. In May, I shipped:

• Rebuilt /skills/ directory with category pages (SEO, Content, Growth, Ops, Automation)
• Added newsletter subscribe form with D1 backend
• Fixed 47 internal broken links
• Added structured data to 50+ pages

My safety rules with Cursor: (1) Never push directly to production. Everything goes through a staging branch. (2) Run diffs through Claude for review before merging. (3) If a change touches more than 3 files, I test it manually on staging first. These rules have prevented every production incident since I started using Cursor in April.

Layer 4: Operating System (Documentation) — $0/mo

Notion (Free)

Single source of truth for: content pipeline (47 articles in various stages), SOPs (12 documented processes), weekly reviews (searchable decision log since January), and idea backlog.

Airtable (Free)

Structured data only: keyword rankings (200+ keywords with weekly snapshots), content inventory (every URL with metadata), and backlink tracking.

What I Replaced and When

What I'm Evaluating for Next Month

• Claude Code ($20/mo): Considering adding this for more complex site changes that Cursor struggles with (multi-file refactors, database migrations). Would add $20/mo to the stack.
• Cloudflare R2 ($5-10/mo): Currently storing images in the repo. Moving to R2 would keep the repo cleaner and give me a CDN. Small cost, likely worth it.
• Dropping Airtable: As my keyword tracker grows past 200 rows, Notion's database features might be sufficient. Airtable's value is the API, but I can replicate the GSC data pull to write directly to Notion.

Stack Rules I've Learned

1. Add tools one at a time, not all at once. I built this stack over 4 months. Every time I tried adding two tools in the same week, I spent more time configuring than using.
2. Every tool must replace something specific. I don't add a tool because it's "interesting." I add it because a current bottleneck is costing me measurable time or money.
3. Review the stack monthly. Last month's review caught that I was still paying for Zapier ($29/mo) even though I'd already migrated everything to n8n. That was $29 of pure waste.
4. Free tiers are real until you hit volume. Notion and Airtable are still free for me. When I outgrow them, I'll pay — but only when the free tier actually breaks.
5. The tools are 30% of the outcome. Process is 70%. My content pipeline, editing checklist, and weekly review cadence matter more than which specific LLM I use. A bad process with the best tools still produces bad output.

FAQ

How much does a complete AI stack cost for a solo founder?

My actual bill is $247/month. Most solo founders can start at $20-50/month (ChatGPT Plus + Notion) and scale up as they prove the workflow. Don't build my stack on day one — I got here over 5 months.

How long does it take to set up?

The core (ChatGPT + Notion) took one afternoon. The full stack with n8n automation took about 2 weeks of part-time work. The hardest part was n8n — I spent 3 nights debugging the GSC API integration before it worked reliably.

Can I run this without coding?

The foundation layer (ChatGPT, Claude, Perplexity) and operating system (Notion, Airtable) require zero coding. n8n and Cursor need basic technical literacy: you should understand APIs, be comfortable reading error messages, and know how to use a terminal. You don't need to be a software engineer, but you can't be completely non-technical.

Related Articles

• The 7 AI Tools I Actually Use (May 2026 Bill: $247)
• n8n vs Make vs Zapier: 6 Months of Real Usage
• AI Coding Assistants: What Actually Works for Non-Developers
• Claude vs Cursor vs Copilot: Real-World Comparison

made with Tycoon.us · superagent