The best, worst and strangest ways AI is really being used at work
By Gammatek ISPL, Industrial Systems & Compliance Analyst at Gammatek ISPL
Last updated: September 2026 | 11 min read
Author block: Gammtek ISPL covers workplace technology adoption and its operational implications for manufacturing and industrial clients at Gammatek ISPL, drawing on direct conversations with plant managers and IT leads evaluating AI tools in 2026.
Why This Matters to You
If your company rolled out an official AI tool this year, there's a good chance most of your employees aren't using it. Instead, they're pasting work into their personal ChatGPT or Gemini account on their phone, because it's faster and nobody's watching<cite index="2-1">, with the real daily use skewing toward mundane tasks like first drafts, meeting notes, and summarizing long documents rather than anything transformative</cite>. That gap between what's approved and what's actually happening — often called "shadow AI" — matters because it's invisible to IT, invisible to compliance teams, and quietly becoming the real story of how AI is used at work, even as companies spend heavily on official rollouts. If you manage a team, a plant, or a compliance program, you're likely underestimating how much of this is already happening under your roof.
The Best: Where AI Is Quietly Making Work Better
The genuinely useful uses of AI at work in 2026 are almost boringly practical. Sales teams draft follow-up emails and ask AI to tighten the tone. Marketing staff condense long competitor reports into a handful of bullet points before a meeting. Customer service reps paste an angry customer's email into an AI tool to get help drafting a calmer response<cite index="2-1">, and HR staff have used AI to help draft difficult scripts, like preparing for a redundancy conversation</cite>.
On the plant floor, the pattern is similar but higher-stakes. In our own client conversations at Gammatek, the AI use cases that plant managers actually value aren't flashy — they're things like AI-assisted anomaly flagging on sensor data (catching a pattern a human might miss across hundreds of readings) or AI-drafted incident report summaries that get reviewed and finalized by a real safety officer before submission. The pattern across every genuinely useful case: AI drafts, a human decides. The moment that flips — AI deciding, human rubber-stamping — is where things start going wrong.
Original data point: Enterprise AI adoption has moved fast — <cite index="4-1">by 2026, a majority of enterprises have deployed AI tools across multiple departments, with employees typically using several different AI applications in an average day</cite>. But adoption speed and adoption quality are two different things, which is exactly what the "worst" section below gets into.
The Worst: Where AI Use Is Creating Real Problems
The worst uses of AI at work aren't dramatic — they're the quiet, ungoverned ones. <cite index="4-1">The biggest risks organizations report are shadow AI adoption happening without IT visibility, data being shared into these tools without oversight, and AI-related costs that catch finance departments by surprise</cite>.
Trust is also fraying. <cite index="1-1">A majority of workers now expect employers to use AI mainly to cut jobs and reduce costs rather than to benefit employees directly, and most workers say AI should have no role in deciding who gets hired or promoted</cite>. That distrust shapes how people actually behave — including the shadow AI pattern above, since employees using AI without approval are often doing so precisely because they don't trust official channels or don't want their AI use tracked.
For regulated industries specifically, this is where it gets serious. A manufacturing or pharma plant that discovers, after the fact, that an employee pasted proprietary process data or a compliance document into a personal AI account has a real data governance problem — not a hypothetical one. This is the exact blind spot that structured AI usage policies and audit trails are built to close, and it's a growing reason compliance software now needs to account for AI tool sprawl, not just traditional document control.
The Strangest: Where AI Use Gets Genuinely Odd
Not every strange AI use case is dangerous — some are just bizarre. <cite index="3-1">One manager reportedly fed employees' LinkedIn headshots into an AI image generator to create themed "personas" for each person, like turning an IT specialist into a cartoon "Tech Wizard."</cite> In another case, <cite index="3-1">a manager known for blunt, aggressive communication started using AI to soften and rewrite their emails and messages — which employees found unsettling rather than reassuring, since the tone didn't match the person they knew</cite>. And on the hiring side, <cite index="3-1">some companies have started reconsidering long-standing practices, like sending interview questions to candidates in advance, because candidates are now running those questions through AI before the interview even starts</cite>.
These stories share a common thread: AI isn't just changing what gets done at work, it's changing trust — between managers and employees, between candidates and interviewers, between what a message says and who (or what) actually wrote it.
Comparison: How These Play Out Differently in Office vs. Industrial Settings
Office/Corporate Setting | Industrial/Plant Setting | |
Most common "good" use | Email drafts, meeting summaries | Anomaly detection on sensor/equipment data |
Biggest risk | Shadow AI leaking business data | Shadow AI leaking proprietary process or safety data |
Governance gap | IT can't see personal AI account use | Compliance teams often have no AI usage policy at all yet |
Consequence if unmanaged | Data privacy incident, brand risk | Regulatory non-compliance, safety audit failure |
What This Means If You Run a Plant or Manage Compliance
The honest takeaway isn't "ban AI" or "adopt AI everywhere" — it's that most organizations don't actually know how much AI is already being used inside their walls, in what form, or with what data. For a manufacturing, chemical, or pharma operation under regulatory scrutiny, that blind spot is a bigger liability than the AI use itself.
A practical starting point: treat AI tool usage the same way you'd treat any other data-handling policy — define what's approved, where data can and can't go, and who reviews AI-assisted output before it becomes an official record (an incident report, an audit document, a customer communication). That's not a technology project, it's a governance one — and it's exactly the layer that sits on top of whatever tools your team ends up using.
[See how Gammatek's compliance platform helps plants build audit-ready policies around new technology adoption, including AI →https://www.gammateksolutions.com/post/your-boss-tech-companies-and-police-can-read-your-chatbot-conversations https://www.gammateksolutions.com/post/new-google-ai-model-said-to-narrow-gap-on-coding-ability
Comments