As IT integrates AI into workflows, documentation is more critical than ever

It’s a tale as old as time: A superstar IT pro has been running an entire company’s digital systems for years. She built the asset management, remote monitoring, and ticketing systems herself, and whenever anyone has a question, she’s the one to ask.

But what happens when she puts in her two weeks’ notice or announces her retirement? What if leadership requires that IT offload some of its day-to-day work to AI? How can all the human expertise be passed on to others, or partially automated?

As boomers retire en masse over the next few years and AI continues to integrate into IT workflows, situations like these will become a reality for many IT pros—for some, they already are. The solution is consistent documentation and, importantly, documenting workflows as a continuous, accessible, and centralized process.

Write it down

Modern companies run on technology, and that technology is maintained by robust documentation, Kevin Sequeira, GM of IT operations at IT software company Kaseya, told Morning Brew.

“It’s the operating manual for a company’s tech stack,” Sequeira said. “It’s how you structurally go in and record how everything sits in your company—what the systems are, how are they built, how do you configure them? How do you connect them, how do you secure them, how do you back them up?”

That operating manual should be accessible: Documentation needs to be in one place or on one platform that’s integrated into the larger workflow, not a system that employees need to log into separately, Sequeira said. It also needs to be standardized, meaning it follows templates that the entire company is aware of and can understand, and have role-based permissions. And most importantly, it should be frequently updated.

“[Documentation] has to be continuously maintained. You just can’t go in and create documentation once. Let it be up and running,” Sequeira told Morning Brew. “Because the only thing worse than no documentation is stale [documentation].”

Why aren’t IT pros documenting?

Sequeira said some IT professionals don’t see documentation as “mission critical,” or they might think their company is too small for organized documentation—but that’s not true.

“People need to think of [it] as a co-foundation that you need to have, not something that you take on when you feel you’re along that maturity curve,” Sequeira said.

Additionally, because documentation doesn’t directly generate revenue, investing in preventive measures when everything is going well can feel futile. But Chirag Mehta, a VP and principal analyst at Constellation Research, told Morning Brew that solid documentation can improve IT pros’ mean time to resolution, or MTTR, which is one of the main metrics used to measure success in that role. To ensure that outcome, though, employees need to be incentivized to document.

“You have to make it a clear priority as a senior leader,” Mehta said. “It has to be integrated in the culture itself.”

If company higher-ups aren’t already requiring documentation, the first step is to start documenting workflows and processes immediately, Mehta said.

“If you have to document something that happened 10 years ago, it’s a much harder problem to solve,” he said, “versus, ‘Can I ensure that at least you are documenting things from now on?’”

AI for documentation

Integrating AI auto-documentation into workflows is a great way to stay on top of recording processes and retain institutional knowledge, Mehta said, but to do so, humans need to create foundational documentation to prompt AI.

“AI needs context. AI [can’t rely] on what’s in your head,” Mehta said, adding that failing to document institutional knowledge can prevent companies from successful AI transformation.

Similarly, Jessie Posilkin, executive director of the Technology Modernization Fund, told Morning Brew that while she’s seeing “a real interest in throwing an LLM” at IT documentation, humans still need to work alongside AI. At the TMF, Posilkin helps federal agencies modernize their IT systems—some of which were built in the 1970s—and she’s found that using an information architect and content designer to organize documentation has been helpful.

“Whether you’re organizing procurement information or the socialsecurity.gov website, both of those require people to understand how others think about problems,” Posilkin said. “Then you want to train a large language model…but you need to have that piece done right first.”

And even when an LLM for documentation is in use, Posilkin said she advises federal agencies to always “go back to human expertise.”

“The more we connect lots of disparate systems,” she said, “the more you need to have both humans and documentation explaining how they are connected.”