How the AI Enabled Safety Professional is Empowering Organizations

The Best Part of Speaking

I recently spoke at the BC Construction Safety Alliance's annual safety conference. My talk, AI in Construction Safety: The Good, the Badass, and the Sloppy, covered the current landscape of AI capabilities and the implications for professionals who are curious enough to jump into the AI pool with both feet.

The part I love most about speaking at conferences is the time afterwards. This is when people approach me, wanting to talk about their own growing fascination with AI and how they are using it to transform their work. These people have caught the bug. They have discovered that AI isn't just a chatbot. It is a capability multiplier that allows them to do things they would never have thought possible because the costs were too high or the skills required were too advanced.

Now these people are vibecoding apps, designing workflows and integrating AI into their daily work processes. They are creators, dreamers and strategists, and they are insatiably curious. They are driven by the thrill of turning their ideas into creations, things that actually work and produce value. They have extended beyond safety within their organizations, and they are often bringing value by improving business processes and even driving sales.

A Drink With Two Attendees

After this particular talk, I was invited for a drink with two of the conference attendees. I have given both of them fictional names here. One of them, Marcus, was a new safety professional at a rock scaling company and was deeply immersed in AI technology. He wore a Bluetooth headset around his neck, which he said he used to have conversations with AI agents when he was brainstorming ideas during his drives to work. He also had a Plaud pendant hanging from a necklace, which he used during our conversation to transcribe portions of what we said or to quickly jot down thoughts for later.

He often looked down at his phone to give orders to his Dot, OpenAI's always-on agent inside ChatGPT, which had access to the two open laptops he told us were running in his home office. His Dot could read his emails, send emails, schedule events and meetings, manage his LinkedIn, access his folders and much more.

Marcus was the safety manager for the company. He oversaw several advisors during the busy months and was now helping build and expand the company's rope rescue division. He had gone all in on AI. He had gone into full AI immersion, and his company was supporting and empowering him. They purchased a Pro ChatGPT license and his Plaud transcriber, and they were unleashing his energy and AI competency onto serious business challenges.

I had no doubt he was bringing value, and I wondered to myself how many other safety professionals like him were out there. These are people who have discovered the power of AI and have become enthralled exploring its strange capabilities and genius.

Where Do I Go to Learn More?

Our other guest, Tom, was not as AI-pilled as me and my new comrade, but he was curious about the magical world we were describing and he was taking notes in a notebook. At one point he asked a question I often get: "Where do I go to learn more about this?" He seemed to be implying that there was a course that would teach him everything he needed to know.

The answer, like the answer to many other questions these days, is that you go to the model. You ask the model to interview you so you get a sense of what you want to do. Then you start building and exploring. You start coding even though you're not a programmer. You build a website or an app. You connect things to it, like other software. You connect your large language model to other software. You watch YouTube videos, or ask AI to watch the video or read its transcript. You give your chosen AI access to folders, files and a web browser so it can navigate complicated websites on your behalf. In other words, you explore and take things in whatever direction you feel driven toward.

Most of all, you have to be ambitious. Ethan Mollick has said he can guarantee that we are not being ambitious enough when it comes to using AI. Push this technology as far as you can and watch in amazement as it delivers results. Don't worry about perfect prompting. Just explore, and give the AI the information, tools and access it needs to get the job done.

Your Own Executive Assistant

Many people are just discovering how powerful this technology is when it is connected to the appropriate tools and given the right context, but many more have not begun to figure this out. Marcus was seeing the enormous advantage of this technology. He told us how he speaks into his Plaud throughout the day and how his AI agent pulls the transcription through MCP. From there, the agent infers whether there is work to be done, meetings to schedule or actions to take. When he comes across a workflow that really works, he tells his AI to build a skill so it is repeatable.

He is connected to his own personal executive assistant, and his assistant happens to have genius level intelligence. Imagine for a minute how incredible that is. A single person can now have their own second brain, or any number of brains, and use that brain to take actions on their behalf. Dot is one of several always-on agents. Meta has Muse, Anthropic has Duet and xAI has Grok Bot.

At one point in our conversation, Marcus said, incredulously, "I don't even sit down at a computer anymore."

This is why I suggest connecting your agent to your safety program. Strawboss is the AI-native safety platform I am building. Think of it as a safety manager in a box. It has safety skills built in and access to your company's data, including all the forms your people fill in. Connect your agent to Strawboss through its MCP, and your agent can read that data and take actions in Strawboss on your behalf. Check it out at strawboss.ai.

Take It Back to Your Company

Companies would be wise to find these champions in their ranks. Employees, regardless of their rank or years of service, should take advantage of this moment where the technology has not fully diffused throughout our workplaces. Where can you show value beyond your current role or status in the company?

Build something, demonstrate that it works, then get buy-in to implement a pilot. These are exciting times, and fortune favours the curious.

Want to build the setup Marcus described? The step by step instructions, with tools and links, are in the appendix below.

Appendix: Build the Setup Yourself

Here is a version of what Marcus described that you can build in an afternoon. It has five pieces: an agent, a connector, a recorder, tools and skills. They fall into three layers: capture, connect and act. Start small, keep a human in the loop, and widen access as you build trust.

