Helios Core Unveils Self-Healing AI to Bypass ServiceNow Tech Debt

AI Disruption in IT Service Management, the Future of IT Ops in an AI Agent World With Helios Core — Ai4

Ai4

Automating basic help desk tickets is now table stakes, forcing enterprise IT to deploy autonomous agents that generate remediation tools on the fly to resolve outages before engineers wake up.

AI-assisted summary of AI Disruption in IT Service Management, the Future of IT Ops in an AI Agent World With Helios Core — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published on YouTube by the Ai4 channel on August 15, 2026, this presentation features Scott McIsaac, CEO of Helios Core AI. According to the session description, the talk examines the shift from reactive AI service desks that merely deflect tickets to proactive AI operations agents capable of monitoring environments, correlating signals across enterprise stacks, and remediating issues autonomously.

## TL;DR Drawing on over two decades of enterprise IT and MSP experience, Scott McIsaac introduces Helios Core as an autonomous operations platform designed to eliminate the chaos of middle-of-the-night production outages. The system ingests signals from monitoring platforms like Datadog and PagerDuty, cross-references cloud dependencies across AWS and Azure, and autonomously triages systemic issues before alert fatigue sets in. To ensure safe remediation, the AI agent draws from an approved tool catalog or generates new scripts on the fly for human approval, escalating unresolved failures through context-rich voice calls rather than blind alerts. The platform also enables conversational natural language runbooks for complex operational workflows and supports over 150 enterprise integrations, including legacy ITSM suites like ServiceNow and Freshservice.

## Atmosphere & style The presentation carries a grounded, practitioner-led tone that resonates immediately with experienced IT professionals familiar with 2:00 AM outage panics and legacy technical debt. McIsaac speaks with candid humor and seasoned authority, acknowledging the noise in the AI startup ecosystem while drawing credibility from his decades scaling managed services and leading business units at NTT. The pacing is concise and direct, creating a pragmatic product debut that feels less like speculative tech hype and more like an overdue architectural overhaul for enterprise operations.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=4s) Enterprise IT operations remain burdened by fragmented automation tools and isolated playbooks that lack unified cross-stack context. - [01:25](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=85s) Incumbent ITSM platforms accumulate substantial technical debt by bolting AI onto legacy architectures rather than designing for native autonomy. - [01:25](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=85s) Basic Tier 1 automation like password resets is now table stakes; true operational value lies in autonomous resolution of complex, off-hours production outages. - [01:25](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=85s) Deploying the agent directly inside enterprise communication channels like Slack and Microsoft Teams enables it to act as an autonomous digital teammate. - [03:42](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=222s) Aggregating signals from Datadog, PagerDuty, tickets, and voice calls allows the platform to correlate alerts against full AWS, Azure, and CMDB dependency maps to eliminate alert fatigue. - [03:42](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=222s) Strict governance prevents rogue execution by restricting autonomous remediation to a curated and pre-vetted tool catalog. - [03:42](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=222s) When a necessary remediation tool is missing, the AI generates the script on the fly, requests human engineer approval, and saves it permanently to the catalog for future reuse. - [03:42](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=222s) For incidents requiring human intervention, the agent places outbound voice calls with diagnostic findings and root-cause context rather than sending blind alert notifications. - [06:00](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=360s) Operational teams can author and execute complex multi-system workflows, such as employee offboarding, using natural language runbooks. - [06:00](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=360s) Alongside out-of-the-box support for platforms like ServiceNow and Freshservice, the agent can autonomously code custom integrations on the fly for unsupported legacy systems.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=4s) The Roots of Helios Core and the Legacy ITSM Challenge

Scott McIsaac introduces himself as the CEO of Helios Core, acknowledging with a bit of humor that they are "an AI startup like everyone else in this building," currently about a year and a half into their journey. The problem space—disruption in IT service management (ITSM)—hits remarkably close to home. McIsaac co-founded a managed service provider (MSP) back in 2005, operated it for over a decade, and sold the company to NTT in 2018, subsequently running several business units inside the global telecommunications giant. When leaving NTT, the driving motivation was to address the persistent structural inefficiency that has plagued enterprise IT operations for decades.

Enterprise IT service management remains deeply tethered to legacy processes. While the market is flooded with tools promising automation—from custom scripting to Ansible playbooks—the ecosystem has historically lacked any real coherence. McIsaac reflects that had this modern AI capability existed fifteen years ago when they were scaling their MSP, the organization would have operated far leaner, with radically lower dependency on headcounts and significantly higher operational velocity across the board. The mission behind Helios Core was sparked by a desire to bring genuine, end-to-end intelligence to an industry still stranded in reactive workflows.

- **Founding timeline:** MSP founded in 2005, acquired by NTT in 2018; Helios Core launched circa early 2025 (~1.5 years prior to the talk). - **Core issue with existing tooling:** Disparate tools and isolated playbooks (e.g., Ansible) lack unified context and coherence.

