How to Build a One-Person AI Agency with 10x Output

TL;DR: Real case study: how Toi La Tung runs a professional agency alone with AI workflows — from design and coding to content and automated marketing.

In March 2025, I had a strange conversation with a prospective client.

They were looking for an agency to build a landing page and lead capture system. After we finished discussing requirements, they asked: "How many people are on your team? I need to know to assess your capacity."

I replied: "Right now I'm solo — but I have AI."

Silence. They seemed unsure.

Two weeks later, I delivered: a polished, responsive landing page with A/B testing setup, lead capture with email automation, and an analytics report. The client was happy enough to refer three more clients.

That was when I realized: a One-Person Agency with AI is not only viable — it can compete head-to-head with a traditional 5–10 person agency.

This post is a real case study of how I built and run Toi La Tung — from daily workflows to the specific AI tools I use for each type of task.

1. Why is a One-Person Agency viable in 2026?

Traditional agencies were built on one assumption: creative and technical work cannot scale without more people.

Want more landing pages? Hire more designers and developers. Want more content? Hire more copywriters. Want more clients? Hire more account managers.

AI broke that assumption.

In 2026, with the right tools, one person can do the work of ten in these domains:

• Web development: Claude/Cursor handles ~80% of the actual coding
• UI design: AI generates layouts; humans review and refine
• Content creation: AI drafts; humans edit and inject personality
• SEO research: AI synthesizes; humans set strategy
• Email automation: AI writes sequences; humans approve and deploy

The only things AI still cannot replace? Relationships, trust, and strategic judgment. And those are exactly what a One-Person Agency founder should focus on.

2. Capability map: what AI handles, what humans must do

This is the framework I use to allocate work:

AI does well (>80% quality, <10% of normal time):

• Write code from descriptions/wireframes
• Generate copy drafts from a brief
• Create A/B test variations
• Synthesize and analyze data
• Generate images/visuals from text prompts
• Set up automation workflows
• Debug and fix bugs
• Refactor code for better performance

Humans must do (cannot fully delegate to AI):

• Discovery calls with clients — listen and understand the real problem
• Strategic recommendations — propose solutions based on business context
• Creative direction — decide the right "feel" for the brand
• Quality judgment — know when AI output is good enough
• Relationship management — follow-up, support, upsell

Once you know this boundary, you can structure a workday for maximum leverage.

3. Toi La Tung's tech stack

Design & Prototyping

• Claude.ai: Wireframe idea → HTML/CSS prototype right in chat
• Figma: Only when the client requires a Figma deliverable
• Gemini Flash: Generate hero images and illustrations for landing pages

Development

• Cursor IDE: Primary code editor with AI integration
• Next.js: Framework for all new web projects
• Vercel: Deploy, preview URLs, edge functions
• Cloudflare: DNS, CDN, domain management

Content & Marketing

• Claude Sonnet: Long-form content (blog posts, email sequences)
• Claude Haiku: Short-form copy (headlines, CTAs, ads)
• Make.com: Automation — connect forms, email, CRM
• Airtable: Simple project management + CRM

Client Communication

• Loom: Demo videos instead of meetings
• Notion: Shared project documents
• Telegram/Zalo: Day-to-day communication

4. A real daily workflow

8:00 — Morning review (30 minutes)

• Check email and Telegram from clients
• Review today's tasks
• Identify which tasks AI can run alone vs. which need me

8:30 — Deep work block 1 (2.5 hours)

• The most important technical task of the day
• Cursor IDE open, CLAUDE.md already set up
• Work in a Zero-Gravity loop: prompt → review → adjust → ship

11:00 — Client communication (30 minutes)

• Reply to all messages in one batch (no interrupting deep work)
• Send updates, previews, or questions to clients

11:30 — Meeting/call block

• Only book meetings in this window
• Prefer Loom demos over live screen share when possible

13:00 — Lunch + recharge

14:00 — Deep work block 2 (2 hours)

• Content creation or a second technical task
• Usually AI drafts; I edit and finalize

16:00 — Admin & automation (1 hour)

• Check reports and analytics
• Set up / improve automation workflows
• Invoices and billing

17:00 — Learning & R&D (30 minutes)

• Read updates on AI tools
• Test new tools
• Capture insights into the knowledge base

Total: 8 productive hours, no overtime, no burnout spiral

5. Case study: a 3 million VND landing page in 3 days

Client: A B2B coaching company that needed a landing page for a new course

Requirements:

• Long-form landing page (hero + benefits + testimonials + FAQ + CTA)
• Signup form connected to Google Sheets
• Automatic confirmation email
• Analytics setup
• Deadline: 3 days

Day 1 — Discovery & Design (4 hours)

Morning: 45-minute discovery call to understand value proposition, target audience, and goals. I took notes in Notion.

