AGENTIC OPERATING PROCEDURES # A living standard to govern every AI agent action ![Aop ](https://cdn.sanity.io/images/3xg3qj5k/production/5c6d82bb0ffa62aad6fa7012ae8a0f260013d367-817x527.svg) Static operating standards only map the happy path and go stale by the hour, leaving agents unable to handle exceptions and edge cases. Shift to dynamic, agentic operating procedures, powered by the Context Graph of Work, to give agents the most efficient path forward. The Skan AI platform builds process models on real decisions so the agentic execution layer matches the reality of work and evolves with the enterprise. [Request a demo](/request-demo) Modeled on 25B+ events. 500+ processes and 500+ enterprise apps. Agentic operating procedures are the validated execution standard for Skan AI Agents in production. ## Tracing decisions and rules to real work for agentic AI ![Icon](https://cdn.sanity.io/images/3xg3qj5k/production/e8605089e46c0bd0fdd99f3db3ab303bc9e316e6-45x46.svg) ### Skan VA captures and aggregates context The Skan Virtual Assistant, a lightweight, privacy-safe AI sensor, captures work activity across desktops and screens to aggregate into patterns and process models. ![Icon](https://cdn.sanity.io/images/3xg3qj5k/production/e8605089e46c0bd0fdd99f3db3ab303bc9e316e6-45x46.svg) ### Context Graph of Work models processes Process variants, exceptions, and workarounds are extracted and mapped on Skan AI's Context Graph of Work, providing all the context AI needs to understand how work gets done. ![Icon](https://cdn.sanity.io/images/3xg3qj5k/production/e8605089e46c0bd0fdd99f3db3ab303bc9e316e6-45x46.svg) ### AOPs keep agent execution grounded in work context Together, case-level traces, decision inputs, and normalized workflow paths become the evidence behind executable workflows. Each action traces back to work context that continuously updates. How agentic operating procedures work ## How agentic operating procedures work - [Process models](#process-models) - [Operating procedures](#operating-procedures) - [Continuous training](#continuous-training) ![Govern agent](https://cdn.sanity.io/images/3xg3qj5k/production/18459b6baab0ef22b35cb77085eb86e8211de568-889x621.svg) ## Govern agent actions with process models Skan AI's context graph captures work across end-to-end processes and builds models that include exceptions, workarounds and rework loops. These models are the core of how your agentic operating procedures ensure AI agents handle the work correctly. ![validate context derived](https://cdn.sanity.io/images/3xg3qj5k/production/df7b750f2024a1356e96b765de9194298c8b4681-889x621.svg) ## Validate context-derived procedures for execution Unlike SOPs, agentic operating procedures represent the most efficient path through a process, derived from real work and structured context. So when the context graph adapts to new patterns in real time, so does the procedures that agents execute. ![image](https://cdn.sanity.io/images/3xg3qj5k/production/c6fd9ee8695c805d89c2531cb343c1161df3fd54-889x621.svg) ## Adapt procedures and re-train AI agents in real time Unlike SOPs, agentic operating procedures change in real time as the Skan AI platform captures more work context, including the outputs your operations create — documents, records and business metrics. Throughout it all, these procedures maintain clear evidence and traceability between work context, agent actions, and outcomes. ## Built for the enterprise that can't afford to get it wrong [Learn more](https://trust.skan.ai/) ![Built for the enterprise that can't afford to get it wrong](https://cdn.sanity.io/images/3xg3qj5k/production/4120f79e8a11e683d3f14058e79d16681b2e4979-1518x1312.png?q=80&fit=max&auto=format) Continuous context capture Delivers continuous insights, allowing for dynamic adjustments and traceable outcomes. Work intelligence at enterprise scale Skan is one of the few work intelligence platforms that can be deployed to observe work at scale, spanning tens of thousands of operators at once. Advanced AI activity and process extraction Uses Computer Vision, NLP, and proprietary deep learning models for high fidelity process mapping and context extraction. Client-owned, privacy-first architecture All data, context and insights remain client-owned. Sensitive data stays within the client’s secured environment; only anonymized, abstracted metadata is sent to the Skan AI cloud. - ![AICPA SOC2 Logo](https://cdn.sanity.io/images/3xg3qj5k/production/d33f4ae0a52b897fdf2d7297bfd14653222da519-62x62.svg) ![AICPA SOC2 Logo](https://cdn.sanity.io/images/3xg3qj5k/production/d33f4ae0a52b897fdf2d7297bfd14653222da519-62x62.svg) - ![ISO Logo](https://cdn.sanity.io/images/3xg3qj5k/production/600af9c049422659a6c3a49f55c82b0944bb8f2b-62x64.svg) ![ISO Logo](https://cdn.sanity.io/images/3xg3qj5k/production/600af9c049422659a6c3a49f55c82b0944bb8f2b-62x64.svg) - ![CCPA Logo](https://cdn.sanity.io/images/3xg3qj5k/production/aa25bd8f86549aa95f78684946c3f688afb65d1e-62x62.svg) ![CCPA