The Anthropic Stack: How I Run 17 Websites as a One-Person Operation (2026)

What is the Anthropic Stack, and what does running 17 websites on it actually look like?

The Anthropic Stack is three Claude products used in concert — Claude Chat for thinking and writing, Claude Cowork for multi-app file and knowledge work, and Claude Code for code, deploys, and infrastructure — wired together with a small set of connective tools (N8N for automation, Airtable for data, Telnyx for voice, Fastmail for email, Notion as a second brain). I run all 17 active Bowman Web Services properties on it as a one-person operation. Real numbers, real workflow, real tradeoffs in this article. No inflation, no marketing.

I’m an operator. Trades background, 25 years in business, the last 15 of those in SEO and web operations. I’m not an Anthropic employee. I have no relationship with Anthropic. They don’t pay me. I’m a paying customer (Claude Max plus the API for production workloads) like everyone else. I’m writing this because the cluster of articles I’ve published on Claude AI, Cowork, Claude vs ChatGPT, and Claude Code for non-developers all pointed at the same question — what does it actually look like when you stitch the three together and run a real business on top of them?

This is the answer. The synthesis piece. What the stack is, why three tools instead of one, what the day-in-the-life looks like, what it costs me, where it falls short, and who should and shouldn’t build something like this.

The three-mode mental model (the cluster recap)

If you’re new to the cluster, the foundation reads in this order:

• What Is Claude AI? — the cornerstone. Defines Claude itself, compares it honestly to ChatGPT and Gemini, and frames why a small business owner should care.
• What Is Claude Cowork? — the niche-term article. Cowork is the desktop file-and-knowledge-work surface, sometimes called Claude Operator or Claude on the Desktop depending on which release notes you read.
• Claude vs ChatGPT for Small Business — the honest head-to-head. Where Claude wins (writing, analysis, longer context, lower hallucination rate). Where ChatGPT wins (image generation, voice mode, plugins).
• Claude Code for Non-Developers: The Contrarian Case — the surprise. Claude Code is officially marketed for developers and is increasingly used by non-developer operators who run code repositories whether they call themselves developers or not.

The stack is the synthesis of all four. Three modes of Claude, picked by use case:

Chat is the thinking surface. Browser-based, conversational, the one most people start with. Right for drafting, brainstorming, research summaries, planning, content writing, and general decision support. If a task starts with “help me think through” or “write me a draft,” Chat is the answer.

Cowork is the desktop surface. Has a filesystem, can move documents around, can integrate with apps via connectors, can run multi-step workflows that touch multiple tools in sequence. Right for document creation, file organization, knowledge-base updates, client onboarding workflows, and anything that lives in actual files on actual disks. If a task starts with “process these files” or “pull these documents and produce a summary,” Cowork is the answer.

Claude Code is the terminal surface. Reads code repositories, edits files based on plain-English instructions, runs commands, ships deploys. Right for website updates, schema work, deploy fixes, multi-repo refactors, infrastructure changes, and anything with a git history. If a task starts with “push this fix to all 17 sites” or “add this schema block site-wide,” Claude Code is the answer.

The decision rule I use is simple: Chat to think, Cowork to handle, Code to ship. Most actual work touches all three.

The portfolio (what runs on the stack)

Here’s what’s currently under management. Sites are mixed across vertical and intent — content, lead-capture, journalism, travel, and a satellite city-AI network. The variety is the point. It stress-tests every workflow.

