Blog October 10, 2026 12 min read

How to Run Your Business With Claude Skills and Routines (No Code Needed)

By Alfonso Briceño
Published Last updated
How to Run Your Business With Claude Skills and Routines (No Code Needed)

In short: you can run most of an online business through Claude or ChatGPT. Turn each job you repeat into a skill (a long prompt with your voice, the steps and the tools), put the skill on a routine so it runs by itself, and connect your other apps with MCP connectors. You don't need to code.

This week I ran the first AI Club in Siargao. About eight people came, from complete beginners to people already building with AI: a social media agency owner, a former trader, a coach in the middle of a career change, someone who works in property management, and a website freelancer.

It was a free-flow session. I showed how I run my businesses through Claude and ChatGPT, and then we built a plan, a landing page and a first app for one of the agencies in the room, live.

This is everything we covered, written up so you can come back to it.

Who I am and why this matters

I have two startups. One is Gobo, which helps local businesses with AI by making little apps personalized for them. The other is Tribed, where experts and coaches get their own personalized app to monetize their audience.

Everything I do, I run through Claude and ChatGPT. I don't use any other software or interface now, and that's the only software I pay for.

I think it can get to the point where it runs an online business end to end: it finds your customer, signs them up, does the onboarding and the support, collects the feedback and improves your product.

1. Why natural language is the best UI in the world

Whatever you're doing on a computer, AI can now do it without you touching it. You just need to give it good guidance on how to reach the goal.

That changes how you use software. You don't need to learn every app, because Claude or ChatGPT can connect to almost every other app in existence. Software changes so fast now that keeping up with every interface is a waste of time.

A few examples from my own week:

  • Ads. Meta Ads' interface is horrendous. I don't go in there anymore. I say "make me an ad about this, upload it."
  • Flights. I even have it buy my flight tickets. I don't want to spend an hour finding the cheapest price, so I say "go buy it." My bank still asks me to verify the payment, which is how it should be.
  • Community. With a WhatsApp connector and a routine, you can say "every Thursday, remind my community about the dinner," and it happens every week.

What is an MCP connector?

MCPs are just APIs for AI. MCP stands for Model Context Protocol, an open standard Anthropic introduced on November 25, 2024. An MCP is a plug that lets Claude or ChatGPT read from and act inside another app: your email, your calendar, WhatsApp, your ads account, your own software. Once it's connected, you talk to the app in plain language instead of clicking through it. Browse what's available in Claude's connector directory, or add your own with a custom connector.

2. What is a Claude skill?

A Claude skill is a long, reusable prompt saved as a file. It holds the voice, the task, each step in order and the tool for each step, so a job a person used to do becomes a digital job. Anthropic launched Agent Skills on October 16, 2025.

I used to have a virtual assistant doing my outreach. Every day, a personalized message to a coach: "hey, I made this little app for you."

Now that's a skill.

A skill is a long prompt. It holds:

  • the voice (how it should sound)
  • what it needs to do
  • the first step, the second step, and so on
  • which tool it uses for each step

Instead of one person doing it, it's a digital job.

To see how real ones are written, read the Agent Skills docs and the example skills on GitHub.

3. What is a routine?

A routine is a skill on a schedule. It runs by itself every day, every week or on a set day, without you starting it. My outreach routine runs every day: it does the outreach, the follow-up, and makes a demo for whoever is interested.

The pattern works for almost any repeated task:

  1. Write down how you do it today, step by step, as if you were training a new hire.
  2. Put that into a skill.
  3. Put the skill on a routine (daily, weekly, every Thursday).
  4. Check the output until you trust it.

4. How do you make AI output reliable? Fix the skill, not the output

When AI gives you a bad result, don't fix that one result. Write the fix into the skill, so the same mistake never comes back. This is the most important idea from the night.

At the beginning, your skill will create outputs you don't like. The instinct is to open that one bad message and fix it by hand. Don't. If you fix the output, you'll fix the same output again tomorrow.

Go back and fix the skill. Every bad output is an edge case the skill didn't know about yet, so you write it in. Keep working on it, see all the edge cases, until you arrive at the ideal skill. Then you can trust it: "okay, go for it, I don't need to check you anymore."

Instead of training a person, you're training a skill file.

My advice is to skillify everything. Think in systems.

5. Why does AI write generic content?

If you ask Claude "make me some social media posts," it will give you the most mainstream posts possible, not adapted to you at all. AI is trained on all the mainstream data, so if you give it nothing about you, it gives you the average of everybody back.

The fix is two things:

  1. A good skill file with your voice, your structure, the phrases you'd never use, and examples of good and bad output.
  2. Your own input: your words, your stories, a voice note, a call, a transcript of you teaching. This article started as me talking in a room.

It may not get it right on the first try, but with a good skill and your own material it produces something decent, and it gets better every time you fix the skill.

You don't have to write every skill from zero. Good free ones exist, like Corey Haines' marketing skills and the "taste" skill for design. Judge a public skill by its stars and installs, the same way you'd check reviews. skills.sh lists public skills with their install counts.

6. How do you start building with AI without code?

You don't need to be technical, and you don't need to know code. Start from a messy idea, ask the AI for a plan before it builds anything, and expect the last 20% to take longer than the first 80%. Here is the process I showed:

  1. Brain-fart your idea into a new folder. Dump everything you're thinking, messy is fine.
  2. Ask for the plan, and don't let it implement yet. Read the plan and push back on anything that feels off.
  3. Then build.
  4. Expect the 80/20. The first 80% comes in one day. The skills, the refinement and the little edge cases are the last 20%, and that's what takes the longest.

