Skip to main content
Fygment
Get started

ResourcesOpinion

Your company's best know-how leaves when your people do

Most of what makes a company good at its work lives in a few people's heads. AI skills are the first practical way to capture that knowledge so it stays, gets used every day and keeps improving.

The Fygment team

, 5 min read

Every company has a few people everyone goes to. The project manager who knows how to scope a fixed-price deal without losing money. The finance lead who knows which supplier invoices always need a second look. The account manager who knows exactly what each large client cares about. None of it is written down in a way anyone else could use.

When one of those people leaves, retires or simply goes on holiday, part of how your company works goes with them. Most leaders know this. Very few have a plan for it.

The knowledge that runs your company is in people's heads

The surveys point the same way. In a 2025 study by APQC, a knowledge-management research body, 92% of organisations said they do not consistently capture knowledge from people about to retire, and 58% of executives said they were very worried about losing it. A 2018 survey by Panopto and YouGov found that employees consider 42% of what they know for their job to be known only to them.

Even while people stay, that knowledge is expensive to reach. In Atlassian's 2025 survey of 12,000 knowledge workers, leaders and teams said they lose about a quarter of their time just searching for answers.

Where documentation stops and skills take over

Documentation still matters. Policies, product information, the reasons behind decisions, how a system is set up: people need to read and understand these, and that will not change. Keeping documentation findable and current is a problem of its own, and it needs its own solution.

There is one kind of knowledge documentation has always handled badly: how to actually do the work. The steps, the exceptions, the checks an expert applies without thinking. Written down, that knowledge has to be found and applied by someone who is busy doing the job, and at that moment asking a colleague is faster. That is the gap AI skills fill. They do not replace the documents: a good skill points to them as its sources, so better documentation makes better skills.

What changes with AI skills

An AI skill is a set of instructions that tells an AI tool such as Claude, ChatGPT or Copilot how your company does a specific task. Anthropic, which created the format, describes skills as the way to give AI the procedural knowledge and company context that real work requires, and compares writing one to putting together an onboarding guide for a new hire. Its guidance for enterprises describes companies capturing their own workflows, compliance rules and institutional knowledge in skills.

The difference lies in what happens at the moment of work. A document waits to be read. A skill gets used every time the work is done. When someone asks the AI to scope a fixed-price project, the skill applies the senior project manager's method, including the exceptions and the checks, without anyone having to look it up.

That also changes what happens to the knowledge over time. A skill that is used every day shows its gaps quickly, so it gets corrected instead of quietly going stale. Analysts have noticed: in a July 2026 note, Gartner argues that companies need a library of agent skills, partly to preserve business knowledge.

“A document waits to be read. A skill gets used every time the work is done.”

Knowledge that grows in value

Treated well, an AI skill turns your organisation's knowledge into an asset that grows in value. Every use is a chance to improve it. Every improvement reaches everyone at once. A new hire starts on day one with the best way your company knows to do the job, instead of whatever the person next to them happens to do. And when an expert leaves, their way of working stays.

Treated badly, skills become one more thing that rots. A skill copied around on laptops, with no owner and no tests, has the same problems as an outdated wiki page, only now the AI acts on it. We describe what goes wrong in AI skills aren't one-and-done. The difference between an asset and a liability is whether the company manages its skills like it manages any other important asset.

What leaders should do now

Start by finding the knowledge you cannot afford to lose. Ask who in each team gets asked the most questions, which tasks only one person can do, and whose departure would hurt the most. That list is where your first skills come from.

Then start small. Pick one process, check that its data is in order (our data checklist takes an afternoon), and have the expert build the first skill with someone who knows how to structure and test it. Give the skill an owner and a place in a company library, not on someone's laptop. The full method is in How to turn your organisation's knowledge into AI skills, systematically.

That library is what Fygment provides. Every skill has an owner, a version history and an approval step. Colleagues send feedback and improvements to the owner, the approved version reaches every AI tool your team uses, and each skill carries test cases that a new version has to pass. Your company's knowledge stays in the company, current and in use.

If you want to start with one skill, Fygment is free to start.

Questions we often get

Can AI skills capture knowledge that experts find hard to explain?
Much of it, yes, if the skill is built with the expert rather than from documents. Watching the expert do the task with an AI tool, and noticing what context and corrections they keep adding, surfaces rules they would never think to write down.
How is an AI skill different from a wiki page?
A wiki page has to be found and read by a person. A skill is applied by the AI every time someone does the task, so the knowledge is used without anyone looking it up, and its gaps show up quickly.
Who should write the skills in a company?
The people who know the work best, usually subject matter experts rather than engineers, supported by someone who knows how to structure and test a skill.
Do AI skills replace knowledge management?
No. Skills cover how work is done: processes, rules, checks and standards, applied at the moment of work. Documentation and knowledge management cover everything else people need to read and understand. Each makes the other better: skills point to documents as their sources, and using skills every day shows where the documents are missing or out of date.

Keep reading