• BowmanWebServices.com — the agency main site. Services, pricing, blog (this article lives here), about, contact.
• SandosPromo.com — vacation club promotions for the Sandos Hotels & Resorts portfolio. Bilingual EN/ES with two Telnyx voice agents handling phone leads in both languages.
• DaytonaHandy.com — local handyman lead capture for the Volusia County market. Telnyx AI voice agent (Erica) answering calls 24/7.
• AllegedFraud.com — investigative journalism property focused on consumer protection and fraud watchdog work. Citizen-journalism stack with content-reporting workflows.
• IWantToTravelTo.com — travel destination property, 13 years old. Mix of editorial content and lead capture for travel partners.
• SelfGrowthVideos.com — curated personal-development video library. Six teacher profile pages live, more in the pipeline.
• MexicanTravelVIP.com — bilingual EN/ES Mexico travel demo with the Maya voice agents.
• VillaPromo.com — partner-pending villa rental property. On hold.
• VacationClubPromo.com — multi-resort showcase with eight built-out resort doctrine pages plus a 25-article Puerto Plata things-to-do series.
• FlFenceAndScreen.com — Florida fence and screen contractor lead site (client property, AI-first rebuild from WordPress).
• DaytonaBeachMassages.com — local massage therapy lead site with Telnyx AI booking agent.
• ResortRevenueEngine.com — 6-division white-label resort revenue platform (concept demo).
• RFKTeamReport.com — recently acquired chronicle property with ESPN-style scorecard format.
• The Volusia AI Network — five satellite city-AI lead-capture sites under one monorepo: DaytonaBeachAi.com, PortOrangeAi.com, OrmondBeachAi.com, NewSmyrnaBeachAi.com, DelandAi.com.

That’s seventeen active properties. Different stacks underneath, different audiences, different conversion goals. All of them deployed on Hugo and Netlify with GitHub as the source of truth, and all of them maintained by one operator (me) using the Anthropic Stack as the work surface.

A day on the stack

Here’s what an actual workday looks like. This is not aspirational. This is the rhythm I’m in three to five days a week.

5:30 AM — coffee plus the morning briefing.

Telegram fires overnight alerts from @BWS_alerts_bot. Last night’s lead pipeline arrivals, GSC anomaly alerts (any site that lost more than 30 percent of its clicks day-over-day), N8N workflow execution health, and any inbound voice calls that the Telnyx agents handled while I was asleep. I read those on my phone before I’m at the desk. By the time I’m in the chair, I already know which fires need attention and which days look quiet.

6:00 AM — Chat session, the planning conversation.

I open Claude Chat and ask something like “What’s pending across the portfolio today? Pull from my Notion ops hub and my last three days of session logs.” Chat reads the relevant Notion pages (the ops hub, the session log, the active queue) and gives me a prioritized list. I push back, refine, lock the day’s plan. Twenty minutes total. I have a written list of 5 to 8 things I’m shipping today, ordered by the dependency chain.

7:00 AM — Cowork session, the file work.

Overnight, three new lead intakes landed in Airtable and a client emailed me a logo update plus three blurry product photos. I open Cowork and walk it through: pull the lead intakes into a daily summary doc, draft personalized first-contact emails for each (using the right tone for each segment), update the client’s Hugo content file with the new logo path and resized photos, and stash everything in the right Notion project folders so the session log is clean. Cowork reads files, edits files, sends emails through Fastmail, writes to Notion, and reports back when the multi-step workflow is done. Forty-five minutes for what used to be a half-day of clicking around in seven different apps.

8:30 AM — Chat, the writing block.

The day’s article (or client content piece, or service page rewrite, or longer-form proposal) gets drafted in Chat. Long-context conversational writing is where Claude shines and where I do my best work with it. A 4,000-word article like this one is typically about 90 minutes of back-and-forth in Chat, with me steering voice, calling for revisions on specific sections, and locking the structure as we go. I don’t accept drafts wholesale — every article gets a manual editing pass for voice and accuracy. But the heavy lift on outline-to-draft is Chat’s territory.

10:00 AM — Claude Code, the ship block.

