{"id":25620,"date":"2026-11-19T11:17:40","date_gmt":"2026-11-19T11:17:40","guid":{"rendered":"https:\/\/thisbiginfluence.com\/?p=25620"},"modified":"2026-11-19T11:17:40","modified_gmt":"2026-11-19T11:17:40","slug":"what-is-agentic-ai-in-healthcare-and-how-do-you-keep-it-safe","status":"publish","type":"post","link":"https:\/\/thisbiginfluence.com\/?p=25620","title":{"rendered":"What Is Agentic AI in Healthcare and How Do You Keep It Safe?"},"content":{"rendered":"<p> <br \/>\n<\/p>\n<div>\n<figure class=\"wp-block-image size-full is-style-rounded\"><img loading=\"lazy\" decoding=\"async\" width=\"320\" height=\"480\" src=\"https:\/\/hitconsultant.net\/wp-content\/uploads\/2026\/11\/Sanjay-Kumar-high-res-8x12_BW.jpg\" alt=\"What Is Agentic AI in Healthcare and How Do You Keep It Safe?\" class=\"wp-image-98749\" srcset=\"https:\/\/hitconsultant.net\/wp-content\/uploads\/2026\/11\/Sanjay-Kumar-high-res-8x12_BW.jpg 320w, https:\/\/hitconsultant.net\/wp-content\/uploads\/2026\/11\/Sanjay-Kumar-high-res-8x12_BW-200x300.jpg 200w, https:\/\/hitconsultant.net\/wp-content\/uploads\/2026\/11\/Sanjay-Kumar-high-res-8x12_BW-193x290.jpg 193w\" sizes=\"auto, (max-width: 320px) 100vw, 320px\"\/><figcaption class=\"wp-element-caption\"><strong>Sanjay Kumar, Chief Expertise Officer at XiFin<\/strong><\/figcaption><\/figure>\n<p class=\"wp-block-paragraph\">Twenty-eight months. That\u2019s how lengthy it took AI to go from a punching bag to a outcome that shocked mathematicians.<\/p>\n<p class=\"wp-block-paragraph\">In 2023, prime AI scientists ridiculed massive language fashions (LLMs), declaring {that a} 2-billion-parameter neural community was \u2018dumber than a $5 calculator\u2019 as a result of it couldn\u2019t reliably multiply multi-digit numbers.<\/p>\n<p class=\"wp-block-paragraph\">By July 2024, one earned a silver medal on the Worldwide Math Olympiad, although people first needed to translate the issues right into a computing language the fashions may use, and it took the fashions as much as three days to resolve some questions. By July 2025, a mannequin solved these issues utilizing plain language in the identical time restrict as the scholars and reached gold-medal stage. Then in Could 2026, a general-purpose mannequin overturned an 80-year-old open conjecture in geometry (Erd\u0151s downside 90, also called the planar unit distance downside) from a single immediate.<\/p>\n<p class=\"wp-block-paragraph\">Demonstrating this velocity of studying\u2014<em>from failing arithmetic to a frontier conjecture in 28 months<\/em>\u2014is unfathomable for many issues<em>.<\/em> And this stage and velocity of studying is already reaching the enterprise. Gartner tasks that the typical massive enterprise ran fewer than 15 AI brokers in 2025 and can run round 150,000 by 2028. On the similar time, Gartner additionally finds that solely about 13% of organizations consider they&#8217;ve the suitable agent governance in place in the present day. Functionality is transferring a lot sooner than our means to control it, and healthcare, with much less room for error than most industries, feels that hole acutely.<\/p>\n<p class=\"wp-block-paragraph\">When folks say \u201cAI,\u201d they might imply any of the next 5 layers, every constructed on the one beneath it, and all nonetheless working in manufacturing in the present day:<\/p>\n<ol class=\"wp-block-list\">\n<li><strong>Professional methods:<\/strong> people write the principles and the system applies them precisely.<\/li>\n<li><strong>Machine studying:<\/strong> the system learns patterns from historic information, similar to what rejected claims appear like, and returns a rating.<\/li>\n<li><strong>Deep studying:<\/strong> multi-layered networks that learn unstructured enter, just like the system that reads addresses on envelopes or unlocks your cellphone.<\/li>\n<li><strong>Generative fashions:<\/strong> methods like ChatGPT, Claude, and Gemini that produce new textual content, code, and pictures.<\/li>\n<li><strong>Agentic methods:<\/strong> a generative mannequin given instruments, reminiscence, and a purpose, which acts after which checks its personal work.<\/li>\n<\/ol>\n<p class=\"wp-block-paragraph\">Nothing was changed alongside the way in which. The helpful query isn&#8217;t \u201cwhich AI?\u201d however \u201cwhich layer suits this downside?