Skip to content

AWS re:Invent 2025 - Automate insights and drive innovation with cloud and AI solutions (IND384)

This presentation at AWS re:Invent 2025 focuses on how leading companies are leveraging cloud and AI solutions to automate insights and drive innovation.

Aperna Galaso (0:05-4:57) opens the presentation by highlighting the rapid evolution of Generative AI and the challenges CPG companies face in keeping up with technological advancements while managing tight margins and evolving consumer expectations. She emphasizes the need for IT and business leaders to collaborate and for IT to become a strategic partner in shaping the business direction.

Chris Hesse's Presentation: Building an Automation Platform at Mondelēz (5:05-25:20)

Chris Hesse, from Mondelēz, discusses their cloud journey and the development of an automated, self-service platform to enable agility and innovation within the company.

Key Takeaways from Chris Hesse:

  • Agile Foundation (5:09): Mondelēz, with its aggressive acquisition and divestiture strategy, required a sound and agile foundation to enable the business to move quickly.

  • Cloud Journey Evolution (6:50): Their cloud journey started in 2018 with a lift-and-shift approach and managed service providers. In 2021, they established a dedicated cloud engineering team, which grew from one person to over 30 engineers by 2024.

  • Guiding Principles for Green Field Environment (8:26): When building their cloud engineering capability from scratch, their principles included:

  • Self-service platform (8:55): Empowering developers to do the right thing quickly and easily.
  • Secure by default (9:04).
  • Automation of everything (9:07): Everything as infrastructure as code.

  • Micro-segmentation (9:11): Planning for thousands of accounts from day one.

  • Golden Path and Enabling Constraints (9:50): Providing guardrails and controls that enable developers while ensuring security and compliance.

  • Infrastructure as Code (10:26): The core of their environment is that all provisioning, including GitHub repositories, branches, standard roles, and CI/CD pipelines, is automated through infrastructure as code (10:34-11:00).

  • Automated Provisioning (11:17): A single request automatically provisions GitHub repositories with branch protection rules, Terraform deployment tools (SpaceLift), and standard roles with defined capabilities (11:19-12:00).

  • GitOps Enforcement (12:27): They strictly enforce GitOps, where all changes are made via pull requests to static branches, triggering automated plans and applications (15:27-16:20). This ensures a controlled path to production.

  • OPA Policy Enforcement (16:37): Open Policy Agent (OPA) policy enforcement is crucial for attaching security standards and guardrails to every pipeline, ensuring all plans and applies meet the latest standards (16:40-17:02). This allows for centralized updates impacting thousands of CI/CD stacks and hundreds of AWS accounts at once (18:22-18:30).

  • Read-Only Access (19:49): Developers primarily have read-only access to AWS, preventing manual edits and enforcing the infrastructure as code approach (19:49-20:02).

  • Micro-Segmentation for Cost Transparency (22:09): Separating costs by AWS account for each application provides clear cost transparency and ownership, allowing product owners to track their spend by the second, day, or month (22:50-23:15).

  • Future Steps (23:24): Mondelēz is moving towards a product and platform organizational journey, leveraging their built platform to drive more transparency, wrapping up legacy cloud migration, refactoring and re-engineering applications to be serverless where possible, and deepening their strategic partnership with AWS (23:24-25:20).

JC's Presentation: Revolutionizing Cost Governance with AI Agents at W.W. Grainger (25:34-47:11)

JC from W.W. Grainger discusses their FinOps journey and how they used generative AI to scale cost governance and deliver actionable insights to business leaders.

Key Takeaways from JC:

  • Strategic Challenge (28:01): Critical insights were trapped in complex systems, and stakeholders lacked the expert knowledge to interpret data. Manual communication couldn't keep up with enterprise growth, especially as federated cloud adoption accelerated.

  • Target Demographic and Solution (29:11): They identified product and technical domain leaders as the highest return-on-value demographic. Their solution was to deliver executive summaries, familiar to leaders, directly to their inboxes (29:57-30:04), requiring no new tools or behavior changes.

  • Impossible Math Problem (30:42): Manually generating these summaries was time-consuming (one full engineering day per summary), making it impossible to scale to their audience of initially 30, growing to 50 recipients.

  • Generative AI as the Solution (31:49): They chose generative AI because programmatic solutions lacked the nuance and tailored analysis of their FinOps specialists. With the right constraints, templated prompts, and organizational data, generative AI could meet the specialist quality at enterprise scale (32:54-33:10).

  • Challenges in Productionizing AI (33:19):

  • Terraform limitations (33:56): Traditional IaC tools like Terraform were not keeping up with the pace of generative AI innovation. They pivoted to Strand's agents for defining agents in source code (34:07-34:15).
  • Deterministic Outcomes from Probabilistic Models (34:27): They needed consistent, deterministic outcomes from probabilistic models, which they achieved by developing a headless MCP client with no human interaction (34:39-34:48).
  • Enterprise Constraints (34:51): They had to find a best-in-class solution that met security, procurement, and legal requirements, leading them to Agent Core (35:06-35:39).

  • Lack of Documentation (35:58): Agent Core was in preview and lacked documentation, requiring deep dives into its source code (36:00-36:07).

  • Project Nightingale (36:19): This project leverages Agent Core and Amazon Bedrock with cloud models to construct analyst-quality FinOps cloud cost intelligence at enterprise scale. It consists of two key agents: a FinOps analyst and an executive summary writer (36:35-36:50).

  • Multi-Agent Architecture and Workflow (36:51): Their solution uses a structured, sequential workflow (agentic pipeline) with templated prompts as input, ensuring consistent and accurate output. This involves a FinOps agent performing analysis using billing and optimization hub data, which is then handed off to a summary agent that synthesizes the analysis into an executive-friendly HTML narrative and sends it via email (37:07-41:47).

  • Output and Impact (43:06): The executive summaries provide headline metrics, key points, a deep dive into cost increases/decreases, highlight anomalies, and offer actionable recommendations tied to a dollar amount (43:32-45:20). This allowed them to produce summaries in minutes instead of a full day, driving engagement and action among leaders (44:39-44:41).

  • Future Outlook (46:01): The approach is not unique to FinOps and can be applied to other teams like DevOps, SRE, and platform engineering. They plan to extend the tool to serve multiple personas (product owners, finance, procurement, executives) with different signal channels (46:05-47:11).

Presenters: Mondelez International and W.W. Grainger representatives