Today’s article needs to go live. That means:

1. The markdown file with frontmatter, body, internal cross-links, and JSON-LD schema (Article + FAQPage)
2. The featured image (1200x630, family signature, on-pattern with prior cluster articles)
3. An atomic two-file commit via the GitHub Trees API (one Netlify build, no rolling-deploy 404 risk)
4. Live verification — the URL returns 200, the schema validates, the featured image fetches, the sitemap updates

Claude Code does all four in one session. I describe the article, paste the body, describe the visual signature, and Claude Code writes the markdown, generates the image with PIL, builds the Trees API payload, commits both files atomically to bowman-web-services/bws-site, and verifies live within ninety seconds. Total time from “article ready” to “article live and indexed”: about ten minutes.

11:00 AM — voice agents handle inbound.

While I’m in the build, the phone is covered. SandosPromo’s English and Spanish lines route to two different Telnyx voice agents. DaytonaHandy’s number routes to Erica. The BWS main line routes to Alex. Each agent has a post-call webhook that fires a transcript at N8N, which extracts the lead via Claude Haiku, normalizes the fields, writes to the right Airtable table, and Telegrams me a summary. By the time I look at my phone after the build, I have three new leads with structured data in Airtable and I’ve read the conversation transcripts already.

12:30 PM — break, walk, real food.

The stack runs on its own for a couple of hours. I’m not chained to it.

2:00 PM — Cowork or Code, whatever the afternoon needs.

This block flexes. Some afternoons it’s a multi-site campaign push — schema rollout across all 17 properties, policy page refresh, footer initiative, IndexNow key deployment. That’s a Claude Code afternoon, often using the same atomic Trees API pattern across multiple repos. Other afternoons it’s a content factory run — pull a list of subjects from Airtable, research each one through Cowork, build the playlists or content pages, stage them for review. Afternoon flex is where the productivity ceiling actually is.

5:00 PM — review, ship, walk away.

End of day, I look at:

• What shipped (commits to which repos, articles published, leads converted)
• What’s blocked or carried forward (always something — the day never ends clean)
• The session log gets a timestamped summary written to Notion
• The next morning’s Active Queue gets updated

Then I close the laptop. The voice agents keep running. The N8N workflows keep firing. The deploys keep building. The stack keeps doing its job overnight, and tomorrow morning I do the briefing again.

That’s the day. Not glamorous. Not magical. Disciplined repetition of “use the right tool for the right thing, ship work, sleep.”

The numbers (real, no inflation)

Here’s the part I have to be careful with, because the temptation is always to inflate. I’m not going to.

• 17 active websites under management. Seven of those are revenue-bearing. Ten are content, journalism, lead-capture, or platform plays.
• One operator. Me. No employees. One sales partner (Greg) on SandosPromo, on commission.
• Five paid Anthropic tools active in the stack. Claude Max for Chat and Cowork in the desktop app, Claude Code via the same plan, Claude API for production extraction work in N8N, and the API for some heavier enrichment runs in Cowork projects.
• About 25 custom Cowork Skills installed. These are little procedural rulebooks I’ve written over time — when this kind of work comes up, follow these steps, use these patterns, avoid these traps. Every BWS skill is on file in the BWS Operations Hub on Notion.
• 137+ active N8N workflows. Not all of them fire daily. Some are weekly reports, some are on-demand webhooks, some are health checks. But the count is real and growing.
• Seven Telnyx voice agents in production. Two SandosPromo (EN, ES), two MexicanTravelVIP (EN, ES), Erica for DaytonaHandy, Alex for BWS Main, and the Daytona Beach Massages agent.
• Site builds. Articles like this one go from draft to live in roughly an hour total. Full website rebuilds (the Volusia Network monorepo with five spokes, for example) used to be a six-week project for a small agency. The five-spoke monorepo went from blank repo to all five sites live and serving 200s in two work-weeks of evenings.
• Multi-property campaigns. A site-wide schema rollout across all 17 properties was three days. Pre-stack, it would have been six weeks of grinding through sites one at a time.

The honest gain isn’t “I’m 100x more productive.” The honest gain is “I can work on the things that need attention each day instead of being permanently stuck in maintenance debt.”

