An AI agent runs on what your company knows. It looks up the procedure to follow the steps, the price list to build a quote, the contract to answer a client. If that material is scattered, out of date or owned by nobody, the agent doesn’t stop. It gets things wrong with complete confidence.
IT Glue, Kaseya’s IT documentation platform, published a short checklist to help IT teams get ready before switching on its AI features. It was written for IT, but the logic holds for any team. Here’s the roadmap, with our take on applying it to the rest of the business.
What the checklist proposes
It has three blocks, in this order: strengthen the foundation, manage your process documentation and keep an accurate picture of your environment. Each block has yes-or-no questions and open questions that turn into a plan. It closes with three short ones: one improvement for this month, one AI feature to try and the blockers that could get in the way.
The order matters. An AI tool won’t tidy up messy documentation, it speeds up whatever is already there. Adding AI before the foundation just spreads mistakes faster.
Foundation first
The first block asks whether the essentials are written down and in one place. Outside IT, that looks like this:
- Is what the team looks up to do its job documented? In IT that’s passwords, licenses and configurations. At a firm it’s contract templates, price lists, client records and internal policies.
- Are the critical procedures written down and available to everyone? If the steps live in one person’s head, the agent has nothing to follow.
- Is everything in one place, with related items linked? Three copies of the same policy in three folders is an invitation for the agent to use the wrong one.
- Do folders follow a structure, global or per client? One folder per client, matter or team, the same everywhere, also keeps one client’s data out of another client’s answer.
- Is there a naming convention? Dates in one format, the version in the file name, agreed abbreviations. That’s what lets search find the right version instead of a PDF from 2019.
- Are there tags and categories? And above all, is someone responsible for keeping all of this in shape?
Then, a process for procedures
The second block is about how procedures are born and retired. With AI writing drafts, this matters even more.
- A written process for creating and approving procedures, including the ones AI generated.
- A way to find gaps: the things the team does every week that nobody has documented.
- A clear call on whether to audit what exists or start fresh.
- Someone responsible for new procedures, and audits on a schedule: monthly, quarterly or twice a year.
The most important question in the block is the last one: how will we review and approve what AI wrote? Without an answer, a machine’s draft becomes company policy before anyone has read it.
Finally, a picture that stays current
Documentation only helps if it matches what exists today. The third block covers cleanup and the links between items.
- Cleanup on a set schedule: archive what has expired, merge duplicates, flag what nobody opens. Weekly, monthly or quarterly, but on the calendar.
- Linked items: the contract tied to the client, the process and the owner. IT Glue suggests these links automatically; either way, it gives the agent the full context instead of a stray file.
- Automatic summaries of the most-used items, for anyone who needs the quick view.
- Someone approving every suggestion for cleanup and links. AI suggests, a person decides.
What ties it together is an owner
All three blocks ask the same question in different words: who is responsible? Who maintains the structure, who audits the tags, who approves the AI-written procedure, who signs off on cleanup. The checklist doesn’t treat AI as a replacement for whoever looks after documentation. It treats it as a fast helper that needs supervision.
It’s the same design we use for agents inside an operation: the agent does the repeatable work, and someone from the team reviews it where it matters.
Beyond IT, it’s the same list
Finance, legal, sales, support and procurement are just examples. If a task repeats and depends on documents, it can be automated, and the documentation needs to be at this level first. In practice, this is where we start:
- The AI Analyst runs the assessment with these questions: what is written down, where it lives, who owns it and what nobody has documented.
- The Forward Deployed Engineer organizes the knowledge base the agent will use, with names, tags and links, and puts human review at the right checkpoints.
- The team learns to keep it up: who approves what AI wrote, how often to clean up and how to spot the gaps.
A good agent on bad documentation just gives bad answers faster. Fix the foundation, name an owner, then switch on the AI.
Enjoyed this article?
Get the next one before it hits the blog.
A short email per issue with what actually matters about AI agents at work. No spam.