Running an AI business sounds like it should be cheap. Automate everything, scale without hiring, keep your margins fat. That pitch is mostly true. But the cost structure is nothing like what most people expect when they start.
This is not a generic breakdown. These are the actual numbers from running Xero AI, a solo company built almost entirely on AI agent infrastructure, with a human (me) doing the strategic direction and approvals. No team. No office. Evo, my AI co-founder, handles research, content, customer discovery, publishing, and operations.
Here is what it actually costs, line by line.
What Does It Actually Cost Per Month to Run an AI Agent Business?
The total monthly burn to run the full Xero AI system sits between $280 and $420, depending on API usage. That covers a live AI co-founder running daily content, Reddit discovery, social scheduling, and newsletter ops with no additional headcount. Here is every line item, no rounding.
Infrastructure (always-on)
| Tool | Monthly Cost | What It Does |
|---|---|---|
| OpenClaw | $97 | AI agent runtime, memory, cron, tool routing |
| Supabase | $25 | Database, blog storage, API |
| Netlify | $0 (free tier) | Site hosting + Netlify rebuilds |
| Postiz | $29 | Social scheduling (Twitter, TikTok, threads) |
| MailerLite | $0 (free tier, <1k subs) | Newsletter send |
| Xero Scout (internal) | $0 (we built it) | Reddit discovery and draft replies |
| PostHog | $0 (free tier) | Analytics |
| Cloudflare | $0 | DNS, CDN |
AI APIs (variable)
| API | Typical Monthly | Notes |
|---|---|---|
| Anthropic (Claude) | $40-$90 | Primary reasoning model for long tasks |
| OpenAI (GPT) | $20-$50 | Image generation, fallback tasks |
| Perplexity API | $10-$20 | Research and web search within agents |
| ElevenLabs | $0 (paused) | Voice experiments, not active |
One-time or annual costs amortized
| Item | Cost | Amortized/mo |
|---|---|---|
| Domain (xeroaiagency.com) | $15/yr | negligible |
| Lovable (site builder) | $32/mo | $32 |
| Figma (design, rarely used) | $15/mo | $15 |
Total: $269 minimum, $420 upper range on heavy content months.
That number buys a system that publishes daily blog posts, cross-posts to dev.to, monitors Reddit for customer opportunities, schedules social content, sends newsletters, and runs a full discovery loop without anyone babysitting it.
Where Does the Real Cost Actually Come From?
The actual constraint in running this kind of business is not the monthly bill. It is configuration overhead upfront and ongoing judgment calls. The tools are cheap. The time to build the system correctly, define the decision rules, and calibrate what the agent can do autonomously is the real investment.
Evo runs most days without me touching anything. But when something breaks, or a new workflow needs defining, or a product decision needs making, that lands on me. The human-in-the-loop cost is real even if it is not a line item.
The first month building this out took about 40 hours of my time. Setting up the memory files, defining the decision frameworks, writing the identity document, calibrating what Evo approves autonomously versus what comes to me in Telegram. That is a genuine investment. Most guides skip this part.
After that initial build, the ongoing time cost sits around 30-60 minutes a day. Reviewing what shipped, approving social content queued by Evo, adjusting strategy based on what is working. Some days zero minutes if nothing needs review. Heavy weeks when I am building something new might hit 3-4 hours.
If your time is worth $100/hour, 30 minutes a day is roughly fifteen hundred dollars per month in implicit cost. The economics still work if the business generates more than that. But the idea that you just turn it on and walk away is not accurate for most people at the beginning.
What Surprised Me Most About Running This Stack?
Three things caught me off guard in the first three months: API costs were far lower than expected, integration breakage was the hidden time sink, and the upfront context-building work took longer than any tool setup. Each one is worth unpacking.
API costs are not the problem. Before building this I assumed AI API costs would be brutal. They are not. Most Evo operations are relatively lightweight in token terms. The expensive tasks, like long research loops or generating full blog post drafts, happen once and produce durable output. You are not re-running them constantly. My Anthropic bill has never exceeded $90 in a month.
The hidden cost is integration tax. Every tool that touches another tool costs time when it breaks. Supabase schema changes, Netlify rebuild hooks, dev.to API quirks, the Postiz scheduling window. Each integration works until it does not, and debugging broken integrations is not glamorous work. Budget 2-3 hours per month for things that stop working unexpectedly.
Storage costs nothing until it suddenly matters. The blog now has nearly 50 posts, each with an OG image sitting in Supabase storage. At current scale, storage is free. At 500 posts with video assets, that changes. It is not a problem now but worth knowing the trajectory.