Afternoon: Used Claude to draft the content outline:

Prompt: "Tôi đang làm landing page cho khóa học [X]. Target audience là [Y]. 
Pain points: [A, B, C]. Hãy tạo content structure cho một long-form landing page 
với: Hero (headline + subheadline + CTA), 3 Benefits sections, Testimonials, 
FAQ (5 câu hỏi phổ biến nhất), và Final CTA. Viết bằng tiếng Việt."

Claude drafted it in 5 minutes. I edited for 30 minutes. Sent the outline to the client for review.

Day 2 — Development (5 hours)

Used Cursor to build the full HTML/CSS/JS:

• Session 1: Structure and hero section
• Session 2: Benefits and testimonials
• Session 3: Form with validation and Make.com webhook
• Session 4: Mobile responsive + animations

Each session ~1 hour; AI wrote ~85% of the code; I reviewed and refined.

Day 3 — Polish & Delivery (3 hours)

• Set up Google Analytics 4
• A/B test variant for the headline
• Email automation with Make.com (3 emails: confirmation, day-3 follow-up, day-7 promotion)
• Final QA: mobile, browser, form submission, email flow
• Deployed on Vercel with a custom domain

Result: Delivered on deadline, client was happy, I worked a total of 12 hours for 3 million VND in revenue.

6. Real numbers: costs and revenue

Monthly operating costs:

• Claude Pro: ~$20 (~500k VND)
• Cursor Pro: $20
• Vercel Pro: $20
• Make.com: ~$10
• Other tools: ~$30
• Total: ~$100 (~2.5 million VND/month)

Achievable revenue:

• 4–6 landing page projects/month × 3–5 million VND = 12–30 million VND
• 1–2 web application projects/month × 10–20 million VND = 10–40 million VND
• Monthly content retainers × 3–5 clients × 3–5 million VND = 9–25 million VND

Potential total: 30–95 million VND/month at ~2.5 million VND in costs

These are not guaranteed figures — but they are numbers I see in Vietnam's AI One-Person Agency community. Real results depend on your skill, network, and market.

7. Limits of the One-Person Agency model

I want to be honest: this model has real limits.

Limit 1: Bandwidth You only have 8–10 hours a day. Even with heavy AI leverage, review, decisions, and relationships remain bottlenecks. Past ~5–7 active projects at once, quality starts to drop.

Limit 2: Enterprise clients Large clients (500+ employees) often need an agency with a team, SLAs, and multiple points of contact. A One-Person Agency is a hard fit.

Limit 3: Specialized domain skills AI is strong at web, content, and baseline design. Some areas still need deep human expertise: legal compliance, industry-specific strategy, crisis management.

Limit 4: Dependency risk If a core AI tool goes down (outage, policy change, pricing spike), your workflow is hit immediately.

Knowing these limits helps you pick the right clients and set the right expectations — instead of overcommitting and under-delivering.

8. How to start if you want to do the same

Step 1: Pick one service line to start

Don't try to do everything on day one. Choose the thing you are strongest at and AI can augment best. Example: "Landing page + lead capture" is a common, learnable combo.

Step 2: Learn the AI workflow for 2 weeks

Before taking clients, spend 2 weeks practicing:

• Run 3 simulated projects with the full process
• Test different tools
• Build a library of prompts and templates that work

Step 3: Get one real case study (even if free)

Your first case study matters more than the money. Do 1–2 free or low-price projects for people you know to get portfolio work and real testimonials.

Step 4: Price correctly from the start

Many beginners underprice because "AI did most of it." That is wrong — you charge for expertise, time, and outcomes, not "coding hours." Don't bill by the hour — bill by project value.

Step 5: Build systems, not just projects

After every project, record: which prompts worked, which workflows were efficient, which mistakes to avoid. Over time you get a personal playbook — and that makes the next project 30–40% faster.

9. Conclusion

A One-Person Agency with AI is not about "doing less" — it is about working smarter. You still need to invest time, keep learning, and deliver quality. AI is leverage, not a substitute for craft.

The question is not "Will AI replace agencies?" The question is "Who will use AI to build a better agency than everyone else?"

If you want to learn this whole framework systematically — from setting up AI workflows and writing prompts for each task type, to scaling a One-Person Agency — it is all inside Claude Code Mastery Pro.

Start today. This industry is moving fast, and the advantage goes to people who act early.

To strengthen your operating systems and systems thinking, also read: What is an AI Agent? The Clearest Explanation and apply it to your work right away.