What you need

• A wearable recorder. Plaud NotePin, or the newer Plaud NotePin S. Any recorder with an app that produces transcripts will work, but Plaud has an official MCP connection.
• An AI assistant. Claude or ChatGPT. Marcus used a Pro ChatGPT license, the tier that includes a Dot. Always-on agents include ChatGPT's Dot, Meta's Muse, Anthropic's Duet and xAI's Grok Bot, described below. Pick the one you will actually open every day.
• The Plaud connection (MCP). MCP is the open standard that lets an AI tool connect to other software. Ask your AI to find Plaud's setup guide and install it for you, or follow the guide yourself at docs.plaud.ai.
• A folder and a browser. Claude Desktop can work with files on your computer, and so can ChatGPT's Work mode in the desktop app. Claude in Chrome can navigate websites for you.
• A voice only option. You can skip the hardware, download Wispr Flow and speak into your phone or earbuds to dictate to your AI assistant.

The five pieces

• Agent. The AI that does the work. The newer agents run on their own cloud computer, so they can keep working when you are not watching.
• Connector. The link that gives the agent access to an app or data source, such as Plaud, your calendar or your email. Most are built on MCP.
• Recorder. The Plaud device captures what you say and turns it into transcripts and notes.
• Tools. The apps and browser the agent uses to act, such as calendar, email, files and websites.
• Skills. Saved instructions for repeatable jobs, so the agent does them your way every time.

Choose your agent

Several AI companies now sell always-on agents. These four are worth knowing. Some are in preview or beta, so check availability in your region.

• Dot (OpenAI). An always-on agent inside ChatGPT with its own cloud computer and browser, launched September 29, 2026 for ChatGPT Pro and Business Premium. Getting started with your dot.
• Claude Duet (Anthropic). A proactive agent that sends a morning brief, flags what needs your attention and drafts replies for you to approve. It carries over your Claude memory, skills and connected apps. Research preview with a waitlist. Claude Duet.
• Muse (Meta). A personal agent that runs in a dedicated virtual machine in Meta's cloud, available on iOS, Android and the web. Muse.
• Grok Bot (xAI). Named, persistent bots that share a cloud computer, with plugins, MCP, skills and approvals. Early beta since August 2026. Grok Bot overview.

Before you commit, confirm your agent can connect to Plaud. Plaud is a connector in Claude and an app in ChatGPT, and it works with other MCP compatible clients.

Step 1: Capture

1. Set up your Plaud device and the Plaud app, then pair them.
2. Make a short test recording and confirm the transcript and summary appear in the app.
3. Decide the rules before you wear it on site. Tell people when you are recording, and check your company policy and the privacy and consent rules that apply to you.

Step 2: Choose your agent

1. Pick the agent you will open every day. Use Dot if you are on ChatGPT Pro, Duet if you are on Claude Pro or Max and get off the waitlist, or a regular Claude or ChatGPT chat to start.
2. Sign up for the paid plan. If you chose Claude, install Claude Desktop, the version that can reach your files.
3. Run the interview prompt from earlier so the agent learns your role before you connect anything.

Step 3: Connect Plaud to the assistant

In Claude (web or desktop):

1. Open Settings, then Connectors.
2. Search for Plaud Web MCP and click Connect.
3. Authorize with your Plaud account.

In ChatGPT:

1. Go to chatgpt.com and open Apps (labelled Plugins in some versions) in the left sidebar.
2. Search for Plaud and click Connect.
3. Sign in with your Plaud account.

Prefer the command line, or using Claude Code? Run npx -y @plaud-ai/mcp@latest install, sign in when the browser opens, then restart the app. This needs Node.js installed.

Test it by asking: "List my Plaud recordings from today and summarize the action items."

A note on data: with ChatGPT, your recording data passes through Plaud's US-hosted MCP server. Plaud's documentation states it does not store the data after the request completes. Read the current documentation and decide whether that fits your site's confidentiality requirements.

Step 4: Add the tools your assistant will act through

1. In Settings, then Connectors, add the apps you use every day, such as your calendar and email. The Claude connector guide explains how to add others.
2. Start with read access and drafts only. Let the assistant draft emails and propose meetings, and you press send.
3. Give it one folder, not your whole drive. Create a folder such as Assistant Work and grant access to that folder only.
4. Install Claude in Chrome if you want the assistant to work inside websites, such as portals and forms.
5. Connect your agent to Strawboss through its MCP, so it can read your company's safety data and take actions in Strawboss for you. If you are not sure how, ask your AI to find the Strawboss MCP details and connect it.

Step 5: Give it skills and standing instructions

A skill is a saved set of instructions for a repeatable job, such as turning a toolbox talk recording into a finished record. Write down one task you do every week, the steps you take and what a good result looks like, then save it as a skill or standing instruction in your agent. Duet carries over the skills from your Claude account, and Grok Bot can draft a skill from a task you show it once.

Paste this into a new Claude Project or a custom ChatGPT, and edit it for your work:

Step 6: Run it for one week

1. Wear the recorder every working day.
2. Each evening, review what the assistant produced and correct it.
3. Note what it got wrong, then add a rule to the standing instructions.
4. After a week, decide what to widen: more connectors, more autonomy, or a scheduled daily run.

How the transcript reaches the agent: the Plaud connection lets your agent pull your transcripts and notes. Whether that happens the moment you stop speaking or on a schedule depends on the agent. Always-on agents such as Dot and Duet can check on their own, while a regular chat pulls only when you ask. Start by asking manually, then move to a daily schedule.

Practical takeaway

• Capture, connect, act. Do them in that order.
• Draft only until the assistant has earned more.
• One folder, one week, one measured result.