Logo](https://cdn.sanity.io/images/3xg3qj5k/production/aa25bd8f86549aa95f78684946c3f688afb65d1e-62x62.svg) - ![GDPR Logo](https://cdn.sanity.io/images/3xg3qj5k/production/18fc79d0270f00a7ac897fb08b524fb938b39d76-63x63.svg) ![GDPR Logo](https://cdn.sanity.io/images/3xg3qj5k/production/18fc79d0270f00a7ac897fb08b524fb938b39d76-63x63.svg) - ![HIPAA Logo](https://cdn.sanity.io/images/3xg3qj5k/production/1e702a7fedc7f26c6971ee5f24febcd90d8a0d85-70x38.svg) ![HIPAA Logo](https://cdn.sanity.io/images/3xg3qj5k/production/1e702a7fedc7f26c6971ee5f24febcd90d8a0d85-70x38.svg) - ![42001](https://cdn.sanity.io/images/3xg3qj5k/production/49954f865f070fe9f2ffa141b669943da62983c9-196x203.svg) ![42001](https://cdn.sanity.io/images/3xg3qj5k/production/49954f865f070fe9f2ffa141b669943da62983c9-196x203.svg) ## The enterprises already 
laying the foundation for agentic operations [See all stories](/case-studies) [ ![Enterprise AI platform dashboard showing process intelligence data and operational workflows](https://cdn.sanity.io/images/3xg3qj5k/production/da847716989e8a92d985d7aeb057d4cc869e8c1b-984x549.jpg?q=80&fit=max&auto=format) 14,000+ 14,000+ hours/year of automation-ready effort Before, it was turn on AI and hope. Now we know exactly which processes to automate, what we'll save, and how to prove we got there. Jennifer Redden Head of Technology ![Definiti](https://cdn.sanity.io/images/3xg3qj5k/production/4bd9a3daea270bb97b789fc4d776f8a1d415d1e1-224x28.png?q=80&fit=max&auto=format) ](/case-studies/definiti-map-the-work-then-automate-it) ## Every Skan AI product, powered by the context graph of work Blueprint [ ![Your AI masterplan](https://cdn.sanity.io/images/3xg3qj5k/production/33ba832e4a26326626c548e23e88a16d205e6896-536x464.svg) ](/blueprint) Your AI masterplan. Recommendations, roadmaps and ROI. [Learn more](/blueprint) See how work really happens Maps, metrics and opportunities, built from real work. [Learn more](/process-intelligence) See where engineering value is built. Signals, patterns and productivity across the full developer day. [Learn more](https://www.skan.ai/engineering-intelligence) Deploy agents grounded in your reality Decisions, actions and outcomes, grounded in operational context. [Learn more](/agents) Your AI masterplan. Recommendations, roadmaps and ROI. [Learn more](/blueprint) Frequently asked questions ## Frequently asked questions What is an agentic operating procedure (AOP)? An AOP is a living, continuously updated standard of how work should happen within an enterprise process. These procedures are built from real-time operator behavior and work metadata. Where a [standard operating procedure](https://www.skan.ai/blogs/how-to-build-standard-operating-procedures-sops-that-power-agentic-ai) (SOP) describes how a process should run, an AOP captures every variation, workaround, and rule operators use to make decisions while handing real cases, giving AI agents the real-world context they need to execute reliably. How is an AOP different from a traditional SOP? SOPs are static documents, written once and updated periodically or not at all. Often, SOPs don’t actually reflect the operational reality of executing processes to handle real cases. In contrast, AOPs are modeled on context derived from real work — captured through the Skan Virtual Assistant and mapped onto the Context Graph of Work — so the procedures reflect exceptions, process variants, and operator judgment calls, and update automatically as the business changes. Where does the data behind an AOP come from? AOP generation starts as soon as work execution data starts to be captured via the Skan Virtual Assistant (VA). Work activity is enriched with case ID tags through the platform’s decision discovery services and stored in Agentic Object Storage. A case reconstruction engine rebuilds anonymized journeys for each individual case to ensure traceability of agentic decisions, and a procedure generation engine combines those cases into a single, decision-aware process flow. What does an AOP actually contain? An AOP is a structured output that includes the process flow graph (nodes and edges), control-flow logic (branching, merging, synchronization), the event system (triggers, timers, errors), data context (variables and objects), responsible actors (pools/lanes), and exception handling (boundary events, escalations). It does not include tool or skill invocation logic or task-execution semantics. Those are handled separately in Skan AI Agents. How do AOPs get used once they're generated? AOPs are ingested into Skan AI Agents as the basis for building, testing, and running process automations, or viewed read-only in the process viewer. Each generated AOP is versioned (starting at v1.000) so incremental edits made on the platform canvas can be tracked as minor versions. Re-running the pipeline produces a new major version, keeping a clear, traceable link between real work and agent behavior.