The AI Communication Stack: The Tools and Workflows Comms Teams at Cisco, NASCAR and GitHub Actually Use

Which AI tools for communications are teams actually using, and how are they getting people to adopt them? That was the focus of the closing panel on Day 1 of ALI Conferences’ AI for Communications Virtual Certificate Series on August 18: The AI Communication Stack: Tools, Workflows and Resources Teams Actually Use.

Moderated by chairperson McClain Houston of AArete, the panel brought together five communicators working in very different environments, from a firewall-protected tech giant to FDA-regulated medical devices to a sports organization building its AI program from scratch:

• Austin Roth-Eagle, AI and Automation Lead, Global Communications, Cisco
• Melody Warner, Global Content and Communications Manager, Carestream
• Kam Hodge, Internal Communications Manager, NORC at the University of Chicago
• Tyrese Manigault, Employee Engagement, NASCAR
• Ari LiVigni, Senior Developer Advocate, GitHub

Here’s what their AI stacks look like and what they’ve learned.

Five AI stacks, five different starting points

Cisco: model-agnostic, with data as the differentiator

Cisco gives its communications team access to the leading frontier models through an internal tool that sits behind the company firewall. Roth-Eagle said the model matters less than what sits on top of it. “The consistent differentiator between mediocre AI output and high quality output is access to the right data,” such as custom projects or Gems connected to your own content.

His team has built “playbooks,” step-by-step instructions that run the same way every time, for recurring work like a monthly newsletter. Now it is moving from chatbots (ask a question, get an answer) to agent skills that take action. His example: an agent that researches how a competitor is positioning a topic in the media, delivers a brief every day, saves it to SharePoint for the whole team, and improves based on replies. Cisco’s internal model guide recommends Claude for writing, ChatGPT for structured reasoning and Gemini for research.

Carestream: making Copilot work in a regulated industry

For Warner, whose company makes X-ray devices, Microsoft Copilot is the most valuable tool precisely because it lives inside the company’s secured environment and follows industry rules. Her team loaded an official brand asset library into SharePoint so that when employees prompt Copilot in PowerPoint, it defaults to approved logos, colors and images. Her tips for setting up a brand kit in Copilot:

• Include only foundational truths about your brand.
• Keep each image library to five or six images so neither people nor Copilot get overwhelmed.
• Choose assets you’ll never need to chase licensing for.
• Set a review cycle.

She has also found many requests for “AI” are really requests for automation, such as Outlook rules or better OneDrive permissions.

NORC at the University of Chicago: an in-house AI platform

NORC built its own internal AI platform, Nova, which offers several models, alongside Microsoft 365 Copilot and Wrike for project management. As a research organization, NORC treats staff as its first test subjects: AI use is part of performance evaluations and the five-year plan. Legal and compliance review AI addendums for tools like its newsletter platform every year. Hodge said the internal platform grew partly out of the communications team being ahead of the curve with its own AI subscriptions.

NASCAR: starting from zero

Manigault said NASCAR essentially started with nothing. A 2025 task force of teams most likely to benefit tested use cases before rollout, with legal and IT involved throughout and a new Director of AI as a partner. Many employees didn’t realize how much the Copilot they already had could do. Manigault uses an agent that sends a summary of priorities and red flags every Monday. The guiding principle is that AI “comes alongside you,” not something forced on people or meant to replace them.

GitHub: agentic workflows

LiVigni works in GitHub Copilot with multiple models, using agentic workflows and an “autopilot” mode to hand off whole tasks for later review, plus Gemini notebooks to organize context. His biggest lesson is that the context you give AI determines the quality of what you get back. He has even bundled his agent skills and custom agents into a shareable plugin.

Where to start when Copilot is your only approved tool

One attendee asked what to prioritize in a highly regulated organization where Copilot is the only approved AI tool. The panel’s advice:

• Start with the tools people already use every day, like Word, Teams and SharePoint, rather than the most complex capabilities (Manigault).
• Bring IT and security in from day one, so a red flag doesn’t appear right before launch. AI is only as good as the information it can access, so don’t give it access to what you don’t want it to use.
• Brand Copilot and PowerPoint first. Houston said branded templates and approved images removed her organization’s biggest point of friction.
• Create a clear request process and point of contact for premium AI licenses and questions (Hodge).

Training that actually drives AI adoption

• Skip the long documents. “People don’t like to read,” Manigault said. NASCAR uses short videos, lunch-and-learns, high-level graphics and a dedicated AI segment in a town hall, with deeper resources for anyone who asks.
• Office hours and recordings. NORC runs monthly Microsoft 365 office hours with its resident Microsoft MVP and monthly AI trainings, then promotes the recordings in the weekly newsletter and in emails to directors.
• A five-minute tip of the week. Cisco replaced long workshops with a live screen share on weekly calls that runs a real playbook or skill from start to finish. Roth-Eagle said there are plenty of conversations about AI’s long-term impact and not enough about “how do you use AI better in the next 30 minutes.”

One piece of advice from each panelist

• LiVigni: Stay curious and play with the tools. You’re still the human in the loop.
• Manigault: Meet people where they are. Not everyone embraces AI the same way.
• Hodge: Upload your brand bible and editorial guide to your AI tool, and practice your prompting.
• Roth-Eagle: Break your workflow into small steps. That’s where it becomes obvious which parts need AI and which need human creativity and judgment.
• Warner: Be friends with IT.

Frequently asked questions

What AI tools do communications teams use?

Panelists used Microsoft 365 Copilot, Claude, ChatGPT, Gemini, GitHub Copilot and in-house platforms that give secure access to several models. Most said the data and context connected to the model matter more than which model you choose.

How do you set up Copilot for brand consistency?

Store a small, approved brand asset library in SharePoint so Copilot defaults to official logos, colors and images, keep each library to a handful of images, avoid assets with licensing limits, and review it regularly.

How do you drive AI adoption on a communications team?

Short, practical demos work better than long documents. Panelists recommended weekly five-minute tips, recorded office hours, town hall segments and starting with tools employees already use every day.

Build your own AI workflow

Want practical prompts and workflows you can use right away? Join the Practical AI for Communications Virtual Conference on January 20-21, 2027, or the AI for Communications Virtual Masterclass on December 10, 2026. Government communicators can join the AI for Government Communications Virtual Conference on October 14-15.