The first day feels like magic, and that's why people think they're almost done. Plan for the second phase. That's where the real product gets made.

If you want to build apps and tools, Claude Code is what I use every day. Anthropic's prompt engineering guide is the best free reference on writing instructions that work.

7. Which AI model should you use?

Use the most capable model for the system and cheaper models for the doing. A simple rule:

  • System work (the plan, the routine, the skills everything else depends on): use the highest model.
  • Doing work (running the steps): use the lower, cheaper models.

Someone in the room put it well: you hire the best expert to write the business plan, and a cheaper person to carry out steps one to six.

Anthropic's model overview compares the models by capability and price. This matters because tokens run out. This week I burned way too many, and I was at 72% of my weekly usage by the day of the club.

8. How do you get clients with AI? Give value before you ask

Websites are cheap now. Part of my sales process is a message that says "hey, I made you a demo," with a landing page made for that person. It costs me almost nothing, and it gets me in the door with people who would ignore a pitch.

For coaches, the same move is an attractive lead magnet made for one person. You want them to say "wow, you did all that?" before you've asked for anything. Give value for free from the start, and let them think: imagine what you could do if I paid you.

That's what we did live at the club: a plan, a landing page and a first app for an agency in the room, in one session. (If you do cold outreach, I wrote up the full AI outreach system we built at an earlier club.)

9. How do you turn a service into a product? Use it yourself first

Build tools for your own business first. Run on them, find the edge cases on yourself, and eventually they can become your product. Instead of just being a service, you are a product.

Go narrow when you do it. Make a vertical SaaS for your very specific audience instead of asking them to use five different tools. You know that workflow from the inside, and that's what nobody else can copy.

10. Should you build your own tools or pay for software?

AI makes it tempting to rebuild every app you pay for. Only replace an app if:

  • it's taking a big share of your income, or
  • you plan to sell what you build.

Some things are hard to replicate (ElevenLabs voices, for example) and are worth paying for. If your product is an app, here's what an app needs to make money before you build it yourself.

11. What still needs a human?

AI does a lot, but not everything:

  • Posting on Instagram. The platforms have measures against automated posting, so somebody needs to go and post it.
  • Trend watching, and anything client-facing on social media.
  • Video editing. AI won't reach the level of someone who spent two hours on an edit. It can enhance the work, and every model gets better and cheaper.
  • The idea itself. AI is trained on mainstream data. An out-of-the-box idea is still the human part. Ideas matter more now, because building got cheap.

12. Is the free version of Claude or ChatGPT enough? Pay the $20

If you're doing something that should make you money, pay for at least the $20 plan (Claude plans, ChatGPT plans). If you use the free model and decide "it can't do anything," well, yes. You're judging AI by its weakest version.

13. Is it safe to put business data into AI?

My honest take: if you think a trillion-dollar company is going to get hacked and leak that your client likes vegetables, then you shouldn't use any software at all. Your email, your CRM and your booking calendar already live on someone else's servers.

If you want the details, Anthropic's Privacy Center explains how your data is used and what you can turn off.

I still keep a line. Anything that moves money still asks me to verify, and passwords and verification codes stay with me. If you handle regulated data, like medical records, follow the rules for that data whatever tool you use.

14. How I actually work

Here's the honest version. I work on six different things at the same time, poking each one. I work very intensely for about four hours a day, then I meditate, go to the gym and touch grass. It's probably unhealthy.

I didn't take a course for any of this. I just kept asking: can I do this? Let's try. You can't buy being smart, but you can put in the reps.

Your checklist for this week

  • Pay for Claude or ChatGPT (at least the $20 plan).
  • Pick one task you repeat every week and write down its steps.
  • Turn those steps into a skill: voice, steps, tools.
  • Run it, and when the output is wrong, fix the skill, not the output.
  • Put it on a routine once you trust it.
  • Connect one app you use every day (email, calendar, WhatsApp) and talk to it in plain language.
  • Have an idea for a product? Brain-fart it into a folder and ask for the plan first.

Go deeper

Skills

Connectors and MCP

Building and prompting

Plans and privacy

From the Tribed blog

FAQ

What is a Claude skill?

A skill is a long, reusable prompt saved as a file. It describes the voice, what the task is, each step in order, and which tool to use for each step. Instead of explaining a job every time, you write it once, and Claude follows it. You improve it by adding every edge case you find.

What is the difference between a skill and a routine?

A skill describes how to do a job. A routine is that skill running on a schedule, for example every morning or every Thursday, without you starting it. Most people should build the skill first, check its output by hand for a while, and only then put it on a routine.

What is an MCP?

MCP stands for Model Context Protocol. In practice, MCPs are APIs for AI: connectors that let Claude or ChatGPT read from and act inside other apps like email, calendars, WhatsApp or an ads account. Once an app is connected, you manage it in plain language instead of through its interface.

Do I need to know how to code to build with AI?

No. You need a clear idea and the patience to refine. Brain-fart the idea into a folder, ask the AI for a plan before it builds anything, and review that plan. The first 80% often comes in a day. The last 20%, the edge cases, takes longer, and that is where your judgment matters most.

Which AI model should I use?

Use the most capable model for system work: planning, writing skills and designing routines that everything else depends on. Use cheaper, faster models for the doing: running steps the system already defined. That keeps quality high where it matters and stops you from burning through your weekly usage.

The AI Club is a hands-on meetup in Siargao where coaches, founders and freelancers build with AI together. I'm Alfonso, founder of Tribed and Gobo.

Turn your expertise into a branded app business

Your programs, community, payments, content, automations, and an AI coach trained on your method and voice—built into one app for the App Store, Google Play, and web.

Get started