The connective tissue (what makes the stack a stack)

The three Claude products are the work surface, but they sit on a small set of connective tools that turn isolated work into running systems.

N8N is the automation backbone. It’s a self-hosted workflow tool that connects everything to everything. It fires the daily GSC pull at 6 AM, processes voice agent post-call webhooks, writes leads to Airtable, sends Telegram alerts, syncs Brevo signups, runs anomaly checks, fires content generation drips, and dozens of other things I don’t have to babysit. Self-hosting on a $9/month Hostinger VPS means I own the infrastructure outright instead of being capped by Zapier or Make tier limits.

Airtable is the operational database. Every BWS property has a base. Leads, content queues, video catalogs, GSC pulls, health metrics, citation status, every structured data store I work with lives in Airtable. It’s not the right answer at scale (the 50,000-row cap will bite eventually and the next move is Supabase), but for an operation this size it’s the right ratio of usability to power.

Telnyx is the voice and phone layer. Phone numbers, AI voice agents, post-call webhooks, programmable routing. Native MCP integration with Claude means I can audit assistant configurations from Chat and patch them via the API without leaving the work surface. Lead capture happens through Telnyx voice while I sleep.

Fastmail is the email layer. JMAP API access means I can have Cowork send emails on my behalf without navigating to a webmail UI. Email aliases per property mean each business has its own inbox without me paying for nine separate Google Workspace accounts. About $5 a month, all in.

Notion is the knowledge base. Operations hub, session logs, repo registry, master infrastructure reference, skill registry. Every fact that the stack needs persists in Notion. When I start a new Chat session, the very first thing the Chat reads is the BWS Operations Hub root page. That’s how an AI tool with no permanent memory still feels continuous.

GitHub is the source of truth for code. Seventeen private repos under the bowman-web-services organization. Every site auto-deploys from GitHub to Netlify on push to main.

Netlify is the hosting. Free tier handles most of it (with the occasional brush against the bandwidth cap on busy months). Auto-deploy from GitHub main is the canonical workflow. Pro tier waits for revenue.

The stack metaphor is literal: Claude products on top, connective tools in the middle, infrastructure (GitHub, Netlify, the Hostinger VPS) on the bottom. Each layer has a job. Each layer talks to the layer above and below through clean APIs. None of it is unique to me — every piece is documented, replaceable, and standard.

Where the stack falls short (the honest part)

I don’t run an Anthropic ad agency. The stack has real limits and real failure modes.

Mode confusion. The hardest skill in the stack is knowing which Claude surface to reach for. New operators reach for Chat for everything (because Chat is what they know) and end up doing slow file work in a chat window. Or they reach for Code for content drafting (because Code feels powerful) and end up overcomplicating simple writing. The decision rule “Chat to think, Cowork to handle, Code to ship” is something I had to learn through misuse.

Concentration risk. Five Claude products and you’ve got a single-vendor dependency. If Anthropic’s API has a bad day, my whole work surface goes degraded. I’ve had two such days this year — both resolved within hours, but it’s real. The mitigation is partial: have at least one ChatGPT subscription as a fallback for thinking, keep the connective tissue (N8N, Airtable, Telnyx) on diverse vendors, and have the discipline to sometimes step away from the keyboard when the stack hiccups instead of trying to fight it.

Cost compounds. The Anthropic tools alone run me about $200 a month between Max and API. Add the connective tissue (Hostinger $9, Telnyx variable but roughly $50-100, Airtable $20, Fastmail $5) and the all-in is $300-400 a month. That’s real money. It’s also less than a part-time admin would cost, but it’s not free, and somebody has to be making at least a little revenue from it for the math to work.

