Hi!

There’s a lot of interest right now across the T2B community in how to build and use AI agents. But before we get to the how-to, there’s some important foundational thinking to do.

At our Q3 AI Roundtable event, Skye King and incoming roundtable chair Monica Aguilar helped us understand how agentic AI changes the way we need to think about workflows: what’s worth delegating, what an agent needs in order to do it well, where human judgment belongs, and what guardrails and oversight need to be in place.

Through frameworks, examples, and live demonstrations, they gave us a foundation for thinking about agentic workflows before jumping into the tools themselves.

T2B Pro and Student members can catch the replay here.​

10 takeaways from How Agentic AI Is Moving Strategic Communicators From Prompting to Delegating

1. Look for a useful place to start. Repetitive work like audits or compiling information from multiple sources can be good agentic use cases. Delegation frees leaders to focus on judgment, creativity and strategy, but sometimes simpler AI tools or processes are better.

2. Own agentic AI as a leadership decision. As AI systems become capable of taking on more complex work, leaders still need to determine the outcome, decide what should be delegated, establish appropriate guardrails and oversight, and remain accountable for the results.

3. Set data, compliance, and risk boundaries first. Define what agents can access, what stays restricted, which systems they can use, and what approvals are required. Existing rules for sensitive or regulated data still apply. Clear guardrails matter.

4. Run any workflow through 4 questions before delegating. Ask whether AI should solve this at all, where human judgment needs to stay, who gets hurt if the output is wrong, and who is accountable for it.

5. Fix the workflow before you automate it. Start from the business need rather than the tool. Garbage in, garbage out still holds. Map the process as if AI weren't in the picture, because automating a broken workflow only makes the mess move faster.

6. Start small with one task and one agent. Start with a bounded, low-risk use case where delegation can add value. A manageable scope makes it easier to test assumptions, spot failures, refine instructions and oversight, assess ROI, and learn before scaling.

7. Lean on AI to help you build the agent. A skill is an SOP that gives an agent your knowledge and goal. Use AI to turn that knowledge into clear instructions. Then define its brain (LLM), tools (email, drives, repositories), and loop to observe, plan, and act.

8. Give clear instructions and verify what comes back. Clear boundaries and trusted sources can improve agent performance, but they don’t prevent hallucinations or guarantee compliance. Test outputs, add human or QC review where needed, and monitor performance over time.

9. Keep people in the loop wherever the stakes are real. The higher the risk to patients, reputation, or regulatory standing, the more human judgment and stronger guardrails you need. Build stage gates and a separate (outside the loop) check to catch drift.

10. Watch the meter and the maintenance. Agents cost tokens and dollars, and can drift over time. Set run frequency, review outputs, and monitor based on risk and complexity. Test before relying on workflows, and plan for changing costs and future budgets.

One of the biggest things I took away from the conversation is just how much thinking needs to happen BEFORE building an agent. Agentic AI asks us to understand our processes differently, define the outcome we’re trying to achieve, decide what should and shouldn’t be delegated, and be deliberate about where human judgment stays.

Ready for the how-to? Join our Virtual Biopharma Comms Forum next month

This last AI Roundtable session was intentionally foundational. At our October 6th & 7th Virtual Biopharma Comms Forum, Owning the Outcome, Skye will be back for a 75-minute, hands-on AI workshop that picks up where this conversation left off.

We’ll go deeper into the practical steps behind building an agentic workflow, from identifying the right use case and mapping the process to thinking through the instructions, components, testing, and oversight needed to make it useful.

>> More details + RSVP here.​

Thank you to everyone who joined us for the AI Roundtable webinar and especially to our speakers – Monica Aguilar and Skye King – and our knowledge partner, Syneos Health Communications

For those of you in the U.S., enjoy the holiday weekend!

Lynnea

​
​