About $200 a month is what it costs me to run a solo creative-tech practice that ships DTC infrastructure, runs an agent council, maintains this site, and sends branded email. That number is not a marketing claim. It's a monthly line item total across the tools I actually pay for. Here are the field notes on what the stack contains, what each tool costs, what it replaced, and what I've tried and cut.
The stack serves a specific shape of practice, not every practice. If you run this shape (solo operator across brand, frontend, martech, and cloud ops), the numbers may be useful. If your shape is different, the structure still translates; the specific tools probably won't.
2026-04-10: the line items
I pulled my last statement and totaled the recurring work tooling. Not the hardware, not the office, not anything consumed outside work. Just the monthly AI and infrastructure subscriptions that let the practice operate.
Claude Code Max at $100 per month. Vercel Pro at $20 per month. Supabase Pro at $25. Stripe is pay-as-you-go so the fixed cost is zero; transaction fees are passed to buyers on the product pages. Resend at $20 for the email tier I need. GitHub at $4. A handful of smaller tools round out the rest. The total hovers around $175-210 depending on the month.
The number that carries the most weight is Claude Code. It's half the stack's cost and roughly all of the productivity differential compared to working without agents. Cut that and the stack is $100 a month but the practice produces materially less. Keep it and the other tools exist to make Claude Code useful.
2026-04-11: what each tool actually earns
Claude Code Max earns its line item because it runs the agent council, the development work, and most of the writing drafts. I described the council pattern in the agent council pattern article. On a typical day the council produces the morning triage in under ten minutes and the dev work runs through Claude Code's file-aware workflow for three to five hours. The $100 is paid back before lunch.
Vercel Pro hosts this site, the product checkout surfaces, and a couple of client preview environments. The Pro tier is about preview URLs, larger concurrency limits, and longer build minutes. I could run this work on the free tier for a while. The moment I added token-gated delivery pages, the upgrade paid for itself in one webhook that did not stall. The architecture I use is documented in the one-person studio stack.
Supabase Pro replaces a database subscription, an auth service, and a file storage bill that would individually cost more than the $25 combined. Row-level security lets me ship multi-tenant products without a dedicated auth architecture session before each launch. The dashboard is good enough that I don't need a separate DBA in my head.
Resend at $20 runs transactional and lifecycle email. I've tried SendGrid, Mailgun, and Postmark for this role over the years. Resend's developer experience is the only one I've enjoyed. Deliverability is currently fine. If it drops I will re-evaluate; deliverability is the only thing that matters for transactional email, not price.
2026-04-12: the tools I tried and cut
Cursor at $20 per month. I used it for six weeks. It's a good editor. It is not Claude Code. The difference is whether the agent holds the full file tree and can coordinate changes across five files in one invocation. Cursor's agent mode has improved, but the workflow still felt like assisted coding rather than agent-driven coding. I cut it.
GitHub Copilot at $10 per month. Same reasoning as Cursor, more granular. It's a fine autocomplete. It does not replace Claude Code for architecture work. Cut.
A dedicated AI writing tool at $30 per month. The council produces drafts that are roughly as good as this tool's output, for the price I was already paying Claude. The tool was redundant. Cut.
Notion AI at an additional $10 per month on top of Notion itself. I don't use Notion as a primary work surface so the AI layer was theater. Cut.
An analytics tool at $49 per month that claimed to unify Meta, GA4, and Shopify data. It did, but I can get most of what I need from the DTC stack audit methodology approach and Shopify's own reporting for the kind of clients I work with. Cut for now; may revisit at higher engagement volume.
Net effect of cutting: about $120 per month saved, minimal productivity loss, and one less set of logins to maintain.
“Keep what earns its line item every week. Cut everything else. The stack that works is smaller than the stack you think you need.
”
2026-04-13: the tools I keep thinking about adding
A dedicated observability tool (Datadog, Sentry Pro, LogRocket) at $50-100 per month. For the current scale, Vercel's built-in logging plus Supabase's own observability covers the 90% case. I'd add one if the practice took on a client with a five-figure monthly ad spend and real uptime stakes. For now it's optional.