Voice and image still need ChatGPT. I covered this in the Claude vs ChatGPT comparison article, but it stays true. For image generation that has to integrate with brand work, I still pay for ChatGPT. For voice mode (real-time conversational AI on the phone, hands-free) I still use ChatGPT. The Anthropic Stack isn’t trying to be everything — it’s trying to be the operations spine. Voice and image are wings, not core.

Echo chamber risk. Claude is excellent at agreeing with you, asking clarifying questions, and sounding reasonable. If you’re not careful, a stack-heavy day can leave you talking only to Claude, your own past notes, and your own data. You stop talking to humans. That’s bad for business and bad for you. I make a point of getting on calls, going to local AI meetups (the Daytona AI Explorers Facebook group I run is partly self-medication for this), and keeping at least one human conversation a day on the calendar.

Things break in invisible ways. A Hugo Future-Date Timezone trap silently dropping pages. A Netlify rolling-deploy queue leaving the latest commit in 404 state. A function-invocation quota cap freezing the entire portfolio for hours. The stack is fast, but when it breaks, it breaks in subtle ways that don’t trigger alerts. The skill that compounds over time is “verify live, don’t trust the deploy state.” Read every commit’s actual rendered URL. Check for ghost pages. Run the audits.

These aren’t dealbreakers. They’re real costs of operating this way that you should know before you decide to.

What it costs me to run

Putting numbers on it for honesty:

Compare that to hiring even one part-time virtual assistant at $25/hour for ten hours a week — that’s $1,000 a month, and a VA can’t write your code, deploy your sites, or answer your phones in two languages.

The math works. It works because the stack absorbs the kind of work I’d otherwise be paying multiple humans to do.

Should you build a stack like this?

Honest answer, by situation:

You’re a solo operator with multiple online assets (multiple websites, content properties, courses, services) — yes. This is the use case the stack is built for. The productivity ceiling moves up dramatically once you stop being the bottleneck on every individual property.

You’re a small agency (one to three operators, multiple clients) — probably yes. The connective tissue (N8N, Airtable, etc.) still pays off. The challenge is permission boundaries — multiple humans on the stack means more coordination, more skill-writing, more documentation discipline. Worth it, but the setup cost is real.

You’re a non-technical small business owner with one website — not yet. Start with Chat. Get fluent. Maybe add Cowork after six months for documents and email. Don’t reach for Claude Code until you have at least one repository you understand. The stack is a power tool for people who already manage technical work; it’s not a beginner platform.

You’re a fast-scaling business with a real team — hybrid. Use the stack for ops (briefings, content drafting, analytics, automations) and hire humans for the relationship work the stack can’t do (sales calls, partnerships, big creative bets, leadership). The stack is a force multiplier for individual operators; it’s not a substitute for human team-building when team-building is the right move.

You’re an enterprise — different conversation. The Anthropic enterprise tier is real, the API has volume pricing, and the build pattern at that scale is more about wiring Claude into existing systems than building a personal stack. Not my domain. Talk to Anthropic directly.

The single best test: do you currently have at least one online asset where you’re permanently behind on maintenance? If the answer is yes, the stack will pay for itself within a quarter. If the answer is no, you don’t need it yet.

Where to go from here

The cluster wraps with this article. If you’ve followed along, here’s the recommended reading sequence to revisit any piece:

• The cornerstone: What Is Claude AI? and What Is Claude Cowork? — the product foundations.
• The decision frame: Claude vs ChatGPT for Small Business — the head-to-head with concrete tasks.
• The contrarian piece: Claude Code for Non-Developers — the surprise. Most “non-developers” already manage code work in 2026.
• The AI search context: What Is GEO?, What Is AEO?, and the GEO vs SEO vs AEO comparison — why this article exists in this form, optimized for AI engines as well as human readers.
• The agency: the BWS services page for the AI-first stack we deploy and the pricing page for what we charge.
• The free audit offer: if you want to know what your business looks like inside ChatGPT, Perplexity, Claude, and Google AI Overviews, drop a line via the contact page and I’ll run your domain through all four. No pitch, just the receipts.