---

[01:25](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=85s) Beyond Chatbots and the 2:00 AM Outage Reality

McIsaac poses an all-too-familiar question to the audience: how many people have received a panic call at 2:00 AM completely blind, lacking any context about why a system crashed? Looking out at a room full of raised hands, he notes that despite minor evolutionary improvements, this chaotic scenario remains the baseline reality across the industry today. Incumbent ITSM platforms attempt to bolt on intelligence—pointing to ServiceNow acquiring and layering on AI tooling, or Freshservice introducing automated assistants. However, these solutions force enterprise customers to wrestle with massive technical debt before seeing real value.

Helios Core aims to bypass that legacy overhead by deploying an autonomous AI agent designed to function like a "digital human"—a colleague sitting directly in enterprise communication channels like Microsoft Teams or Slack to actively resolve incidents. In discussions with platform leaders at enterprise vendors like ServiceNow, McIsaac notes that legacy providers are actively losing ground to nimble startups because customers are exhausted by tech debt. Helios Core only needs access to tickets and operational tools to execute resolution. Critically, McIsaac clarifies that their vision goes far beyond commoditized Tier 1 tasks: basic password resets or adding users to directory groups are simple "table stakes" that everyone can automate. The real differentiator lies in handling high-stakes, middle-of-the-night production application failures without waking an engineer into chaos.

```mermaid flowchart TD accTitle: Incident Processing and Autonomous Remediation Flow accDescription: Workflow showing signals ingesting into the Helios Core agent, dependency correlation, governed tool execution, and contextual escalation.

subgraph Ingestion ["1. Signal Ingestion"] A["Events & Signals<br/>(Tickets, Teams/Slack, Calls, Datadog, PagerDuty)"] end

subgraph Analysis ["2. Correlation & Context"] B["Cross-Cloud Dependency Mapping<br/>(AWS, Azure, CMDB Integration)"] C["Alert Correlation & Triage"] end

subgraph Execution ["3. Governed Resolution"] D{"Existing Tool<br/>in Catalog?"} E["Execute Controlled Command"] F["Generate Tool on the Fly<br/>& Prompt Human for Approval"] G["Add Validated Tool<br/>to Agent Catalog"] end

subgraph Outcome ["4. Resolution & Escalation"] H["Autonomous Self-Healing"] I["Context-Rich Outbound Call<br/>to On-Call Engineer"] end

A --> B B --> C C --> D D -- Yes --> E D -- No --> F F --> G G --> E E --> H C -. Unresolvable.-> I ```

---

[03:42](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=222s) Autonomous Triage, Self-Healing Tool Generation, and Governance

Explaining the platform mechanics, McIsaac walks through how Helios Core processes an operational event. The system ingests triggers across diverse entry points—monitoring alerts from Datadog, PagerDuty alarms, help desk tickets, Microsoft Teams chats, or even voice calls. The platform immediately maps the event against the enterprise's broader cloud footprint across Azure and AWS, drawing on full dependency mappings and deep CMDB integrations to understand connected systems. This solves chronic alert fatigue: if an alarm has fired multiple times throughout the day without human notice, the agent correlates these disparate signals to diagnose systemic risk before an outage unfolds.

Crucially, Helios Core enforces strict governance so the agent does not run uninhibited commands across production infrastructure. When an event translates into remediation steps, the agent draws from a pre-vetted catalog of tools. If the necessary remediation tool does not exist, the agent drafts the tool on the fly, presents it to a human engineer with an explicit approval request ("Is this good for me to run?"), and upon approval, checks that tool into its persistent catalog for future autonomous execution. This dynamic tool-building loop underpins true self-healing IT. If human intervention is mandatory—such as a complex failure at 2:00 AM—the agent will actually place an outbound phone call to the engineer, delivering full diagnostic context and root-cause findings rather than a blind alert.

- **Multi-channel ingestion:** PagerDuty, Datadog, support tickets, voice calls, and enterprise collaboration channels. - **Environment visibility:** Automated dependency mapping across AWS, Azure, and underlying CMDB records. - **Dynamic tool catalog:** Agents generate new remediation scripts on the fly, submit them for human sign-off, and store them permanently for automated reuse. - **Contextual escalation:** Replaces blank pagers with an AI-driven voice call detailing exact triage progress and root-cause analysis.

---

[06:00](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=360s) Natural Language Runbooks and Ecosystem Integrations

Beyond reactive troubleshooting, the platform orchestrates proactive operational workflows using natural language runbooks. McIsaac highlights a practical corporate scenario: an HR department submits a request to offboard an employee at 5:00 PM. Rather than requiring manual human handoffs across identity and licensing systems, the natural language runbook automatically triggers at the appointed time: it locates the reporting manager, reroutes the departing employee's email, disables the user account, and revokes application licensing. Operational staff build, review, and approve these complex runbooks simply by speaking or typing instructions into the interface.

