{"id":558,"date":"2026-03-30T17:47:23","date_gmt":"2026-03-30T17:47:23","guid":{"rendered":"https:\/\/e40c447b5810478002.temporary.link\/ai\/?page_id=558"},"modified":"2026-03-30T19:36:49","modified_gmt":"2026-03-30T19:36:49","slug":"agentic-ellucian-colleague-ethos-api-coding","status":"publish","type":"page","link":"https:\/\/ai.elcamino.edu\/es\/initiatives\/agentic-ellucian-colleague-ethos-api-coding\/","title":{"rendered":"Codificaci\u00f3n de API Ethos de Agentic Ellucian Colleague"},"content":{"rendered":"

At El Camino College, one of the most persistent barriers to data-informed decision-making<\/strong> has not been access to data\u2014but the friction required to extract, transform, and operationalize it<\/strong>. Like many institutions running Ellucian Colleague<\/span><\/span> on-prem, we rely heavily on the Ellucian Ethos<\/span><\/span> API layer to modernize access. In theory, Ethos provides a clean, RESTful interface into Colleague. In practice, it introduces a different kind of complexity\u2014one that traditionally requires deep technical expertise, significant trial-and-error, and time that most operational teams simply do not have.<\/p>\n\n\n\n

This is the context in which I explored a different approach: using an AI coding agent to \u201cvibe code\u201d directly against the Ethos API<\/strong>\u2014not as a novelty, but as a way to fundamentally compress the distance between intent and execution.<\/p>\n\n\n\n

Lo\u00efc Audusseau – Chief Technology Officer<\/em><\/p>\n\n\n\n\n\n\n\n

The Problem: Ethos Is Powerful but Not Frictionless<\/mark><\/h3>\n\n\n\n
\n
\n

Ethos solves a real problem: it abstracts Colleague\u2019s legacy data structures into modern APIs. But several issues quickly emerge when you try to build real applications on top of it:<\/p>\n\n\n\n

1. Documentation vs. Reality Gap<\/h4>\n\n\n\n

Ethos documentation provides endpoint definitions, but it rarely captures:<\/p>\n\n\n\n

    \n
  • Real-world data relationships across modules<\/li>\n\n\n\n
  • Edge cases in financial or enrollment data<\/li>\n\n\n\n
  • Required sequencing of API calls<\/li>\n<\/ul>\n\n\n\n

    You are often left reverse-engineering behavior through experimentation.<\/p>\n\n\n\n

    2. Data Model Complexity<\/h4>\n\n\n\n

    Colleague\u2019s underlying data model is not intuitive. Even with Ethos:<\/p>\n\n\n\n

      \n
    • Financial data requires stitching together multiple endpoints<\/li>\n\n\n\n
    • Enrollment data spans sections, terms, statuses, and student records<\/li>\n\n\n\n
    • Business logic lives outside the API<\/li>\n<\/ul>\n\n\n\n

      This creates a translation problem: understanding what to query is often harder than querying it.<\/strong><\/p>\n\n\n\n

      3. Iteration Cost<\/h4>\n\n\n\n

      Traditional development cycles for reporting apps look like this:<\/p>\n\n\n\n

        \n
      • Write code<\/li>\n\n\n\n
      • Test API call<\/li>\n\n\n\n
      • Debug response<\/li>\n\n\n\n
      • Adjust logic<\/li>\n\n\n\n
      • Repeat<\/li>\n<\/ul>\n\n\n\n

        Each iteration is slow, especially when working across multiple endpoints.<\/p>\n<\/div>\n\n\n\n

        \n
        \"Ellucian<\/figure>\n<\/div>\n\n\n

        Credit: Ellucian<\/mark><\/p>\n<\/div>\n<\/div>\n\n\n\n\n\n\n\n

        The Shift: Using Claude as an Agentic Coding Layer<\/mark><\/h3>\n\n\n\n

        Instead of treating coding as a manual process, I used Claude<\/span><\/span> as an agentic layer on top of Ethos<\/strong>.<\/p>\n\n\n\n

        The goal was not just code generation\u2014it was delegating the reasoning process<\/strong>:<\/p>\n\n\n\n

          \n
        • Interpreting API documentation<\/li>\n\n\n\n
        • Structuring queries<\/li>\n\n\n\n
        • Handling transformations<\/li>\n\n\n\n
        • Iterating rapidly based on feedback<\/li>\n<\/ul>\n\n\n\n

          In effect, Claude became a Colleague-aware coding agent<\/strong>, capable of:<\/p>\n\n\n\n

            \n
          • Generating working API calls<\/li>\n\n\n\n
          • Debugging responses<\/li>\n\n\n\n
          • Refactoring logic in real time<\/li>\n\n\n\n
          • Building full applications from prompts<\/li>\n<\/ul>\n\n\n\n

            This is what I mean by vibe coding<\/em>: describing the outcome and letting the agent handle the implementation details.<\/p>\n\n\n\n\n\n\n\n

            Application 1: Financial Reporting App<\/mark><\/h3>\n\n\n\n

            The first test was a financial reporting application\u2014something traditionally handled through static reports or complex SQL extracts.<\/p>\n\n\n\n

