AI in Healthcare, Meetings, and the Workplace
Dear Friend,
In our AI Smart Lab session last week, we had conversation about where people fit once AI becomes part of the work.
Victoria shared that she knows how to build a newsletter skill, but that skill cannot do much for a client until she understands their process, has access to their documents, and knows who is responsible for each step.
Emma pointed out that when AI writes processes without input from the people doing the work, the results are generic. The company loses the details that make its operations custom for us. Over the last 4 years, we have written more SOPs and workflow documents than ever before. In all that time, only 1 client has come to us with their processes fully documented.
The people who know the work are the ones who can make AI useful fast. Who on your team holds the process knowledge that nobody has written down yet?
What to Check Before You Connect an AI Agent
When you link an AI agent to your tools, it takes on the reach of whoever clicked approve, and that is often the busiest person with the widest access. Our article shows you how to see what each agent can open, and why 74% of firms in a Cloud Security Alliance survey say agents often get more access than the work needs. It also covers old app links, since 47.2% of the links one study reviewed sat unused for 90 days with their access fully intact. It recommends giving each agent its own login and keeping it read-only for the first two weeks. Read the full article here.
Our Most Requested Service: AI Strategy and Consulting
The question I hear from business owners is often about where to begin. They know AI can help their teams. They are trying to figure out which tools to trust, where their information goes, and how to introduce AI without creating more work for everyone.
We start by looking at the work your team already does. We identify repetitive processes, map the information moving between people and systems, and develop a phased plan. Training is part of that work because your people need to understand what the technology does and how to manage it.
One of our medical training clients automated 99% of its customer service work and achieved $177,000 in combined savings. Nobody was laid off. The team gained new skills and took on more valuable work.
That is the kind of result I want to see more organizations pursue.
This week’s stories
Zoom Wants AI to Turn Meetings Into Completed Work
Zoom announced a Unified Context Layer at its October 20 Zoomtopia conference. The technology is designed to connect information from meetings, calls, and other conversations so AI tools can use that context to support follow-up work. Zoom also introduced AI capabilities across its workplace and customer experience products.
My Take: Meeting notes have been one of the easiest places for teams to start using AI. The harder part comes after the meeting, when decisions need to become tasks and someone needs to own them. At AISV, we have worked on connecting meeting information with task management in ClickUp. The important decisions concern who receives each task, what the deadline is, and how the team confirms that the work was completed. I like the direction Zoom is taking because people spend a considerable amount of time transferring information between systems. I will be watching how well its tools handle accountability when a conversation involves several people, competing priorities, and decisions that change over time.
Source: Zoom
AstraZeneca and Oracle Plan AI Tools for Clinical Trials
AstraZeneca and Oracle announced a collaboration on October 20 to explore AI applications in clinical research and healthcare. Their proposed work includes identifying patients who may qualify for clinical trials, supporting safety monitoring, and bringing research findings into clinical workflows. The companies are exploring these applications under a non-binding agreement.
My Take: Clinical trial recruitment is a meaningful place to apply AI. Patients can miss opportunities to participate in research because their information is scattered across systems or because identifying eligibility takes time. I am encouraged by projects that make specialized knowledge more accessible to the people providing care. The distinction between a proposed collaboration and a proven clinical result matters here. We need to see how these systems perform with real patients, including people whose records are incomplete or whose circumstances differ from the populations represented in the data. I want to see patients gain access to better information while healthcare professionals retain responsibility for clinical decisions. Technology has a valuable role in care when it supports the relationships and judgment that patients depend on.
Source: AstraZeneca
Microsoft Raises the Security Question Around AI Agents
Microsoft published an October 19 discussion titled Securing the Agentic Insider, addressing security challenges created when AI agents operate inside organizations.
As companies give AI systems access to internal information and applications, security teams must consider how those systems receive permissions, use data, and perform actions.
My Take: An AI agent with access to your business systems deserves the same attention you give any other human that can read information or take action. We have already seen how an automated process can consume resources without attracting attention because the process continues to work without being checked.. During an internal review of our ClickUp agents, we found that one task intake agent was responsible for a large share of our credit consumption because it was doing more than we needed. That experience reinforced the importance of monitoring what agents do after deployment, how data is impacted, and token usage levels. Security extends beyond protecting a password. It includes deciding what an agent can access, who approves its actions, and who can stop it when something goes wrong. And even how to tell that something went wrong. With these agents operating in the background, a lot of damage can be incurred before any human notices. Give every agent a defined responsibility AND a human boss.
Source: Microsoft
Researchers Examine Who Gets Left Out of The AI Conversation
Hong Kong Baptist University hosted an international conference on AI, technology, and emerging media from October 16 to 18. The university reported more than 200 participants, with discussions addressing responsible AI communication, inequality, and unequal access to technology.
The conference brought together researchers and students from institutions across several regions.
My Take: The information we use to train and evaluate AI deserves more scrutiny. During an earlier AI Smart Lab discussion, we examined AI-generated professional biographies that described men through their credentials while presenting women through different language and assumptions. Ending a woman’s bio with “Yes, she belongs in the room too.” Misogony at it finest, automated and programmed into our tools. That kind of bias can shape how people are represented long before anyone makes a formal hiring or promotion decision. AI systems learn from material created by societies with long histories of unequal access to education, publication, leadership, representation in literature, and recognition. I want more researchers asking whose experiences are represented, whose voices are missing, and how those differences affect the answers people receive. These answers are training our youth and we best not forget that.
Source: Hong Kong Baptist University
OpenAI Models Are Coming to Google Cloud
OpenAI announced on October 20 that its models and Codex workflows are coming to Google Cloud through Gemini’s Model Garden and Google Cloud Marketplace.
The arrangement gives organizations another route for accessing OpenAI technology through an established cloud platform.
My Take: This is a growing trend with some tools allowing you to choose from over 400 models all in one place. When we help clients evaluate technology, we look at the tools they already pay for, the information those tools can access, and how much work it would take to introduce something new. The model is one part of that decision. Billing, data handling, integrations, and ongoing management affect whether the investment makes sense. A wider selection of models inside established platforms may give organizations more flexibility when designing their systems. We are in a moment of consolidation. You will need fewer tools to do all of the work in your organization. You’ll be able to build and replace with simpler solutions. We are fully embracing this moment. Need help? I’m here for you. [email protected]
Source: OpenAI
Tool picks of the week
Read AI: Our favorite An AI meeting assistant that records meetings, prepares summaries, and identifies follow-up items.
Canva: A design platform with AI features for creating and adapting visual content.
Snitcher: Identifies the companies visiting your website, reveals contact details, and triggers outreach automations.
Got a burning question, a fresh take, or just want to share your latest AI wins? Hit us up at [email protected]. Your insights keep this community growing and thriving!
Have a great day!
-Nicole A. Donnelly
Founder, AI Smart Ventures
AI Strategy – AI Training – AI Consulting – AI Implementation