McIsaac presents this talk as Helios Core's formal debut. In addition to offering an end-to-end ITSM product, the platform is engineered to layer on top of an organization's existing software stack. Helios Core integrates with ServiceNow, Freshservice, Zendesk, and Zoho, alongside over 150 out-of-the-box integrations. If a company uses a bespoke or unsupported system, the AI agent can autonomously code and implement the necessary integration on the fly. McIsaac wraps up the session by directing interested attendees to their conference booth, checking in with his team offstage to confirm their location at booth K519, before thanking the room and opening the floor to questions.

- **Runbook authoring:** Zero-code workflow creation driven entirely through conversational natural language prompts. - **Supported ITSM suites:** Out-of-the-box support for ServiceNow, Freshservice, Zendesk, and Zoho. - **Integration breadth:** Over 150 enterprise connectors supported natively, with on-the-fly agentic generation for custom platforms. - **Event location:** Conference booth K519.

## Notable quotes > Every AI service desk closes tickets faster, resets passwords, provisions apps. That is not a differentiator anymore, it is the current expectation. — Scott McIsaac

> The real question is who is running your operation when no one opened a ticket. — Scott McIsaac

> This session is about the shift past the reactive AI service desk to an AI operations agent that watches your environment, correlates signals across your stack, triages what matters, and remediates before it becomes someone's problem. — Scott McIsaac

> Drawing on two decades of running enterprise IT operations, we will separate what is genuinely autonomous from the chatbots wearing an agent's clothes, and show what it takes to put proactive AI into production. — Scott McIsaac

> If you own IT service delivery, this is the ten minutes that moves your thinking from deflecting tickets to running operations. — Scott McIsaac

> [00:04](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=4s) an AI startup like everyone else in this building — Scott McIsaac

> [01:25](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=85s) digital human — Scott McIsaac

> [01:25](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=85s) table stakes — Scott McIsaac

> [03:42](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=222s) Is this good for me to run? — Scott McIsaac

## Mentioned Scott McIsaac, Helios Core, Ai4, NTT, Ansible, ServiceNow, Freshservice, Microsoft Teams, Slack, Datadog, PagerDuty, Azure, AWS, Zendesk, Zoho

## Video description Scott McIsaac - CEO, Helios Core AI

Every AI service desk closes tickets faster, resets passwords, provisions apps. That is not a differentiator anymore, it is the current expectation. The real question is who is running your operation when no one opened a ticket. This session is about the shift past the reactive AI service desk to an AI operations agent that watches your environment, correlates signals across your stack, triages what matters, and remediates before it becomes someone's problem. Drawing on two decades of running enterprise IT operations, we will separate what is genuinely autonomous from the chatbots wearing an agent's clothes, and show what it takes to put proactive AI into production. If you own IT service delivery, this is the ten minutes that moves your thinking from deflecting tickets to running operations.

[Source](https://www.youtube.com/watch?v=2i_W6NyZFIc)

中文简报

