--- title: Netflix Open-Sources Agentic Workflow for Causal Inference description: Netflix has open-sourced an agentic workflow for Observational Causal Inference (OCI) that automates repetitive parts of causal analysis, such as sensitivity... canonical: https://daily.dev/posts/netflix-open-sources-agentic-workflow-for-causal-inference-qsi2krkyg twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Netflix Open-Sources Agentic Workflow for Causal Inference | daily.dev og:description: Netflix has open-sourced an agentic workflow for Observational Causal Inference (OCI) that automates repetitive parts of causal analysis, such as sensitivity... og:url: https://daily.dev/posts/netflix-open-sources-agentic-workflow-for-causal-inference-qsi2krkyg og:image: https://api.daily.dev/og/posts/QSi2krkyg.png og:image:alt: Netflix Open-Sources Agentic Workflow for Causal Inference og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Netflix Open-Sources Agentic Workflow for Causal Inference **[InfoQ](https://daily.dev/sources/infoq)** · 3 min read · 5 upvotes · 1 comments ## Summary Netflix has open-sourced an agentic workflow for Observational Causal Inference (OCI) that automates repetitive parts of causal analysis, such as sensitivity checks and iteration tracking, while leaving question framing and result evaluation to humans. An actor agent fills in a templated Jupyter notebook based on a human-provided analysis plan and executes it, while a critic agent reviews the output, rates it, and recommends spec changes. Evaluated on the ACIC competition dataset, the workflow performed competitively against benchmark systems. In a case study estimating retention impact of new entertainment types, the workflow produced an estimated effect just 25% of a naive Claude-based linear regression baseline, after the critic flagged early adopter bias and a failed placebo test. The oci-agent source code is available on GitHub. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is Netflix's oci-agent and what does it do? Oci-agent is an open-sourced agentic workflow from Netflix for Observational Causal Inference that automates repetitive parts of causal analysis. Given observational data and a human's analysis plan, an actor agent fills in and runs a templated Jupyter notebook while a critic agent rates the output and recommends spec changes, including sensitivity checks like placebo tests, before producing a report and next steps. _daily.dev surfaces releases like this for engineers building agentic workflows around data analysis._ ### How does Netflix's oci-agent workflow compare to just asking an LLM to run causal inference? In a Netflix case study estimating retention impact from new entertainment types, feeding the analysis plan directly to a Claude model produced a simple linear regression estimate, but the oci-agent workflow produced an effect estimate just 25% of that baseline after its critic agent flagged early adopter bias and a failed placebo test. This shows a naive single-model approach can significantly overstate causal effects without structured checks. _Teams evaluating agent-based analysis pipelines can track cases like this on daily.dev before trusting a single model's output._ ### How does the oci-agent critic agent evaluate the actor agent's causal analysis output? The critic reviews the notebook output produced by the actor agent and assigns one of three ratings: not_satisfactory, satisfactory_with_caveats, or fully_satisfactory, then recommends spec changes such as rerunning with tweaked parameters. It follows playbooks covering known failure scenarios like early adopter bias and failed placebo tests, combining process audits with human oversight since ground truth is unavailable. _daily.dev helps developers building actor-critic agent evaluation loops keep up with real-world design patterns._ ## Community discussion Top comments from developers on daily.dev. **@kartiknvj** · 0 upvotes > The design choice I like here is keeping question framing and result evaluation with the human while automating the repetitive sensitivity checks. That split is usually where agentic workflows earn their keep, since the judgment calls are exactly what you do not want a loop guessing at. Worth a read just for how they drew that boundary. ## Similar posts on daily.dev - [My Experience with GitHub Agentic Workflows](https://daily.dev/posts/my-experience-with-github-agentic-workflows-orst3dlvf) · Thomas Thornton · 0 upvotes · 0 comments - [Agentic AI in observability: accelerating root cause analysis](https://daily.dev/posts/agentic-ai-in-observability-accelerating-root-cause-analysis-htepnjdj4) · The New Stack · 1 upvotes · 0 comments - [A GitHub agentic workflow](https://daily.dev/posts/a-github-agentic-workflow-mwq53dvmu) · ITNEXT · 1 upvotes · 0 comments - [QCon London 2026: Ontology‐Driven Observability: Building the E2E Knowledge Graph at Netflix Scale](https://daily.dev/posts/qcon-london-2026-ontology-driven-observability-building-the-e2e-knowledge-graph-at-netflix-scale-3ezjkkaup) · InfoQ · 1 upvotes · 0 comments - [Agentic Workflows: A Complete Guide for 2026](https://daily.dev/posts/agentic-workflows-a-complete-guide-for-2026-4f1kyn0vl) · Orca Security Blog · 2 upvotes · 0 comments --- Tags: [#open-source](https://daily.dev/tags/open-source), [#data-science](https://daily.dev/tags/data-science), [#ai-agents](https://daily.dev/tags/ai-agents), [#netflix](https://daily.dev/tags/netflix) [View this post on daily.dev](https://daily.dev/posts/netflix-open-sources-agentic-workflow-for-causal-inference-qsi2krkyg) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Netflix Open-Sources Agentic Workflow for Causal Inference","url":"https://daily.dev/posts/netflix-open-sources-agentic-workflow-for-causal-inference-qsi2krkyg","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/netflix-open-sources-agentic-workflow-for-causal-inference-qsi2krkyg"},"datePublished":"2026-08-18T13:02:57.386Z","dateModified":"2026-09-14T08:24:34.087Z","description":"Netflix has open-sourced an agentic workflow for Observational Causal Inference (OCI) that automates repetitive parts of causal analysis, such as sensitivity...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/da420f3d5d10a25a4e21852b5500878a?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/da420f3d5d10a25a4e21852b5500878a?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"InfoQ","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"InfoQ","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/afc3bced3e1e4b188dd9127017a60e0c","url":"https://daily.dev/sources/infoq"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/netflix-open-sources-agentic-workflow-for-causal-inference-qsi2krkyg","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":5},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"open-source,data-science,ai-agents,netflix","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"InfoQ","item":"https://daily.dev/sources/infoq"},{"@type":"ListItem","position":3,"name":"Netflix Open-Sources Agentic Workflow for Causal Inference"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/netflix-open-sources-agentic-workflow-for-causal-inference-qsi2krkyg","comment":[{"@type":"Comment","text":"The design choice I like here is keeping question framing and result evaluation with the human while automating the repetitive sensitivity checks. That split is usually where agentic workflows earn their keep, since the judgment calls are exactly what you do not want a loop guessing at. Worth a read just for how they drew that boundary.","datePublished":"2026-08-18T18:15:27.395Z","url":"https://daily.dev/posts/QSi2krkyg#c-WiKvVvjbz","author":{"@type":"Person","name":"kartik-nvjk","url":"https://daily.dev/kartiknvj","image":"https://media.daily.dev/image/upload/s--3gGgsVCw--/f_auto/v1781456774/avatars/avatar_TvTVeiMdkRCqWUDullFmy?_a=BAMAMiWQ0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/netflix-open-sources-agentic-workflow-for-causal-inference-qsi2krkyg#faq","mainEntity":[{"@type":"Question","name":"What is Netflix's oci-agent and what does it do?","acceptedAnswer":{"@type":"Answer","text":"Oci-agent is an open-sourced agentic workflow from Netflix for Observational Causal Inference that automates repetitive parts of causal analysis. Given observational data and a human's analysis plan, an actor agent fills in and runs a templated Jupyter notebook while a critic agent rates the output and recommends spec changes, including sensitivity checks like placebo tests, before producing a report and next steps. daily.dev surfaces releases like this for engineers building agentic workflows around data analysis."}},{"@type":"Question","name":"How does Netflix's oci-agent workflow compare to just asking an LLM to run causal inference?","acceptedAnswer":{"@type":"Answer","text":"In a Netflix case study estimating retention impact from new entertainment types, feeding the analysis plan directly to a Claude model produced a simple linear regression estimate, but the oci-agent workflow produced an effect estimate just 25% of that baseline after its critic agent flagged early adopter bias and a failed placebo test. This shows a naive single-model approach can significantly overstate causal effects without structured checks. Teams evaluating agent-based analysis pipelines can track cases like this on daily.dev before trusting a single model's output."}},{"@type":"Question","name":"How does the oci-agent critic agent evaluate the actor agent's causal analysis output?","acceptedAnswer":{"@type":"Answer","text":"The critic reviews the notebook output produced by the actor agent and assigns one of three ratings: not_satisfactory, satisfactory_with_caveats, or fully_satisfactory, then recommends spec changes such as rerunning with tweaked parameters. It follows playbooks covering known failure scenarios like early adopter bias and failed placebo tests, combining process audits with human oversight since ground truth is unavailable. daily.dev helps developers building actor-critic agent evaluation loops keep up with real-world design patterns."}}]} ```