Process Documentation Nobody Reads vs Skills an Agent Runs Every Time

Process documentation goes unread because nothing depends on it. A skill is the same knowledge written as instructions an AI follows every time the job is done, so errors show up in the work and get fixed. That daily use is what keeps a skill accurate. Turning your best people's methods into skills keeps them alive.

Why does process documentation go unread?

Because reading it is optional and the work gets done without it. Most process documents are written once, often for an audit or onboarding, then filed. People ask a colleague instead, because the colleague is faster and more current. Within months the document no longer matches how the job is actually done.

Several habits make it worse. Documents are written in policy language rather than the words of the person who does the job well. They describe an ideal rather than the real method. And nobody owns keeping them current, so updates depend on someone remembering.

The cost is real. According to Panopto's 2018 Workplace Knowledge and Productivity Report, 42% of institutional knowledge is unique to the individual who holds it. The same report found knowledge workers lose 5.3 hours a week waiting for information or recreating knowledge that already exists. Unread documentation is part of why.

What is the difference between a process document and a skill?

A process document describes how a job should be done, for a person to read. A skill instructs an AI how to do the job, step by step, and is run every time the job comes up. The content can be similar; the difference is that a skill is used, so it is tested continuously.

In Anthropic's terms, skills are "folders that include instructions, scripts, and resources that Claude can load when needed". The idea extends beyond one vendor: the format was published as an open standard in December 2025.

Why do skills stay accurate when documents do not?

Because the work depends on them. If a skill tells an AI to use last year's report template, this week's report comes out wrong, and someone notices and fixes the skill. A process document with the same error can sit unnoticed for years, because nobody follows it closely enough to find out.

That feedback loop is the whole point. A document fails quietly. A skill fails loudly, in the output, which is exactly what keeps it right. Over time, the skill becomes the most accurate description of how the job is done in the business.

There is a second benefit: consistency. When everyone's AI runs the same skill, the job is done the best way every time, not just when the expert does it.

How do you turn a process document into a skill?

Start from the document, then go to the person who does the job best and walk through a real example. Rewrite the method as clear steps, add the checks they make and the mistakes they avoid, include one example of good output, and write a short description of when the skill should be used.

A practical conversion:

1. Pick one document for a job done at least weekly.
2. Sit with the expert and watch them do the job on a real case.
3. Capture the real steps, including the ones the document left out.
4. Add the checks they make before handing work over.
5. Include an example of output they would approve.
6. Write the description: what the skill does and when to use it, in a sentence.
7. Test it on real cases and refine.

According to Anthropic's help documentation, a skill's description can be up to 200 characters and is what decides when Claude uses it, so that one sentence matters. The guide to writing a Claude skill covers the file format.

Which processes should become skills first?

Start with processes that are frequent, depend on expertise and produce output a person can check quickly: weekly reports, proposals, client onboarding steps, content checks and follow-up emails. Avoid rare, high-judgement processes at first; they are harder to capture and slower to test.

A useful signal is repetition of questions. If people keep asking the same expert how something is done, that method is a strong candidate. Management frameworks such as EOS, which ask businesses to document their core processes, give a ready list to start from. The comparison of Company OS and EOS explains how the two fit together.

Skein turns businesses' core processes into skills that run as agents, each reading a shared company brain; the agents page shows them. That suits a business whose best methods live in a few people's heads. A business with stable, simple processes and little AI use may be fine with good documents for now.

Frequently asked questions

Why doesn't anyone read our process documentation?

Because the work gets done without it. People ask colleagues instead, documents are often written in policy language rather than practical steps, and nobody owns keeping them current. Within months, they no longer match how the job is really done, so people trust them even less.

What is the difference between a process document and an AI skill?

A process document describes a job for a person to read when they choose to. A skill writes the same method as instructions an AI follows every time the job is done. Because the skill is used constantly, errors show up in the output and get fixed.

Can AI help with process documentation?

Yes. AI can turn call recordings, walkthroughs and existing documents into clear step-by-step methods, and then run them as skills. The most valuable input is still a real example from the person who does the job best, so the method reflects how work is actually done.

How do you keep process documentation up to date?

Make something depend on it. When a method is written as a skill that AI runs on every job, wrong steps produce wrong output that people notice and correct. Pair that with a named owner for each skill and a periodic review.

Which processes should be turned into AI skills first?

Start with frequent jobs that depend on expertise and produce output a person can check quickly, such as weekly reports, proposals, onboarding steps, content checks and follow-up emails. Processes that people keep asking one expert about are especially good candidates.

Related reading