            \n
            \n

            Objective<\/h4>\n\n\n\n

            Create a dynamic reporting interface that:<\/p>\n\n\n\n

              \n
            • Pulls real-time financial data from Colleague via Ethos<\/li>\n\n\n\n
            • Aggregates by fund, org, and account<\/li>\n\n\n\n
            • Calculates budget vs. actual vs. encumbrances<\/li>\n\n\n\n
            • Presents clean, decision-ready outputs<\/li>\n<\/ul>\n\n\n\n

              What the Agent Did<\/h4>\n\n\n\n

              Claude:<\/p>\n\n\n\n

                \n
              • Identified relevant Ethos endpoints for financial data<\/li>\n\n\n\n
              • Structured multi-step API calls<\/li>\n\n\n\n
              • Built transformation logic to calculate totals and variances<\/li>\n\n\n\n
              • Generated a working front-end structure<\/li>\n<\/ul>\n<\/div>\n\n\n\n
                \n
                <\/div>\n\n\n
                \n
                \"Ethos<\/figure>\n<\/div><\/div>\n<\/div>\n\n\n\n

                Outcome<\/h4>\n\n\n\n

                What would typically take days (or weeks) of iterative development was reduced to hours of guided prompting and refinement<\/strong>.<\/p>\n\n\n\n

                More importantly, the app was not static\u2014it was iteratively co-developed in real time<\/strong>.<\/p>\n\n\n\n\n\n\n\n

                Application 2: Enrollment Dashboard<\/mark><\/h3>\n\n\n\n

                The second application focused on enrollment\u2014another area where data fragmentation is common.<\/p>\n\n\n\n

                \n
                \n

                Objective<\/h4>\n\n\n\n

                Build a dashboard that:<\/p>\n\n\n\n

                  \n
                • Tracks enrollment trends by term<\/li>\n\n\n\n
                • Breaks down headcount, FTES, and section fill rates<\/li>\n\n\n\n
                • Allows filtering by program or demographic dimensions<\/li>\n<\/ul>\n\n\n\n

                  Challenges<\/h4>\n\n\n\n

                  Enrollment data is inherently relational:<\/p>\n\n\n\n

                    \n
                  • Students \u2194 Sections \u2194 Terms<\/li>\n\n\n\n
                  • Status codes and census rules<\/li>\n\n\n\n
                  • Timing dependencies<\/li>\n\n\n\n
                  • FTES calculation logic<\/li>\n<\/ul>\n<\/div>\n\n\n\n
                    \n
                    <\/div>\n\n\n
                    \n
                    \"Enrollment<\/figure>\n<\/div><\/div>\n<\/div>\n\n\n\n

                    What the Agent Did<\/h4>\n\n\n\n

                    Claude:<\/p>\n\n\n\n

                      \n
                    • Mapped relationships across Ethos endpoints<\/li>\n\n\n\n
                    • Proposed data joins and aggregation strategies<\/li>\n\n\n\n
                    • Generated logic for metrics like fill rate and enrollment trends<\/li>\n\n\n\n
                    • Iterated quickly based on output validation<\/li>\n<\/ul>\n\n\n\n

                      Outcome<\/h4>\n\n\n\n

                      The dashboard emerged through conversation, not traditional development cycles<\/strong>.<\/p>\n\n\n\n\n\n\n\n

                      Where It Broke (and What That Taught Me)<\/mark><\/h3>\n\n\n\n

                      This process was not seamless. Several challenges surfaced that are important for any institution considering this approach:<\/p>\n\n\n\n

                      1. Context Drift<\/h4>\n\n\n\n

                      The agent can lose context across complex workflows.
                      Lesson:<\/strong> You must anchor the agent with structured prompts and consistent definitions.<\/p>\n\n\n\n

                      2. Hallucinated Endpoints<\/h4>\n\n\n\n

                      At times, Claude would generate plausible\u2014but non-existent\u2014API endpoints.
                      Lesson:<\/strong> Always validate against actual Ethos documentation.<\/p>\n\n\n\n

                      3. Data Misinterpretation<\/h4>\n\n\n\n

                      Without domain context, the agent may misinterpret:<\/p>\n\n\n\n

                        \n
                      • Financial structures<\/li>\n\n\n\n
                      • Enrollment rules<\/li>\n<\/ul>\n\n\n\n

                        Lesson:<\/strong> Human-in-the-loop is non-negotiable.<\/p>\n\n\n\n

                        4. Over-Generation<\/h4>\n\n\n\n

                        The agent tends to over-engineer solutions.
                        Lesson:<\/strong> Constrain scope aggressively and iterate incrementally.<\/p>\n\n\n\n\n\n\n\n

                        The Bigger Insight: This Is Not About Coding<\/mark><\/h3>\n\n\n\n

                        What emerged from this process is not just a faster way to build apps\u2014it\u2019s a fundamentally different model:<\/p>\n\n\n\n

                        We are moving from writing code to directing systems.<\/strong><\/p>\n\n\n\n

                        Instead of:<\/p>\n\n\n\n

                          \n
                        • \u201cHow do I write this query?\u201d<\/li>\n<\/ul>\n\n\n\n

