You're Probably the One Who'll Bring AI Into Your Fire Protection Company
If you work in the office at a fire protection company, you have probably already noticed that nobody is coming to implement AI for you.
There is no IT department. There is no rollout plan. There is a stack of inspection reports, a customer asking why last quarter's deficiency still shows as open, and an AHJ submittal due Friday.
That vacuum is the actual story — and it is a better one for you than the one you keep hearing.
Small companies haven't moved yet
The U.S. Census Bureau tracks AI use across American businesses in its Business Trends and Outlook Survey. In the survey period running from December 14, 2025 to May 3, 2026, national AI use among firms hovered between 17% and 20%. Firms with 250 or more employees reported 37% current use.
Firms with fewer than 20 employees reported under 20% — and, in the Census Bureau's words, no significant change since December.
That is most of this industry. Independent sprinkler, alarm and extinguisher companies are small businesses, and small businesses have been flat on this for months while the enterprise end of the economy pulled ahead.
Read that as an opening rather than a warning. In a company of twelve people, the person who works out which parts of the job a tool can genuinely help with is not a consultant and not a vendor. It is whoever handles the reports today.
Where the fear actually comes from
It's worth being precise about this, because the fear is not baseless and pretending otherwise doesn't help.
The Federal Reserve published a note on April 3, 2026 monitoring AI adoption in the U.S. economy. Roughly 18% of firms had adopted AI as of year-end 2025. Buried in the same note is the finding that matters most to you: senior leaders "project stronger productivity gains and lower future employment than workers believe will occur." The Fed is explicit that these are expectations, not measured outcomes.
So the gloomiest number in the room is a forecast made by people who are not doing the job.
There is also a real projection worth naming. The Bureau of Labor Statistics expects employment in office and administrative support occupations to fall 3.9%, or about 761,900 jobs, between 2024 and 2034, against total employment growth of 3.1%. BLS attributes part of that to AI-driven efficiency gains.
That is a decline. It is also a decline of roughly four tenths of a percent a year, in an occupational group BLS still projects will see about 2 million openings each year, on average, over the same decade. And BLS says plainly of its own AI assumptions that "developments in AI are proceeding rapidly, and the uncertainty about potential impacts remains very high," noting it applies adjustments from this research conservatively, where there is convincing evidence for a change.
A slow drift with two million annual openings and an explicit uncertainty warning is not the story that gets repeated to you. But it is the story the data supports.
What the evidence says it's good at
Here is the part that should change how you think about your own position.
When researchers have actually measured AI's effect on office work, the gains have landed hardest on the people with the least experience.
Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied customer support agents at a Fortune 500 software company and found a 15% increase in issues resolved per hour overall. The bottom skill quintile gained 36%. The highest-skilled agents saw essentially no significant improvement. The line worth sitting with: agents with two months of tenure and AI access performed about as well as untreated agents with more than six months.
Shakked Noy and Whitney Zhang ran 453 college-educated professionals — marketers, grant writers, consultants, data analysts, HR professionals and managers — through occupation-specific writing tasks: press releases, short reports, analysis plans, professional emails. Time taken fell by 0.8 standard deviations and output quality rose by 0.4 standard deviations, with the larger gains concentrated among lower-performing workers.
Look at that task list again. Short reports. Professional emails. That is a description of Tuesday afternoon in a fire protection office.
Concretely, the work in your day that matches what the research measured:
Turning an inspector's shorthand deficiency notes into a write-up a building owner will actually understand
Summarizing a forty-page ITM report down to the three items the customer has to act on
Drafting the fifth follow-up email this week about unscheduled quarterly inspections
Rewriting a proposal cover letter you have written thirty times before
Reformatting notes into whatever structure a particular AHJ prefers
None of that is the skilled part of your job. All of it is the part that keeps you at your desk until six.
What to keep in your own hands
The same body of research is equally clear about where these tools fail, and the failure mode is specific: they fail on judgment.
Fabrizio Dell'Acqua and colleagues at Harvard Business School ran a preregistered experiment with 758 Boston Consulting Group knowledge workers. On eighteen realistic tasks inside the model's competence, workers with AI completed 12.2% more tasks, 25.1% more quickly, with better quality. On a complex task that sat outside it, workers using AI were 19% less likely to produce correct solutions than workers without it.
METR found something related from the other direction. In a randomized trial, experienced open-source developers working on codebases they knew well took 19% longer with AI tools than without. They had expected a 24% speedup, and still believed afterward that they'd gotten a 20% one. METR is careful to say this does not show that AI fails to speed up most developers, and that the result may not extend to less experienced people or unfamiliar work.
Translate both findings into your building. The judgment calls stay yours:
Whether a finding is a deficiency, a non-compliant condition, or an impairment under NFPA 25 — and what that triggers
Whether an inspection result is wrong or the system genuinely changed
Which AHJ accepts what, in what format, on what timeline
That the riser room at one property is labeled backwards and always has been
That knowledge lives in your head and in nobody's training data. It is the reason the experienced people in these studies gained least — there was less that a general-purpose tool could add. That is not a threat to your position. It is a description of it.
Where to start
Pick one recurring writing task you resent. Not a compliance decision — a writing task. Draft it, then read every line before it leaves the building, exactly as you would check a new hire's first report.
Then tell your owner what you found, in your own numbers: this took forty minutes, now it takes twelve, and here is the one place it got the NFPA language wrong so I fixed it.
You will have done something the Fed's data says most people at the top are only forecasting. And you will have done it as the person who knows which parts of this job can safely be handed off — which is knowledge nobody above you has.
The office staff at small fire protection companies are going to decide how AI actually gets used in this industry. Not because anyone appointed them. Because they are the only ones close enough to the work to tell the difference between the parts that need judgment and the parts that just need typing.
Joyfill is built for that gap — mobile inspection forms and reporting that cut the rekeying out of the office so the work that reaches your desk is the work that actually needs you. See how Joyfill fits your inspection workflow.
