--- title: "79 Workflow Tools, Analyzed – Lessons Learned" newsletter: "MLOps Community" date: 2025-02-18 source: https://aaif.live/newsletters/mlopscommunity/2025-02-18-79-workflow-tools-analyzed-lessons-learned --- # 79 Workflow Tools, Analyzed – Lessons Learned *Plus, making MLOps pipelines more flexible with artifacts, a tasty blog on data transformations, Tooling Tuesday, and the sandbox.* *MLOps Community — Agentic AI Foundation, 2025-02-18* https://x.com/untitled01ipynb/status/1891122010245255398 I'm writing this before the launch of Grok 3, so the above is a bit of a gamble. But the man who delivered full self-driving in 2017, a Hyperloop in Chicago, a million robotaxis in 2020, and a $25,000 Optimus robot is surely a man of his word. And good for our word, here's the first of two newsletters this week! Expect new features and your old favorites over two editions a week. Let us know what you want to see more of here [https://go.mlops.community/NewsletterFeedback], or email steve@mlops.community ## Evolving Workflow Orchestration 2 min read Evolving Workflow Orchestration Alex Miłowski // Entrepreneur and Computer Scientist @ Self Open laptop ➡️ check ticket board ➡️ sigh ➡️ close laptop ➡️ make coffee Workflows are everywhere, but structuring them well is a real challenge. Alex talked me through the evolution of workflow systems, from rule-based engines to modern DAG-based tools, how they fit into MLOps, and the distinction between business process modeling tools and ML-specific orchestration. He shared what he found after analyzing 79 workflow tools, comparing their differences, adoption, and trends, including: * Business process automation dominates – Around 46% of tools focused on enterprise workflows rather than ML pipelines. * ML and data science workflow tools are growing – Most emerged after 2015, reflecting the rise of ML automation. * Newer tools focus on developer experience – Many have shifted toward Python-based interfaces instead of DSLs or YAML. We also covered agent-driven workflows, balancing automation with human oversight, and the challenges of governing AI-powered processes—particularly in regulated industries where auditing and control are critical. Open link ➡️ make coffee ➡️ listen to episode Video [https://go.mlops.community/11jwmu] || Spotify [https://go.mlops.community/1a09bd] || Apple [https://go.mlops.community/d2tslo] ## Lessons Learned from the Gemini Long Context Kaggle Competition 2 min read Lessons Learned from the Gemini Long Context Kaggle Competition With thanks to Médéric Hurier for their contribution. If I said you'd enjoy a blog about windows, you might question my sanity, but context is key because this is a blog about long context windows. Gemini 1.5’s expanded 2M-token context window allows AI to process entire codebases, long documents, and complex datasets in a single pass. A Kaggle competition explored its potential, with one entry focused on turning open-source textbooks into an interactive AI tutor. The system retrieved relevant textbooks, extracted text using pypdf, and created a personalized learning assistant. One of the most useful features was context caching, which significantly reduced costs and improved efficiency: * Instead of reprocessing the full textbook for every query, the system stored previous context, cutting costs by 75%. * This made it more practical for real-world applications where long documents need repeated analysis. Despite its strengths, the approach had challenges. Response times were slow—around 25 seconds per query—and Gemini’s file API struggled with large textbooks. These limitations highlight the need for optimization in long-context AI applications, particularly in education. The experiment also suggested that traditional RAG pipelines might become less necessary as models improve at handling large-scale direct context ingestion. Don't let this window of opportunity close: click below to read it! Read it here [https://go.mlops.community/b4idfg] ## Job of the Week [https://go.mlops.community/ky65x8](https://go.mlops.community/ky65x8) ## Coming Soon: Tuesday Tool Time 🔧 1 min read Coming Soon: Tuesday Tool Time 🔧 MLOps tools, tested and reviewed - every Tuesday We’re kicking off a new feature - Tuesday Tool Time - where we’ll review MLOps tools, test what’s worth your time, and help you navigate the ever-growing stack. But before we get started, we wanted to hear from you. What tools are on your radar? What categories do you want us to compare? Any overhyped tools you’re skeptical about? Drop your suggestions here [https://go.mlops.community/ToolTalkFeedback] or email steve@mlops.community - looking forward to them! ## The Sandbox 1 min read The Sandbox A little place to test some ideas Trying out a few things here - let us know what you think here [https://go.mlops.community/hx6agg] or email steve@mlops.community Back to the Feature A highlight from last week A lot of people slept easier after this episode on AI SREs and knowledge graphs. Willem shared how mapping system dependencies can speed up root cause analysis and prevent those 3am wake-up calls. Video [https://go.mlops.community/qto1p4] || Spotify [https://go.mlops.community/k7jxr3] || Apple [https://go.mlops.community/cy73ap] Tech Teaser A mini MLOps mindbender Roughly how much high-quality video could OpenAI's Sora model generate in the time it takes to watch the recent BAFTA Best Film winner, Conclave? a) About 1 hour 30 minutes b) About 50 minutes c) About 3 hours 30 minutes d) About 2 hours 15 minutes Click here [https://go.mlops.community/TechTeaserAnswers] for the answer. Interested in partnering with us? Get in touch: partners@mlops.community Thanks for reading. See you in Slack [https://go.mlops.community/slack], YouTube [https://www.youtube.com/channel/UCG6qpjVnBTTT8wLGBygANOQ?view_as=subscriber], and podcast [https://home.mlops.community/public/content/] land. Oh yeah, and we are also on X [https://twitter.com/mlopscommunity] and LinkedIn [https://go.mlops.community/linkedin]. The MLOps Community newsletter is edited by Jessica Rudd [https://www.linkedin.com/in/jmrudd/]. --- Source: https://aaif.live/newsletters/mlopscommunity/2025-02-18-79-workflow-tools-analyzed-lessons-learned