# How ML teams ship together A pack of 8 sessions from the MLOps Community YouTube channel, in the order to watch them. 7h 38m of video. Page: https://mlopstalks.com/packs/how-ml-teams-ship-together The model is ready, but the application team has no time to integrate it. A data problem keeps returning because the producer and consumer expect different things. Everyone supports the project, yet nobody will change the business process that would use its predictions. Begin by agreeing on the outcome and examining the actual work between disciplines. Then consider how to staff missing capabilities, assign responsibility for a running service and transfer maintenance without abandoning its new owner. The later accounts connect everyday collaboration to leadership: small releases, room to revise decisions and written responsibilities. Finish by distinguishing stakeholder agreement from a commitment to act. The point is a team that can deliver and improve a useful system, with enough shared understanding to keep doing so when people move on. ## This pack is for you if - Your models stall between data science, engineering and the team expected to use them. - You are adding people but have not identified which missing capability delays delivery. - A system reached production without clear responsibility for its data, maintenance or business result. ## The talks, in order ### 1. What Business Stakeholders Want to See from the ML Teams Peter Guagenti, Tabnine | 1:21:28 | MLOps Podcast Video: https://www.youtube.com/watch?v=8zZ7-TCdMLM Summary: https://mlopstalks.com/talks/what-business-stakeholders-want-to-see-from-the-ml-teams.md Why first: Guagenti asks what the business should do differently before discussing an algorithm or implementation. His practical distinction between data, insight and action exposes missing requirements on both sides of the conversation. Establish that shared outcome before reorganizing the people who will deliver it. ### 2. Data Scientists & Data Engineers: How the Best Teams Work Beverly Wright, Wavicle Data Solutions & Sadie St. Lawrence, Human Machine Collaboration Institute / LinkedIn Learning & Joe Reis, Ternary Data & Victor Cuadros, Microsoft | 27:55 | DE4AI 2024 Video: https://www.youtube.com/watch?v=A9oLe3bqEpY Summary: https://mlopstalks.com/talks/data-scientists-data-engineers-how-the-best-teams-work.md Why second: This panel examines the everyday cost of separate disciplines that cannot see the finished product. Pair programming and learning the other role's constraints make upstream dependencies less abstract. Those practices give shared requirements a way to survive the handoff from a conversation to working code. ### 3. Building Better Data Teams Leanne Fitzpatrick, Financial Times | 1:01:40 | MLOps Coffee Sessions Video: https://www.youtube.com/watch?v=JxVS3-4wyKc Summary: https://mlopstalks.com/talks/building-better-data-teams.md Why here: Fitzpatrick identifies engineering capacity as the constraint behind an apparently undersized data science function. Embedded colleagues connect technical work to commercial users as well as helping models reach production. Staffing follows the missing work, rather than assuming that another model builder will remove a delivery bottleneck. ### 4. MLOps as Tool to Shape Team and Culture Ciro Greco, Coveo | 43:02 | MLOps Coffee Sessions Video: https://www.youtube.com/watch?v=OAB-fu9ylZo Summary: https://mlopstalks.com/talks/mlops-as-tool-to-shape-team-and-culture.md Why here: Greco gives domain expertise a concrete responsibility: the final data preparation belongs with the people who understand its use. A small working system then tests that understanding with product and business colleagues. This keeps cross-functional collaboration tied to a deliverable and a feedback opportunity. ### 5. Autonomy vs. Alignment: Scaling AI Teams to Deliver Value Grant Wright, SEEK Ltd. | 50:40 | MLOps Coffee Sessions Video: https://www.youtube.com/watch?v=Gr69acrT8HE Summary: https://mlopstalks.com/talks/autonomy-vs-alignment-scaling-ai-teams-to-deliver-value.md Why here: Wright describes teams accountable for an operating service rather than a model file. Written partner agreements specify behavior, consumers and maintenance, while the reorganization creates real friction for previously autonomous specialists. The account shows what shared responsibility requires beyond drawing a new organization chart. ### 6. Data Governance and AI Alexandra Diem, Gjensidige | 1:05:46 | MLOps Podcast Video: https://www.youtube.com/watch?v=tLlDRApuP7Y Summary: https://mlopstalks.com/talks/data-governance-and-ai.md Why here: Diem's enabling team temporarily pairs with business-unit analytics teams, builds capability and explicitly transfers model maintenance. That offers a different answer to centralization: specialist help can end without leaving an unsupported artifact behind. Starting with one complete use case makes the transfer teachable before the approach spreads. ### 7. Culture and Architecture in MLOps Jet Basrawi, Satalia | 53:41 | MLOps Coffee Sessions Video: https://www.youtube.com/watch?v=uV676_YLP98 Summary: https://mlopstalks.com/talks/culture-and-architecture-in-mlops.md Why here: Basrawi connects small batches and preparation for failure to the leadership conditions that let teams learn. His criticism of ceremonial process matters after defining the roles: a written agreement cannot replace frequent conversation about a customer problem. Teams also need permission to revise decisions when the evidence changes. ### 8. Lessons on Data Teams Leadership Luigi Patruno, 2U, Inc | 1:13:32 | MLOps Podcast Video: https://www.youtube.com/watch?v=GqIFGvj8aLc Summary: https://mlopstalks.com/talks/lessons-on-data-teams-leadership.md Why last: Patruno distinguishes a stakeholder who agrees with an idea from one who will spend time and change operations to use it. His shared, version-controlled handbook preserves the delivery practices as the team grows. Finish by naming the committed partners and documenting how the next project will reach a measurable result.