Helios Core推出自愈式AI以规避ServiceNow技术债 自动化基础服务台工单现已成为基本门槛，迫使企业IT部署自主智能体，实时生成修复工具以便在工程师醒来前解决停机故障。 凭借二十多年的企业IT和MSP经验，Scott McIsaac推出了自主运维平台Helios Core，旨在消除深夜生产环境停机造成的混乱。该系统摄取来自Datadog和PagerDuty等监控平台的信号，交叉比对AWS与Azure之间的云依赖关系，并在警报疲劳产生前自主对系统性问题进行分诊。为确保安全修复，AI智能体可调用经批准的工具目录，或实时生成新脚本供人工审批，并通过包含丰富上下文的语音电话而非盲目警报来升级上报未解决的故障。此外，该平台还针对复杂的运维工作流支持对话式自然语言运维手册，并提供超过150种企业级集成，涵盖ServiceNow和Freshservice等传统ITSM套件。 Helios Core 旨在颠覆传统 ITSM 模式，通过直接部署在 Slack 和 Microsoft Teams 等协作渠道中的自主 AI Agent，解决传统工具由于架构遗留问题和孤立脚本所导致的技术债。 (1:25) 密码重置等基础 Tier 1 自动化已成为行业标配，该平台的核心价值在于处理高风险的非工作时间生产故障，实现真正自主的端到端事件处置。 (2:05) 平台整合来自 Datadog、PagerDuty、工单和语音电话的信号，并结合 AWS、Azure 及 CMDB 依赖关系图进行告警关联分析，从根源上消除告警疲劳。 (3:42) 系统在严格治理下运行，优先调用预审工具库；若缺少修复脚本，Agent 可实时生成代码并提示人工审批，随后永久保存至工具库以备自主复用。 (4:20) 支持通过自然语言构建运维 Runbook 并与 ServiceNow、Freshservice 等主流系统集成；对于必须人工介入的故障，Agent 会外呼提供根因上下文的语音电话，而非发送盲目通知。 (6:00)

Reading summary

Scott McIsaac, CEO of Helios Core, contends that legacy IT service management platforms have trapped enterprises in fragmented tooling and technical debt, arguing that the future of IT operations relies on autonomous AI agents capable of deep dependency triage and governed production self-healing. Shortcomings of Incumbent ITSM and Chatbots (1:25) McIsaac argues that legacy automation tools like isolated Ansible playbooks and bolted-on AI assistants from incumbents like ServiceNow or Freshservice fail to resolve operational chaos. Enterprise teams still endure blind middle-of-the-night outages because incumbent solutions force customers to navigate substantial technical debt before realizing real value. While commoditized Tier 1 automations such as password resets and directory updates are now table stakes, McIsaac asserts that conventional chatbots remain incapable of resolving high-stakes production application crashes. Cross-Cloud Triage and Governed Tool Generation (3:42) To counter alert fatigue, the platform ingests events across monitoring systems like Datadog and PagerDuty, support tickets, and chat channels, correlating disparate alarms against cloud environments in AWS and Azure alongside CMDB dependency records. To execute remediation safely, the agent utilizes a strict governance loop: it relies on a vetted tool catalog, or dynamically drafts new remediation scripts on the fly. The agent requests explicit human approval before running new tools and permanently checks approved scripts into its catalog for future autonomous reuse. If an incident cannot be resolved autonomously, the agent places an outbound voice call to on-call engineers detailing full root-cause findings rather than sending an uninformative alert. Conversational Runbooks and Dynamic Ecosystem Integration (6:00) Beyond reactive troubleshooting, the platform orchestrates scheduled operational workflows through natural language runbooks created without code. For example, an administrator can author a prompt to execute an employee offboarding at a designated time, which automatically identifies reporting managers, reroutes email, disables accounts, and revokes software licenses. The platform is designed to layer over existing stacks, offering native connectivity to over 150 tools—including ServiceNow, Freshservice, Zendesk, and Zoho—while deploying its agent to code bespoke integrations on the fly when encountering unsupported platforms. Helios Core positions autonomous agents as a remedy for long-standing structural inefficiencies in enterprise IT operations, though the system maintains operational boundaries by requiring human review and approval before adopting newly generated remediation scripts.

精炼阅读摘要