If you want to ask questions, the Daytona AI Explorers Facebook group is the local community we’re activating around this stack and adjacent topics. If you’re building something similar and want to compare notes, I’m always happy to.

This is the work I do. This is the stack that makes it possible. The compounding effect across years is what makes a one-person operation able to maintain seventeen properties without being permanently underwater. It’s not magic, it’s not for everyone, and it’s not free. But it’s real, and it works, and the productivity ceiling keeps moving up as Anthropic ships.

— Tom

⚠️ Disclosure: I have no relationship with Anthropic. They don’t pay me. This article isn’t sponsored. I’m a daily paying user (Claude Max plus Claude API) who runs his entire business on the resulting stack. The same disclosure has appeared in body copy on every article in this cluster. If Anthropic ever extends a partner or case-study relationship, I’ll update this notice on every published cluster article and call it out at the top. As of this article’s publish date, no such relationship exists.

Frequently asked questions

What is the Anthropic Stack in plain English? It’s the combination of three Claude products (Chat for thinking and writing, Cowork for desktop file work, Claude Code for code and deploys) plus the connective tools that wire them into a working operation (N8N, Airtable, Telnyx, Fastmail, Notion, GitHub, Netlify). Used together as the work surface for running multiple online businesses solo.

Why three tools instead of one? Different jobs have different shapes. Conversational thinking, multi-app file work, and terminal-based code work each benefit from a different surface. The same underlying Claude model is doing the work in all three; the surface is what changes. Trying to do everything in one surface is inefficient — it’s why most people end up tab-switching and copy-pasting between Chat and other tools anyway.

Can I run a business on the Anthropic Stack alone, with no ChatGPT or Gemini? Mostly, but not entirely. Anthropic has gaps — image generation, real-time voice mode, plugin ecosystem — that other providers fill. I keep a ChatGPT Pro subscription as the gap-filler. The Anthropic Stack is the operations spine; ChatGPT covers the wings.

How long does it take to learn the stack? Plan on six to twelve months to fluency if you’re starting cold. The first month is Chat. The second and third are Cowork. The fourth through sixth are connective tissue (N8N, Airtable). Claude Code comes in once you have at least one repository to manage. The skill of knowing which mode to use when is the slowest part to develop.

What if Anthropic raises prices or removes a feature? Real risk, real possibility, you’ve seen the recent pricing tests. Mitigation: keep your work portable (Cowork files are real files; Claude Code edits real Git repos; nothing is locked in proprietary formats), keep the connective tissue on diverse vendors, and budget that AI tool costs will probably go up not down over the next two years. The stack is more durable than the specific pricing is.

What’s the connective tissue between the three Claude tools? The actual data flow goes: Chat plans the work, Cowork executes the file/document/email side, Claude Code executes the code/deploy side, N8N runs scheduled and event-triggered automations between everything, Airtable holds operational state, Telnyx handles voice, Fastmail handles email, Notion holds knowledge. Every layer talks to every other layer through standard APIs. No proprietary glue.

How do you handle outages? Three answers. (1) Most outages are short — Anthropic’s API SLA is real and recovery is usually under an hour. (2) The connective tissue runs on diverse vendors so a Claude API outage doesn’t take down N8N, Airtable, Telnyx, or my deploys. (3) When the work surface is degraded, I work with what’s still up — focused work on existing material, calls, reading, walking the dog. Avoid the urge to fight the outage. It comes back.

Is this realistic for someone who isn’t technical? Some of it. The Chat and Cowork halves are accessible to non-technical operators today. The Claude Code half assumes you already manage at least one code repository, even if you don’t think of yourself as a developer. If you’re truly non-technical with a single website on a managed platform like Squarespace or Wix, you don’t need Claude Code yet — start with Chat for content and customer communications, add Cowork for document and email work, and revisit Code in twelve months when you’ve outgrown the managed platform.