What we use Claude for: 16+ real jobs at a garment factory

When we started using AI seriously, our first question wasn't how smart it is, or how much it costs.

It was: with the budget we have, what can AI actually help us with, and how do we get the most out of it?

We didn't have the answer on day one. So we picked one job, set a goal for it, let AI try, checked the result, adjusted, and looped. Jobs that worked stayed. Jobs that weren't worth it yet got reworked.

One example: when we first had AI help answer customer chats, we had Claude read through every chat every 4 hours, assess how well the chat AI was doing, then update its work plan and knowledge base. That early phase uses more, because we were tracking so closely.

After looping like this for a while, Claude now handles more than 16 jobs for us. The full detail is much longer. Each job will get its own “experiment log” in the coming weeks, including the parts that broke.

The short version

1. Start from the budget you have, ask which job AI can do best for it, then test, adjust and repeat.
2. The jobs fall into 5 groups: ads, web and content, team and factory reports, systems and projects, and personal assistant work.
3. Most reports arrive on a set schedule, some daily, some Monday/Wednesday/Friday. Nobody has to ask for them.
4. AI helps our team move faster and more accurately. It does not replace people.

1. Ads

• Runs our Meta and Google ads.
• Reports daily on whether the ads are worth it, by reading the messages customers send in, counting how many are genuinely interested, and turning that into a percentage against the ads we ran.
• Picks new content to promote, based on the KPIs we set.

What makes this different from a normal ad report: we don't just look at clicks or message counts. We look at how many of the people who message us are actually interested.

2. Web and content

• Vibe-codes our front-end and back-end web: we tell AI what we want, and AI writes the code.
• Builds our company website in Thai and English.
• Writes our blog and email marketing.
• Market research, branding and content.

Our factory ERP is built by our business partner. Several of the reports below pull their data from that system.

3. Team and factory reports

• Daily: how the sales team and our customer-chat AI are performing.
• Monday, Wednesday, Friday: fabrics running low, so the team knows before we run out, checked against our fabric purchase orders.
• Monday, Thursday: how much production capacity is left, how much work is coming in, how full we are, whether sales can take more work or a rush order, where jobs are stuck, and what needs to happen first.
• Sales by person: against target, how much is left to close, what payments can still be collected, where the pipeline drops off, plus coaching.

What these reports have in common: a clear schedule, and each one answers a question the team has to make a decision on. They don't just send numbers.

4. Systems and projects

• SLAs, SOPs and work instructions for the whole company.
• Internal projects, such as trade show booths and factory improvements.
• Slides for when I speak or teach.

5. Personal assistant

• Booking hotels, trains and museums when I travel abroad.
• Scheduling and other odd jobs.

AI isn't replacing anyone

Every job on this list has a human owner. AI reads, compiles and drafts. The people who decide, take over and stay responsible for the work are still us and our team.

AI helps our team move faster and more accurately. It does not replace people.

If you want to try this, start today

1. List the jobs you or your team repeat every week. Just 10.
2. Circle the ones that are really “repeat questions”: are we running low, are we on target, where is work stuck?
3. Pick one. Give it a clear schedule, like every Monday at 9am, and decide who reads it and what they have to decide.
4. Let AI run that report for 2 weeks, then ask your team whether they could actually decide anything with it. If not, fix the question before adding a new job.

If you had one AI assistant, which job would you give it first? Tell me on LinkedIn.

— Nes · LittleTalk: small stories from a real business that you can actually use.

This article is also available in Thai: เราให้ Claude ทำอะไรบ้าง