Helios Core首席执行官Scott McIsaac指出，传统企业IT服务管理（ITSM）长期受困于分散工具与累积的技术债务，难以应对突发的生产故障；他主张通过部署具备上下文感知能力与严格安全治理的自主AI代理，实现多源警报关联、动态工具自愈以及自然语言驱动的运维流程编排。 传统IT运维的技术债务与被动响应痛点 (0:04) McIsaac结合自身在2005年创立托管服务提供商（MSP）并于2018年将其出售给NTT的从业经历指出，企业IT运营在过去数十年间始终存在结构性低效。尽管市场上充斥着Ansible剧本与定制脚本等自动化方案，但由于普遍缺乏全局上下文和协同机制，现有运维流程依然高度碎裂。面对行业痛点，现有主流ITSM平台如ServiceNow和Freshservice主要依赖收购或外挂AI组件进行修补，但在McIsaac看来，这种方式迫使企业客户必须先行消化庞大的历史技术债务，难以快速获得实际价值。 当前运维领域最具代表性的挑战是凌晨两点的突发系统宕机，值班工程师常在毫无故障上下文的情况下被盲目唤醒。McIsaac强调，诸如密码重置或员工权限配置等基础的一线支持（Tier 1）任务已完全商品化，属于任何自动化产品均能实现的标配功能。该领域真正的分水岭与核心价值，在于能否在深夜核心生产应用发生严重故障时，无需将工程师拉入混乱状态即可实现安全可靠的自主排查与修复。 跨云依赖映射与多源警报智能关联分析 (3:42) 为了解决传统运维长期存在的警报疲劳问题，Helios Core构建了一套覆盖多种异构入口的信号摄取系统。该架构能够统一接入来自Datadog的监控指标、PagerDuty报警通知、工单系统记录、Microsoft Teams或Slack即时通讯消息，乃至突发的语音呼叫信号。在捕获事件输入后，系统会立即对照企业在AWS与Azure跨云环境中的基础设施足迹，深度调用配置管理数据库（CMDB）建立系统依赖映射。 McIsaac解释称，大规模生产事故在全面爆发前，往往伴随着白天多次被忽视或被忽略的边缘报警。通过将这些看似孤立的告警信号与底层复杂的跨云架构拓扑进行实时关联，AI代理能够先于灾难性停机识别出系统性风险并完成分诊归因，从而改变传统运维被动扑火的工作模式，在基础设施故障扩散前精确锁定根因。 受控自愈机制与动态运维工具生成闭环 (3:42) 在推动具体故障的自主修复过程中，平台设计了严格的执行治理边界，以防范自主代理在生产环境中运行未经审查的高危指令。当事件转化为具体的排错动作时，AI代理会优先从预先审核的工具目录中选取经过验证的脚本。若所需工具不存在，代理可在运行时动态编写对应的修复工具，并主动向人类工程师发起确认请求（明确询问“该指令是否允许运行”）；一旦获得人工审批授权，该工具便会被持久化登记至工具目录中，供后续故障处理自主复用。 这一动态编写并审核工具的机制构成了系统实现闭环自愈的基础。同时，针对必须由人类工程师介入的深层次复杂故障，系统摒弃了仅发送机械告警通知的传统方式，转由AI代理主动拨打语音电话联系值班人员，直接口述详尽的诊断上下文、当前排查进度与根本原因分析，确保工程师在接入处置时拥有完整的情报支持。 自然语言工作流编排与异构系统集成 (6:00) 在被动排障之外，平台通过引入自然语言运行手册（Runbook）来实现主动运维与业务流程编排。McIsaac列举了企业常见的人力资源离职交接场景：当人事部门提交在下午5点离职某位员工的工单后，自然语言运行手册可在设定时间自动触发，依次完成查找直属主管、重定向离职员工电子邮箱、禁用身份目录账号以及撤销应用软件许可等跨部门操作。运维人员无需编写底层代码，仅通过日常对话或输入文本提示即可完成流程的创建、审查与上线。 在生态兼容性方面，该平台既支持作为端到端的独立ITSM产品运行，也可直接作为智能化外挂层嵌入企业现有技术栈。系统原生支持与ServiceNow、Freshservice、Zendesk和Zoho等平台的对接，并内置超过150种企业级连接器。McIsaac宣称，若客户使用非标准或小众的自研系统，AI代理还能够在运行时自主编写代码实现对应的API集成，降低接入的技术门槛。 McIsaac主张企业IT运维正在由分散的脚本和繁琐的工单流转，迈向由自主AI代理主导的协作体系。然而，在复杂且高敏感的企业生产环境中，动态生成代码并执行生产级操作仍需面对安全管控和治理信任的实际检验；该模式能否顺利渗透并替代传统ITSM巨头建立的既有工作流，将取决于企业对自主执行权限的开放意愿与合规边界。

Detailed notes — zh

## Background 该内容由 Ai4 频道于 2026 年 8 月 15 日在 YouTube 平台发布，记录了 Helios Core AI 首席执行官 Scott McIsaac 的主题演讲。视频简介指出，单纯快速关闭工单和重置密码已成为基础标配，本次分享探讨了 IT 运维如何从被动式 AI 服务台迈向主动式 AI 运维智能体，实现无需人工提单即可跨技术栈关联信号并自主排障修复。 ## TL;DR Helios Core 首席执行官 Scott McIsaac 结合自身多年运维实战经验，指出传统 ITSM 平台在原有老旧架构上叠加 AI 存在严重技术债务。他主张运维应超越 L1 级基础工单偏转，引入如数字同事般的 AI 智能体，直接负责高复杂度故障的端到端根因排查。该平台整合工单、监控与云拓扑依赖图，在安全受控的环境下支持动态即时生成工具与人在回路审批。最后，他展示了通过自然语言构建自动化 Runbook 以及无缝集成 150 余种主流工具的生态能力，推动运维向事前预防与主动自愈演进。 ## Atmosphere & style 演讲整体风格务实、敏锐且直击痛点，开场以“凌晨两点接到紧急排障电话”的经典场景迅速引发全场运维从业者的强烈共鸣。Scott 凭借深厚的 MSP 实战积淀，用接地气而自信的叙事剖析大厂方案的局限，条理清晰地拆解智能体落地架构。整场分享节奏紧凑、干货密集，既有对行业沉疴的犀利反思，又充满了以技术创新重塑 IT 运维掌控力的高效与务实感。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=4s) 传统 ITSM 工具长期处于各自为政的碎片化状态，即便依赖脚本和 Playbook 依然无法摆脱对高负荷人力的依赖。 - [01:25](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=85s) 运维痛点不仅在于故障频发，更在于凌晨报警时缺乏上下文；下一代 AI 运维应作为常驻协作平台的“数字人”主动接管排障。 - [01:50](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=110s) 老牌 ITSM 平台在原有陈旧架构上“打补丁”式叠加 AI 会背负沉重技术债务，难以达成敏捷的原生自主排障。 - [02:15](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=135s) 密码重置、工单偏转等 L1 基础自动化已是行业门槛标配，AI 智能体的核心竞争壁垒在于应对核心系统崩溃时的端到端根因定位与修复。 - [03:42](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=222s) 打通工单、聊天、Datadog 监控及 PagerDuty 告警等多源数据入口，为全局故障洞察提供跨技术栈的统一上下文。 - [04:10](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=250s) 结合多云架构的依赖拓扑图（Dependency Map）进行动态告警关联，能够精准捕获数日内被人工遗漏的潜在系统隐患。 - [04:40](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=280s) 智能体深度联动 CMDB 将决策转化为系统指令，并在严密的权限边界与安全治理下运行，杜绝失控风险。 - [05:10](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=310s) 具备运行中即时（on the fly）生成新工具的弹性能力，经人工审核授权后沉淀入资产目录，实现系统能力的自主进化。 - [05:30](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=330s) 在保持人在回路（Human-in-the-loop）的前提下，智能体遇到疑难问题可直接致电工程师并同步排查结论，极大缩短恢复时间（MTTR）。 - [06:00](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=360s) 支持通过自然语言快速定义并审批端到端 Runbook，实现跨系统离职回收等复杂周期性 IT 流程的完全自动化。 - [06:30](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=390s) 兼顾独立部署与生态覆盖能力，可无缝叠加在 ServiceNow、Zendesk 等现有体系之上，支持约 150 个预置集成与动态接口拓展。 - [07:00](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=420s) 未来 IT 运维的核心逻辑在于为 AI 智能体注入全栈上下文并赋予执行工具，推动运维模式从“被动关闭工单”彻底转向“主动自愈运营”。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=4s) 传统 ITSM 的局限与 Helios Core 的创立背景 Helios Core 首席执行官 Scott McIsaac 开场介绍了公司的背景与渊源。作为一家成立约一年半的 AI 初创公司，团队在 IT 服务管理（ITSM）领域拥有深厚的实战积淀。早在 2005 年，他们就创办了一家托管服务提供商（MSP），并一路运营至 2018 年被 NTT 收购，随后在 NTT 内部继续管理多项业务。离开 NTT 后，团队立志彻底改变 IT 服务管理行业。 Scott 指出，IT 服务管理行业长期以来极其依赖传统老旧模式。市场上充斥着各种宣称能实现自动化的工具，企业也可以编写 Ansible 等各种 Playbook 来提升效率，但这些工具之间往往支离破碎、缺乏协同体系。如果十五年前构建 MSP 业务时就拥有现在的技术，企业本可以变得更加精简高效、降低对人力的依赖。因此，团队致力于探索如何利用 AI 在该领域掀起真正的变革。 ```mermaid flowchart TD accTitle: 从传统被动运维到 AI 智能体自主运维演进图 subgraph Legacy["传统运维模式 (Legacy Ops)"] A["多源告警/工单碎片化"] --> B["凌晨 2:00 唤醒工程师"] B --> C["无上下文排查与手工执行"] end subgraph Autonomous["Helios Core AI 智能体体系"] D["全渠道接入 (工单/监控/Chat)"] --> E["依赖拓扑映射 & 告警关联分析"] E --> F["受控登录系统 & 根因分析"] F --> G{"工具库是否覆盖?"} G -- 是 --> H["执行自愈 Runbook"] G -- 否 --> I["即时动态生成工具 -> 人工审核"] I --> H H --> J["故障自愈并沉淀经验"] end ``` ### [01:25](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=85s) 超越“打补丁式”AI 与基础 L1 自动化 在探讨现状时，Scott 向全场发问：“在座有多少人曾在凌晨两点接到过运维电话，而且完全不知道发生了什么？”几乎全场都深有共鸣。他直言，虽然行业在逐步改善，但目前依然缺乏能够整合所有上下文并实现深度自动化与高效运维的工具。Helios Core 的目标是引入一个像“数字人”（Digital Human）一样的 AI 智能体——它如同并肩作战的同事，常驻在 Teams 或 Slack 等协作工具中主动解决问题。 针对市场上已有的方案，Scott 进行了尖锐的对比。无论是 ServiceNow、Freshservice 等老牌 ITSM 平台，还是其收购整合的 Moveworks 等产品，大多只是在原有庞杂架构上“打补丁”（bolting on top），企业若想提升效率就必须背负沉重的技术债务。相比之下，敏捷的初创方案无需理会这些包袱，只要接入工单与系统工具就能推进故障解决。Scott 强调，真正的价值绝非停留在重置密码、分配用户组等早已成为“标配入场券”（table stakes）的 L1 级基础服务，而在于当凌晨两点应用全线崩溃时，智能体能够接管事件、在严格管控下登录系统、完成分类排障，并带着根因分析主动向运维人员提供解决方案。 - **传统巨头的痛点**：在老旧 ITSM 架构上叠加 AI 往往受制于庞大的技术债务，导致交付沉重且迟缓。 - **定位差异**：不仅做基础层面的工单偏转（Ticket Deflection），更聚焦于高复杂度的关键故障端到端排障。 ### [03:42](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=222s) 动态事件关联、系统排障与治理闭环 Scott 详细拆解了 Helios Core 平台的运作机制。平台支持多源统一接入，包括 ITSM 工单、Teams 聊天、语音电话、Datadog 监控以及 PagerDuty 告警等。一旦事件触发，AI 智能体会全面审视整个架构（包括 Azure 和 AWS 的云上部署环境），结合完整的依赖拓扑图（Dependency Map）进行告警关联分析，识别出那些在过去几天内反复出现但被人工忽略的潜在风险。 在执行层面，智能体深度集成配置管理数据库（CMDB），明确涉及的后端系统与组件，并将其转化为具体的执行命令。为了防止智能体在生产环境中失控乱跑，系统建立了极其严格的安全治理闭环： - **受控执行**：智能体只允许调用经过授权的工具集。 - **即时工具生成**：遇到工具库未覆盖的新场景时，智能体会在运行中（on the fly）自动生成所需工具，并提示运维人员：“这个工具适合让我运行吗？”经人工审核后，该工具便沉淀入系统目录，以便下次自动调用。 - **自主升级与人在回路**：随着时间推移，IT 服务形成自愈体系；若遇到无法解决的疑难杂症，智能体甚至会在凌晨主动给值班工程师拨打电话，直接说明完整上下文和排查结论，大幅缩减人工介入后的恢复时间。 ### [06:00](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=360s) 自然语言 Runbook 构建与生态集成 系统的另一大核心是基于自然语言构建可执行的 Runbook（自动化运维流程手册）。Scott 以日常运维为例：若 HR 提交工单要求在下午 5:00 离职某位员工，到了指定时间，智能体便会自动执行一整套流程——查找直属主管、将离职人员邮件转发给主管、禁用该账号并收回软件许可。管理员只需在交互界面中使用自然语言沟通，即可快速生成并审批这些 Runbook，直接扩充智能体的工具库。 在生态兼容性方面，Helios Core 既能提供原生的 ITSM 能力，也可以作为增强层无缝覆盖在现有的 ServiceNow、Freshservice、Zendesk、Zoho 等主流工具之上。目前平台已预置约 150 个第三方集成点，若遇到缺失的接口，智能体同样支持即时自动构建集成。演讲最后，Scott 总结道，智能体的核心在于“如何为它注入更多上下文并赋予更多工具”，这也是将故障消灭在宕机发生之前的关键所在。随后他简要介绍了展台位置（K519 展位）并向与会观众致谢收尾。 ## Notable quotes > Every AI service desk closes tickets faster, resets passwords, provisions apps. That is not a differentiator anymore, it is the current expectation. — Scott McIsaac > The real question is who is running your operation when no one opened a ticket. — Scott McIsaac > This is the ten minutes that moves your thinking from deflecting tickets to running operations. — Scott McIsaac > [00:04](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=4s) 如果十五年前构建 MSP 业务时就拥有现在的技术，企业本可以变得更加精简高效、降低对人力的依赖。 — Scott McIsaac > [01:25](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=85s) 在座有多少人曾在凌晨两点接到过运维电话，而且完全不知道发生了什么？ — Scott McIsaac > [01:25](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=85s) 重置密码、分配用户组等早已成为“标配入场券”。 — Scott McIsaac > [01:25](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=85s) 在老旧 ITSM 架构上叠加 AI 往往受制于庞大的技术债务，导致交付沉重且迟缓。 — Scott McIsaac > [03:42](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=222s) 这个工具适合让我运行吗？ — Scott McIsaac > [06:00](https://www.youtube.com/watch?v=2i_W6NyZFIc&t=360s) 如何为它注入更多上下文并赋予更多工具，是将故障消灭在宕机发生之前的关键所在。 — Scott McIsaac ## Mentioned Scott McIsaac, Helios Core, Helios Core AI, NTT, Ansible, ServiceNow, Freshservice, Moveworks, Microsoft Teams, Slack, Datadog, PagerDuty, Azure, AWS, CMDB, Zendesk, Zoho ## Video description Scott McIsaac - CEO, Helios Core AI Every AI service desk closes tickets faster, resets passwords, provisions apps. That is not a differentiator anymore, it is the current expectation. The real question is who is running your operation when no one opened a ticket. This session is about the shift past the reactive AI service desk to an AI operations agent that watches your environment, correlates signals across your stack, triages what matters, and remediates before it becomes someone's problem. Drawing on two decades of running enterprise IT operations, we will separate what is genuinely autonomous from the chatbots wearing an agent's clothes, and show what it takes to put proactive AI into production. If you own IT service delivery, this is the ten minutes that moves your thinking from deflecting tickets to running operations. [Source](https://www.youtube.com/watch?v=2i_W6NyZFIc)

Source quotes

Scott McIsaac: Every AI service desk closes tickets faster, resets passwords, provisions apps. That is not a differentiator anymore, it is the current expectation. Scott McIsaac: The real question is who is running your operation when no one opened a ticket. Scott McIsaac: This is the ten minutes that moves your thinking from deflecting tickets to running operations. [0:04] Scott McIsaac: 如果十五年前构建 MSP 业务时就拥有现在的技术，企业本可以变得更加精简高效、降低对人力的依赖。 [1:25] Scott McIsaac: 在座有多少人曾在凌晨两点接到过运维电话，而且完全不知道发生了什么？ [1:25] Scott McIsaac: 重置密码、分配用户组等早已成为“标配入场券”。 [1:25] Scott McIsaac: 在老旧 ITSM 架构上叠加 AI 往往受制于庞大的技术债务，导致交付沉重且迟缓。 [3:42] Scott McIsaac: 这个工具适合让我运行吗？ [6:00] Scott McIsaac: 如何为它注入更多上下文并赋予更多工具，是将故障消灭在宕机发生之前的关键所在。

Source quotes

Scott McIsaac: Every AI service desk closes tickets faster, resets passwords, provisions apps. That is not a differentiator anymore, it is the current expectation. Scott McIsaac: The real question is who is running your operation when no one opened a ticket. Scott McIsaac: This session is about the shift past the reactive AI service desk to an AI operations agent that watches your environment, correlates signals across your stack, triages what matters, and remediates before it becomes someone's problem. Scott McIsaac: Drawing on two decades of running enterprise IT operations, we will separate what is genuinely autonomous from the chatbots wearing an agent's clothes, and show what it takes to put proactive AI into production. Scott McIsaac: If you own IT service delivery, this is the ten minutes that moves your thinking from deflecting tickets to running operations. [0:04] Scott McIsaac: an AI startup like everyone else in this building [1:25] Scott McIsaac: digital human [1:25] Scott McIsaac: table stakes [3:42] Scott McIsaac: Is this good for me to run?