                          The question becomes:<\/p>\n\n\n\n

                            \n
                          • \u201cWhat outcome do I want, and how do I guide the agent to get there?\u201d<\/li>\n<\/ul>\n\n\n\n

                            This aligns directly with broader trends in AI enablement:<\/p>\n\n\n\n

                              \n
                            • Institutions that embed AI into operational workflows will move faster<\/li>\n\n\n\n
                            • Those that treat it as an add-on will remain constrained<\/li>\n<\/ul>\n\n\n\n

                              As highlighted in the broader AI landscape, there is a clear gap between capability and readiness\u2014students and employees alike need structured pathways to build these skills<\/strong> .<\/p>\n\n\n\n\n\n\n\n

                              Governance and Risk Considerations<\/mark><\/h3>\n\n\n\n

                              This approach also intersects directly with AI governance.<\/p>\n\n\n\n

                              Agentic development touches multiple risk domains:<\/p>\n\n\n\n

                                \n
                              • Data security (API access, sensitive data)<\/li>\n\n\n\n
                              • Vendor and tool governance<\/li>\n\n\n\n
                              • Algorithmic decision-making<\/li>\n<\/ul>\n\n\n\n

                                As outlined in the governance framework, application development and AI-enabled systems must be governed through standards around APIs, testing, and oversight<\/strong> .<\/p>\n\n\n\n

                                This reinforces a key point:<\/p>\n\n\n\n

                                \n

                                Agentic coding is not just a productivity tool\u2014it is a governance concern.<\/em><\/p>\n<\/blockquote>\n\n\n\n\n\n\n\n

                                Lessons Learned<\/mark><\/h3>\n\n\n\n
                                  \n
                                1. Start with a Clear Use Case<\/strong>
                                  Financial reporting and enrollment were ideal because they are high-value and data-rich.<\/li>\n\n\n\n
                                2. Treat the AI as a Junior Engineer (at First)<\/strong>
                                  Guide it, validate it, refine it.<\/li>\n\n\n\n
                                3. Prompt Engineering Becomes Architecture<\/strong>
                                  The quality of outputs is directly tied to how you structure inputs.<\/li>\n\n\n\n
                                4. Iteration Speed Is the Real Breakthrough<\/strong>
                                  The biggest gain is not code generation\u2014it\u2019s the ability to iterate instantly.<\/li>\n\n\n\n
                                5. Human Oversight Remains Critical<\/strong>
                                  Especially in institutional systems with real-world consequences.<\/li>\n<\/ol>\n\n\n\n\n\n\n\n

                                  Where This Leads<\/mark><\/h3>\n\n\n\n

                                  This experiment points to a near-term future where:<\/p>\n\n\n\n

                                    \n
                                  • ERP systems like Colleague are no longer bottlenecks<\/li>\n\n\n\n
                                  • API layers like Ethos become interfaces for AI agents, not just developers<\/strong><\/li>\n\n\n\n
                                  • Institutional teams can build their own tools without traditional development pipelines<\/li>\n<\/ul>\n\n\n\n

                                    In other words:<\/p>\n\n\n\n

                                    The constraint is no longer technical capability\u2014it\u2019s organizational readiness.<\/strong><\/p>\n\n\n\n

                                    And that is precisely where institutions will differentiate.<\/p>","protected":false},"excerpt":{"rendered":"

                                    At El Camino College, one of the most persistent barriers to data-informed decision-making has not been access to data\u2014but the friction required to extract, transform, and operationalize it. Like many institutions running Ellucian Colleague on-prem, we rely heavily on the Ellucian Ethos API layer to modernize access. In theory, Ethos provides a clean, RESTful interface […]<\/p>\n","protected":false},"author":1,"featured_media":561,"parent":78,"menu_order":0,"comment_status":"closed","ping_status":"open","template":"","meta":{"_acf_changed":false,"footnotes":""},"class_list":["post-558","page","type-page","status-publish","has-post-thumbnail","hentry"],"acf":[],"_links":{"self":[{"href":"https:\/\/ai.elcamino.edu\/es\/wp-json\/wp\/v2\/pages\/558","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ai.elcamino.edu\/es\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/ai.elcamino.edu\/es\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/ai.elcamino.edu\/es\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ai.elcamino.edu\/es\/wp-json\/wp\/v2\/comments?post=558"}],"version-history":[{"count":35,"href":"https:\/\/ai.elcamino.edu\/es\/wp-json\/wp\/v2\/pages\/558\/revisions"}],"predecessor-version":[{"id":620,"href":"https:\/\/ai.elcamino.edu\/es\/wp-json\/wp\/v2\/pages\/558\/revisions\/620"}],"up":[{"embeddable":true,"href":"https:\/\/ai.elcamino.edu\/es\/wp-json\/wp\/v2\/pages\/78"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ai.elcamino.edu\/es\/wp-json\/wp\/v2\/media\/561"}],"wp:attachment":[{"href":"https:\/\/ai.elcamino.edu\/es\/wp-json\/wp\/v2\/media?parent=558"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}