[{"data":1,"prerenderedAt":152},["ShallowReactive",2],{"blog-ai-plugin-development":3,"blog-related-\u002Fblog\u002Fai-plugin-development":120},{"id":4,"title":5,"author":6,"authorRole":7,"body":8,"cover":102,"coverAlt":103,"date":104,"description":105,"draft":106,"extension":107,"meta":108,"navigation":109,"path":110,"readingTime":111,"seo":112,"stem":113,"tags":114,"updated":118,"__hash__":119},"blog\u002Fblog\u002Fai-plugin-development.md","Building custom AI plugins for CRMs, e-commerce and SaaS platforms","William Ali","Founder & AI Specialist",{"type":9,"value":10,"toc":94},"minimark",[11,15,20,43,47,80,84,87,91],[12,13,14],"p",{},"The AI feature people actually use is the one that appears inside the tool they already have open. Plugins, whether for HubSpot, Salesforce, Shopify, Zendesk, Slack or your own platform, are how AI stops being a separate destination.",[16,17,19],"h2",{"id":18},"three-patterns","Three patterns",[21,22,23,31,37],"ol",{},[24,25,26,30],"li",{},[27,28,29],"strong",{},"Enrich:"," when a record is created, fill in what's missing (company data, classification, summary).",[24,32,33,36],{},[27,34,35],{},"Draft:"," propose the next email, note, or reply for a human to send.",[24,38,39,42],{},[27,40,41],{},"Act:"," update fields, create tasks, move stages, with approval where the action is hard to undo.",[16,44,46],{"id":45},"the-engineering-that-keeps-it-safe","The engineering that keeps it safe",[48,49,50,56,62,68,74],"ul",{},[24,51,52,55],{},[27,53,54],{},"Scoped credentials:"," the plugin gets exactly the permissions it needs.",[24,57,58,61],{},[27,59,60],{},"Idempotency:"," the same event processed twice must not create two records.",[24,63,64,67],{},[27,65,66],{},"Approval gates:"," anything that touches money, deal stage or customer communication waits for a click.",[24,69,70,73],{},[27,71,72],{},"Audit log:"," who\u002Fwhat changed which field, when, based on what input.",[24,75,76,79],{},[27,77,78],{},"Dry-run mode:"," run in shadow for a week and compare with what humans did.",[16,81,83],{"id":82},"a-worked-example","A worked example",[12,85,86],{},"For a distributor we built a CRM agent that enriches inbound leads, classifies intent, drafts a follow-up and proposes updates. Low-risk fields apply automatically; deal-stage changes wait for the rep. Reps saved ten-plus hours a week and missing CRM fields dropped by around 90%.",[16,88,90],{"id":89},"start-small-expand-on-evidence","Start small, expand on evidence",[12,92,93],{},"Enrich first, then draft, then act. Each step earns the trust that justifies the next.",{"title":95,"searchDepth":96,"depth":96,"links":97},"",2,[98,99,100,101],{"id":18,"depth":96,"text":19},{"id":45,"depth":96,"text":46},{"id":82,"depth":96,"text":83},{"id":89,"depth":96,"text":90},"\u002Fcovers\u002Fblog\u002Fai-plugin-development.svg","Line-art illustration of a plug with a trailing cord connecting into a socket panel, sparks where the prongs meet.","2025-04-15","How to add AI where your team already works, inside the tools they use, with agents that respect approvals and audit trails.",false,"md",{},true,"\u002Fblog\u002Fai-plugin-development",6,{"title":5,"description":105},"blog\u002Fai-plugin-development",[115,116,117],"AI Agents","Automation","Integrations",null,"7n0gbFwFFrqyaayJTCt53GNZZ6harMn8Z8-8FdCCAuc",[121,132,142],{"path":122,"title":123,"description":124,"date":125,"tags":126,"cover":129,"coverAlt":130,"readingTime":131},"\u002Fblog\u002Fmeasuring-ai-roi-honestly","Measuring AI ROI honestly: a method that survives the finance review","How to measure the return on an AI project without fooling yourself: baselines, full costs, holdout attribution, leading indicators and inflated-ROI red flags.","2026-08-11",[127,128],"Strategy","Analytics","\u002Fcovers\u002Fblog\u002Fmeasuring-ai-roi-honestly.svg","Line-art illustration of a ruler measuring a stack of coins beside a rising bar.",7,{"path":133,"title":134,"description":135,"date":136,"tags":137,"cover":140,"coverAlt":141,"readingTime":131},"\u002Fblog\u002Fmultimodal-ai-business-workflows","Multimodal AI in real business workflows: where it earns its place","Vision-language and audio models now read forms, photos, screenshots and calls directly. Where multimodal AI beats OCR-then-LLM, with examples, limits and cost.","2026-06-30",[138,139,116],"Computer Vision","LLMs","\u002Fcovers\u002Fblog\u002Fmultimodal-ai-business-workflows.svg","Line-art illustration of an eye, an ear and a speech bubble converging into a single processing node.",{"path":143,"title":144,"description":145,"date":146,"tags":147,"cover":150,"coverAlt":151,"readingTime":131},"\u002Fblog\u002Fmlops-basics-first-model","MLOps basics for teams shipping their first model","The minimum MLOps a team needs to put a first model into production safely: versioning, reproducible training, a registry, serving, monitoring and rollback.","2026-05-19",[148,149],"MLOps","Product","\u002Fcovers\u002Fblog\u002Fmlops-basics-first-model.svg","Line-art illustration of a looping pipeline with gauges, from data to model to monitoring and back.",1792973686188]