Some tools I bought and never used. Honest accounting includes the wasted spend. ElevenLabs was $22 for a month I experimented with voice content, produced nothing publishable, and paused the account. Figma costs $15/month and I use it maybe once a month. These are small but they add up over 12 months.
What Are the Unit Economics of an AI-Run Solo Business?
Cost per blog post published runs roughly $3 to $5 when you divide the monthly stack across the 20 to 25 posts Evo produces. Reddit opportunity drafting costs between $0.50 and $2 per lead in API usage. Social scheduling costs fractions of a cent per post. These numbers put individual output costs well below any freelancer or agency equivalent.
Cost per blog post published: roughly $3-$5 when you divide the monthly stack across the 20-25 posts Evo publishes. That includes AI API time, hosting, image generation, and cross-posting.
Cost per Reddit opportunity spotted and drafted: roughly $0.50 to $2 in API costs for the Scout workflow.
Cost per tweet scheduled: negligible. Postiz handles scheduling, and the generation cost per tweet is fractions of a cent.
The real question is not the cost per output. It is whether those outputs convert into revenue. One consulting call booked from an organic blog post covers three months of the entire stack. That math is why this model exists.
What most founders underestimate
You need more context management than you think. An AI agent that does not know your business well will produce generic output. The investment in writing identity files, source-of-truth documents, and decision frameworks is not optional overhead. It is what makes the difference between output you can use and output you have to rewrite.
I spent roughly 8 hours in the first two weeks writing the Xero SOUL.md file, product knowledge base, and decision frameworks. That work compounds every day. Every task Evo runs now benefits from that context. Skipping it to save time at the start costs you far more time later. There is a post on how to write an identity file for your AI agent if you want to see exactly what goes into this.
Guardrails take time to calibrate. Early on, Evo would occasionally draft social posts in a tone that did not sound like me, or schedule things I had not approved. The fix was not a different AI model. It was better guardrail rules and clearer approval gates. If you want to see how to build those, the AI agent guardrails guide covers the actual system we run.
You cannot automate judgment. Distribution decisions, pricing, what to build next, how to respond to a customer complaint. These require a human. The agents handle execution and surfacing information. The calls on what to execute are still mine. Any honest account of running an AI agent business should say this clearly.
Is Running an AI Agent Business Actually Worth the Cost?
For a solo founder building in public with a clear offer and specific audience, yes. The economics work at a level that no comparable human team could match at this price point. One consulting call booked from organic search covers three to four months of the full stack. But that only holds if the strategy underneath is sound.
For a solo founder building in public, yes. The leverage is real. One person operating with this infrastructure can produce the content volume, distribution presence, and customer discovery coverage that would otherwise take a small team.
The economics work if you are building something that compounds. Each blog post that ranks brings traffic indefinitely. Each Reddit comment that helps a potential customer builds reputation that does not expire. Each workflow you define and hand to Evo frees you for higher-leverage work.
The economics do not work if you treat it as a magic business generator. The agents are only as good as the strategy behind them. If you have not figured out who your customer is, what problem you solve, and how you will get them, automating your output volume just means producing the wrong content faster.
The real ROI question is not what the tools cost. It is what your judgment is worth when you free yourself from the execution work.
What Is the Cheapest Way to Start an AI Agent Business?
If you are evaluating this model and want to start without the full stack, the minimum viable version costs around $97/month (OpenClaw) plus API costs that start near zero and scale with usage. Add Supabase on the free tier for storage and data. That is the core.
The starter guide at Your First AI Agent is $7 and walks through setting up the memory and identity layer that makes the rest actually work. That is the piece most people skip and then wonder why their agent produces generic output.
For a broader look at how solo founders are structuring these systems, the State of AI Agents report from AIMultiple tracks how usage and cost patterns are shifting across industries. Research from Andreessen Horowitz on AI infrastructure spending consistently shows API costs are not the bottleneck for early-stage AI projects. Context quality is. That is where to invest first.
Build the context foundation first. The tools are the easy part.
Published by Michael Olivieri / Xero AI
Start Building Your Own AI System
- Your First AI Agent - $1 launch-test guide, instant download. The fastest way to get started.
- Build an AI Co-Founder - the full architecture ($19).
- AI for the Rest of Us newsletter - practical AI 3x/week for people with day jobs.
Want to build your own AI co-founder?
I'm building Xero in public — an AI system that runs distribution, content, and ops while I work a full-time job.
- Start here: Your First AI Agent — $7 guide, instant download
- Go deeper: Build an AI Co-Founder — the full architecture ($19)
- Newsletter: AI for the Rest of Us — practical AI 3x/week for people with day jobs
- Site: xeroaiagency.com
Originally published at xeroaiagency.com
Top comments (0)