How are you using AI/Agents to automate workflows without losing human oversight?

Carlos.Medina

Hi all!

I work as a QC Bioanalytical Scientist in a regulated biotech/CDMO environment, and one area I’m interested in exploring is how Smartsheet MCP, Agents, APIs, and automation can reduce administrative work while keeping people focused on the scientific and quality decisions that require human judgment.

A typical QC project can involve multiple methods, milestones, documents, dependencies, stakeholders, and action items across QC, QA, Manufacturing, Analytical Development, Program Management, and external clients.

Today, a lot of the coordination around that work can still depend on manual status updates and follow-up.

I’m interested in exploring workflows where Smartsheet could help with things like:

• Identifying overdue actions and upcoming milestones
• Summarizing project status across multiple sheets
• Flagging dependencies or potential timeline risks
• Preparing concise updates for different stakeholders
• Automating routine follow-up while keeping technical decisions with the appropriate subject matter experts

The design constraint that interests me most is human oversight. In a regulated environment, I don't necessarily want an AI agent making a quality or scientific decision. I want it to help surface the right information so the right person can make that decision faster and with better visibility.

I’m still early in exploring MCP and Agents, so I’d love to learn from others:

Where have you found the best boundary between automation/AI and human decision-making?

And if you’ve built something similar for a regulated or highly controlled environment, what did you learn along the way?

Carlos Medina
Thermo Fisher Scientific
https://www.linkedin.com/in/cmedinachan

Galina Jordanowa

Hi Carlos, welcome, and thank you for framing the question around human oversight. That's the right starting point in a regulated environment, and it's the design principle we hear most from teams in similar settings.

A pattern that tends to work well is to let AI read, summarize, and flag, while people decide and approve. In practice, that could look like:

• Read-only first. Start with use cases where the agent only reads across sheets: overdue actions, upcoming milestones, dependency risks, cross-project status. Nothing changes in the plan, so the risk is low and the value shows up quickly.
• Draft, don't send. Let the AI prepare stakeholder updates and follow-ups as drafts, and have the owner review them before anything goes out.
• Keep decisions in an explicit approval step. Where a quality or technical decision is involved, route it through a Smartsheet approval workflow so a named person signs off and there's a clear record.
• Expand write access gradually, and only for low-risk, routine fields once the team trusts the output.

Your list of use cases maps nicely to that progression. Most of it sits in the "surface the right information" zone you described.

I'd be glad to continue the conversation, and I'd love to hear what you build as you go.