Gotchaa Lab
Back to Blog
AIAI agentsmalaysiaSMEbusiness-strategy

Can AI Run a Business? Hand It One Department, Not the Org Chart

22 July 2026·5 min read·By Gotchaa Lab
Can AI Run a Business? Hand It One Department, Not the Org Chart

TL;DR

  • A viral X post arranged 42 Claude skills into seven departments of a company org chart. It is a real shopping list of installable instruction sets, not a working company.
  • AI can hold one department's repetitive work today. It cannot hold accountability, which is why every pilot still needs a named human owner.
  • Pick your first department with two questions: how many times a week does this task repeat, and what happens when it is wrong? High repetition plus low blast radius wins.
  • 81% of Malaysian MSMEs told Xero they have already implemented AI, but 82% say they need more education to use it confidently, and 30% of users have no governance policy at all.

In mid July a post on X broke Claude down into 42 skills and arranged them like a company org chart. Developers, marketing, finance, legal, every department with its own stack of real, installable tools. A widely reposted version passed 790,000 views. The implied promise was hard to miss: here is your whole company, no staff required. Which lands on the question plenty of owners are now asking out loud. Can AI run a business?

No, not end to end, and not this year. It can hold one department's repetitive work, reliably, under a named human owner. That is the honest version of the org chart, and it is still a big deal.

What the 42-skill org chart actually is

A skill is a folder of written instructions the AI loads only when a task calls for it. Anthropic describes them as modular, file system based capabilities and publishes a set for everyday document work. Community collections go much further: one public repository lists 166 skills sorted into 14 corporate departments, from executive leadership to procurement.

So the chart is real. It is also not a company. Every box on it sits idle until someone triggers it, checks the output, and carries the consequences. An org chart made of skills has nobody in it.

Can AI run a business department by department?

That is the more useful question. Here is the split we see in client projects, not the one the thread promised.

DepartmentWhat AI can hold todayWhat still needs a human owner
Marketing and contentFirst drafts, repurposing, research briefsAny claim that must be true, and brand judgement
Sales adminLead research, CRM notes, follow-up draftsThe price, the promise, the relationship
Finance and adminInvoice data entry, reconciliation prep, expense sortingAnything filed to LHDN, and every number you sign
Customer supportThe same five questions, ticket routing, draft repliesComplaints, refunds, anything a screenshot could make viral
Operations and reportingAssembling the weekly report, tidying dataThe decision the report leads to

None of that right column is beyond what the models can produce. It just needs somebody accountable, and a smarter model does not fix that.

Which department should a Malaysian SME hand over first?

Score every candidate on two questions. How many times a week does this task repeat? And what happens when the output is wrong?

You want high repetition and low blast radius. A weekly sales report assembled from three spreadsheets repeats 52 times a year and fails quietly, costing you an hour and an apology. An e-invoice submitted to MyInvois with the wrong figure fails loudly and costs money. Give the AI the first kind of job and make it earn the second. Switch on all seven departments at once and you end up with 42 installed skills and none in daily use.

Slide from Xero's AI readiness and adoption survey of Malaysia's MSMEs Xero's AI readiness study, published 18 November 2025, covered 1,033 Malaysian MSMEs. Source: Malaysia SME

The Malaysian numbers back this up. Xero's November 2025 study of 1,033 local MSMEs found 81% had already implemented AI somewhere, 82% said they need more education to use it confidently, and 30% of users had no governance policy at all. Owners here have not refused to try AI. Most have tried it thinly, without deciding who owns the output. And governance is not just paperwork: putting customer records into a personal chatbot account is a PDPA problem, not a preference.

Our take: can AI run a business without you in 2026?

The org chart works as a sketch and fails as a plan. We use these tools daily and they have genuinely changed how fast we ship the AI systems we build for clients, but the wins come from one narrow job done properly, then the next. Never from a wall of departments switched on in a weekend.

For concrete first pilots with costs attached, see our five ways an SME can put an AI agent to work. If somebody is pitching you the whole org chart as a service, read the AI automation agency hype first.

If you do one thing this month, draw your own org chart on a single page and circle the most repetitive task nobody would panic about if it came out wrong once. Give that task to a tool for four weeks with one person's name against it, then count the hours saved before you buy anything else.

Thinking about which part of your business is actually ready for this? Let's chat. We will give you an honest read, including the parts we would tell you not to automate yet.

This article is general information and does not constitute legal advice. Check your own PDPA and e-invoicing obligations with a qualified adviser.

References

  1. The 42-skill Claude org chart post on X
  2. Charlie Hills' original org chart writeup
  3. Anthropic: Agent Skills overview
  4. awesome-claude-corporate-skills: 166 skills across 14 departments
  5. Xero survey: Malaysian MSMEs are rapidly adopting AI, but a confidence gap remains

Share this article

Frequently Asked Questions

Can AI run a business on its own?
Not end to end, and not in 2026. AI agents can already carry a large share of research, drafting, data entry, and routine replies. What they cannot carry is accountability: signing a number, making a promise to a customer, filing to a regulator, or owning the outcome when something goes wrong. The realistic version is one department's repetitive work handled by AI under a named human owner, not a company with nobody in it.
Which department should a Malaysian SME automate with AI first?
Score each candidate on two things: how often the task repeats, and how much damage a wrong output does. The best first pilot is high repetition and low blast radius. For most SMEs that means internal reporting, document drafting, or first-line support replies that a human still sends. Finance filing and final customer replies come later, because a wrong output there costs money or trust immediately.
What is a Claude skill, and do I need a developer to use one?
A skill is a folder of written instructions plus any supporting files that the AI loads only when a task calls for it. Anthropic documents them as modular, file system based capabilities. Installing one is closer to dropping a document into a folder than to writing software. Wiring a skill into your actual systems, your inbox, your order data, or your accounting tool, is where a developer usually comes in.
Is it safe to give AI access to customer data in Malaysia?
It depends on where the data goes and what you have written down. Under Malaysia's PDPA you remain responsible for personal data you hand to a processor, so consumer chatbot accounts are a poor fit for customer records. Use a business tier with a data processing agreement, redact what the task does not need, and keep a log of what the AI did. In the Xero survey, 59% of Malaysian MSMEs named data privacy and security as their main AI concern, which is a reasonable worry rather than a paranoid one.

Need help building this for your business?

We help Malaysian companies turn ideas like these into working software. Free consultation, no obligation.