\u201d<\/p>\n<h3 id=\"h-who-decides-the-path\" class=\"wp-block-heading\"><em>Who decides the trail?<\/em><\/h3>\n<p class=\"wp-block-paragraph\">Conventional software program, together with most income cycle administration (RCM) methods, is at present deterministic. Each path is drawn upfront. It&#8217;s quick, repeatable, and auditable, however when it meets a scenario no person anticipated, it stops and throws an error.<\/p>\n<p class=\"wp-block-paragraph\">Agentic AI works in another way. You give it a vacation spot, and it really works out the route. When it hits an impediment, it causes round it, and when it actually can\u2019t, it escalates to an individual with its reasoning connected. The distinction isn\u2019t what the system can do. It\u2019s who decides the trail. In regulated work, the place you need tight guardrails and little improvisation, you select how quickly it palms off to a human.<\/p>\n<figure class=\"wp-block-image size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"1079\" height=\"341\" src=\"https:\/\/hitconsultant.net\/wp-content\/uploads\/2026\/11\/image-9.png\" alt=\"\" class=\"wp-image-98750\" style=\"aspect-ratio:3.149484536082474;width:611px;height:auto\" srcset=\"https:\/\/hitconsultant.net\/wp-content\/uploads\/2026\/11\/image-9.png 1079w, https:\/\/hitconsultant.net\/wp-content\/uploads\/2026\/11\/image-9-300x95.png 300w, https:\/\/hitconsultant.net\/wp-content\/uploads\/2026\/11\/image-9-290x92.png 290w, https:\/\/hitconsultant.net\/wp-content\/uploads\/2026\/11\/image-9-766x242.png 766w\" sizes=\"auto, (max-width: 1079px) 100vw, 1079px\"\/><\/figure>\n<h3 id=\"h-how-an-ai-agent-is-like-a-trip-planner\" class=\"wp-block-heading\"><em>How an AI Agent is Like a Journey Planner\u2026.<\/em><\/h3>\n<p class=\"wp-block-paragraph\">Consider an agent the way in which you concentrate on planning a household journey. The LLM is the reasoning core, which decides the following step. By itself, although, it may well\u2019t see, act, or bear in mind, nevertheless it:<\/p>\n<ul class=\"wp-block-list\">\n<li><strong>Sees<\/strong> by way of connectors, doc parsers, and retrieval, so it may well learn the declare, the calendar, or the flight choices.<\/li>\n<li><strong>Acts<\/strong> by way of instruments, so it may well search, evaluate, and alter one thing on the planet.<\/li>\n<li><strong>Remembers<\/strong> by way of externally equipped reminiscence. Massive language fashions are stateless, so we offer short-term and long-term reminiscence ourselves.<\/li>\n<li><strong>Loops.<\/strong> Every outcome feeds again in and modifications what occurs subsequent. That return path is what turns a pipeline into an agent.<\/li>\n<\/ul>\n<h3 id=\"h-why-isn-t-one-agent-enough\" class=\"wp-block-heading\"><em>Why isn\u2019t one agent sufficient?<\/em><\/h3>\n<p class=\"wp-block-paragraph\">If a single agent dealt with a whole denied supplier declare, from eligibility verification by way of enchantment, it will hit 4 partitions.\u00a0<\/p>\n<ol class=\"wp-block-list\">\n<li><strong>Runs out of room<\/strong>: Its context window fills and the earliest details silently drop out.\u00a0<\/li>\n<li><strong>Turns into a generalist<\/strong>: It is aware of a bit about protection and a bit about coding, however not sufficient depth in both.\u00a0<\/li>\n<li><strong>Reaches all the things:<\/strong> It could possibly learn all the things it must, together with PHI, can question payors, and file appeals, with no method to scope it down.<\/li>\n<li><strong>Leaves a skinny path<\/strong>: When an enchantment comes again flawed, there aren&#8217;t any steps to examine.<\/li>\n<\/ol>\n<p class=\"wp-block-paragraph\">I\u2019m a soccer fan, so I give it some thought this fashion: you don\u2019t desire a staff the place everybody has the identical abilities. The reply isn\u2019t a much bigger agent, however a smaller scope per agent. A coding agent, an eligibility agent, and a denial administration agent every maintain one piece of the declare, whereas an orchestrator plans, routes, and re-plans. Each handoff is a contract, and each step is auditable.