A dedicated AI code review tool. I've looked at several. Claude Code does the review pass cleanly enough that a separate tool feels like duplication. May add if the repo grows past a size I can hold in context.
A design system tool (Figma Pro tier with their AI features). I use Figma for client mockups but most of my design work happens in code now. The free tier works. Will add a paid seat if I take on work that requires handing specs to outside developers at pace.
What the week taught me
Three things about the $200/month stack hold across iterations.
The first. The cost of a tool is not its subscription. It's the subscription plus the context tax of keeping it in your workflow. A $20 tool that requires ten minutes a day to manage costs more than its list price. A $100 tool that disappears into the workflow (Claude Code, Supabase) is cheap at double the price.
The second. The stack is opinionated because the practice is opinionated. If I took on a client that required a different frontend framework or a non-Shopify commerce platform, some of these tools would change. The architecture I described in the creative-tech operator playbook assumes a specific shape of work. The tools fit that shape; they are not universally correct.
The third. The gap between $200/month and $2,000/month of tooling does not buy proportional productivity. Most of the productivity lives in the first $200. The next $1,800 buys options I am not currently exercising. That ratio is worth checking whenever anyone tries to sell you a more expensive stack.
How I review the stack
Once per quarter I do a line-item pass. Each tool has to answer: did you earn your cost this month? If the answer is no for two consecutive quarters, it gets cut. The cut is permanent; if I need the tool back, I add it with a specific reason and an explicit cost target.
The quarterly review is also when I look at what I'm working around. If I keep reaching for a function that exists in a tool I'm not paying for, that's the candidate for next quarter's addition. The most recent example was background job processing, which led me to add a lightweight queue on top of Supabase rather than adopting a third-party job service at another monthly line item.
The review keeps the stack honest. Without it, subscriptions accumulate silently and the stack doubles in size over a year without doubling in output.
Frequently asked questions
Could I run this stack on less than $200?
Yes, with tradeoffs. The Claude Code tier is the biggest line item; you can get by on the smaller plan at $20 per month if your workload is light. Vercel and Supabase both have useful free tiers that work until you hit a specific scale. A minimum viable version of this stack is closer to $30 per month. The gap between that and $200 is the productivity difference, not just the feature count.
Why Claude Code over ChatGPT or Gemini?
Claude Code runs in a terminal and can coordinate file writes across a multi-file codebase. That workflow is materially different from a chat interface. ChatGPT and Gemini are strong models; they are not structured for the architecture work that fills most of my day. If your work is more research and writing than code, a different tool might fit better.
How often does the stack change?
About one substantive change per quarter. Small tools rotate more often. The core (Claude Code, Vercel, Supabase, Stripe, Resend) has been stable for over a year at this point. The infrastructure layer rarely changes once it's working; the layer around it adjusts as the practice adjusts.
Do I need to be a developer to run this stack?
Most of these tools require at least technical-marketer level comfort with code. Claude Code assumes you can read a diff. Supabase assumes basic SQL or at least the willingness to learn it. The stack is not for non-technical operators; a different tool set would suit that audience better.
What's the step up from this stack?
At higher scale the practice adds a dedicated observability tool, probably a CI/CD pipeline beyond Vercel's defaults, and maybe a job queue or cron service. That pushes the stack toward $400-600 per month. Beyond that, the tools start replacing human specialists instead of augmenting one operator, which is a different operating model. I've written about the shape of that transition in the One-Person Studio OS.
Sources and specifics
- Monthly total of approximately $175-210 pulled from the recurring work-tool subscriptions on a personal statement as of April 2026; numbers exclude hardware, office, and non-work services.
- Stack is optimized for solo creative-tech practice shape described in the creative-tech operator playbook; different shapes of practice will have different optimal stacks.
- Claude Code Max tier referenced at $100 per month corresponds to the mid-2026 Anthropic pricing; subject to change.
- Tools cut over the past 12 months include Cursor, GitHub Copilot, Notion AI, an AI writing tool, and a unified analytics service; savings estimated at approximately $120 per month.
- Quarterly line-item review process has been in place for roughly 18 months and has prevented at least four subscriptions from becoming permanent residents.