<\/p>\n<h3 id=\"h-security-must-change\" class=\"wp-block-heading\"><em>Safety should change<\/em><\/h3>\n<p class=\"wp-block-paragraph\">For twenty years, we secured software program by validating enter towards a schema. A malformed declare will get rejected. This method can establish when issues are flawed, however it may well\u2019t <em>instruct<\/em> on decision. New approaches demand a brand new self-discipline. Conventional controls have been constructed for methods that execute, not methods that resolve. An agent is completely different, as a result of <strong>enter turns into context<\/strong>, and a reasoning engine might be persuaded. That creates 4 failure modes. Two require an attacker:<\/p>\n<ul class=\"wp-block-list\">\n<li><strong>Immediate injection:<\/strong> hidden directions planted in a declare narrative.<\/li>\n<li><strong>Jailbreaking:<\/strong> enter framed in order that bypassing a safeguard sounds cheap.<\/li>\n<\/ul>\n<p class=\"wp-block-paragraph\">And two don\u2019t:<\/p>\n<ul class=\"wp-block-list\">\n<li><strong>Confabulation:<\/strong> an actual denial motive and an actual analysis code joined by a coverage quotation that doesn\u2019t exist.<\/li>\n<li><strong>Unfamiliar enter:<\/strong> the agent has no \u201cI can\u2019t deal with this\u201d reflex, so it makes a assured, finest guess.<\/li>\n<\/ul>\n<p class=\"wp-block-paragraph\">Brokers are greater than chatbots. They motive, act, and self-correct throughout multi-step work. Shifting from automation to orchestration is actual and taking place now.\u00a0 The constructing blocks (notion, reasoning, reminiscence, instruments, and orchestration) should all be in place. Governance is a function, not a constraint.\u00a0 Begin slim, measure all the things, and develop from there.<\/p>\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n<p class=\"wp-block-paragraph\"><strong>About Sanjay Kumar <\/strong><\/p>\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/www.linkedin.com\/in\/sanjayjkumar\/\"><strong>Sanjay Kumar<\/strong> <\/a>is <strong>CTO<\/strong> at <a href=\"https:\/\/www.xifin.com\/\">XiFin<\/a>, the place he leads the corporate\u2019s expertise imaginative and prescient, engineering technique, and innovation roadmap for its AI-powered XiFin Empower ecosystem. He focuses on scaling next-generation AI, agentic capabilities, and cloud-native infrastructure to help XiFin\u2019s continued development.<\/p>\n<\/div>\n<p><br \/>\n<br \/><a href=\"https:\/\/hitconsultant.net\/2026\/11\/19\/what-is-agentic-ai-and-how-do-you-keep-it-safe-xifin-cto\/\">Source link <\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Sanjay Kumar, Chief Expertise Officer at XiFin Twenty-eight months. That\u2019s how lengthy it took AI to go from a punching bag to a outcome that shocked mathematicians. In 2023, prime AI scientists ridiculed massive language fashions (LLMs), declaring {that a} 2-billion-parameter neural community was \u2018dumber than a $5 calculator\u2019 as a result of it couldn\u2019t [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":25622,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5],"tags":[13013,153,2197],"class_list":["post-25620","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-health","tag-agentic","tag-healthcare","tag-safe"],"_links":{"self":[{"href":"https:\/\/thisbiginfluence.com\/index.php?rest_route=\/wp\/v2\/posts\/25620","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/thisbiginfluence.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/thisbiginfluence.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/thisbiginfluence.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/thisbiginfluence.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=25620"}],"version-history":[{"count":1,"href":"https:\/\/thisbiginfluence.com\/index.php?rest_route=\/wp\/v2\/posts\/25620\/revisions"}],"predecessor-version":[{"id":25621,"href":"https:\/\/thisbiginfluence.com\/index.php?rest_route=\/wp\/v2\/posts\/25620\/revisions\/25621"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/thisbiginfluence.com\/index.php?rest_route=\/wp\/v2\/media\/25622"}],"wp:attachment":[{"href":"https:\/\/thisbiginfluence.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=25620"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/thisbiginfluence.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=25620"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/thisbiginfluence.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=25620"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}