Original transcript

[0:04] My name's Scott. Um, I'm the CEO of Helios Core.  We're a an AI startup like everyone else in this [0:09] building. Um, but we've been in business about a  year and a half now. Um, the topic today around [0:15] AI disruption in IT service management is pretty  close to our backgrounds. Um, so we started an MSP [0:23] back in like 2005. We operated that business up  until 2018 and we actually sold it to NTT in uh [0:31] in 2018 and then operated different businesses  within that within that corporation. So, [0:36] when we left NTT, we wanted to really figure out  how do we how do we drive change in this industry? [0:42] Um, the service management industry from an IT  perspective is very legacy. Yes, there's tons [0:47] of tools out there that claim automation.  There's tons of tools um, that, you know, [0:52] you can bring into your business that make you  more efficient and you can build playbooks and [0:56] Ansible and all kinds of different things that,  you know, to automate that kind of industry. But, [1:02] none of them really were coherent. Um, we  found a big gap. If we would have had this [1:08] technology 15 years ago when we were building  this business, we would have been, you know, [1:12] a much leaner organization, less people dependent,  um, much more efficient across the board. So, [1:18] we sought out to figure out how do we drive  change in this industry with the use of AI. [1:25] Um, how many people in this room have been called  at 2:00 in the morning and with no context of [1:31] what's going on? Pretty much everybody, right?  And that's that's the state of where it still is [1:38] today. It's getting better, but there's still  no tools that are pulling everything together [1:44] to really drive more automation, more context, um,  you know, driving better service delivery, more [1:51] efficient operations, all those things. So, what  what we're trying to look at is how do we bring in [1:57] an AI agent into your business that operates  like a digital human? Like how do you [2:03] How can we bring in something that actually  looks and feels and operates like another [2:09] colleague of yours? Sits next to you in Teams,  in Slack, whatever tool you might be using there, [2:14] and just solves problems. The problem that is  out there today is there's tons of AI agents that [2:21] claim they do this. ServiceNow has Moveworks, they  acquired on top of that. Freshservice has one. [2:27] All of these different ITSM platforms are doing  it, but they're bolting on top of that. So, [2:32] if you if you really want to drive efficiency,  you have to deal with all your tech debt. [2:37] Um and that becomes very very complicated. We've  talked to uh a lot of different leaders actually [2:43] in ServiceNow that claim they're losing a ton of  business to smaller startups like us because they [2:48] can be more efficient. They don't care about  tech debt. All we need access to is tickets. [2:52] All we need access to is tools to actually drive  in a drive resolution. And I'm not talking about [2:57] just level one service resolution. I'm not talking  about password resets and adding users to group. [3:02] That stuff is table stakes, everyone does  that, it's very very simple. I'm talking about [3:06] that event happens at 2:00 in the morning  um and your application is completely down. [3:12] What happens? In the old world, you get called.  An engineer gets called with no context. [3:17] What we're driving to is the ability to have an  AI agent that will take that event that came in, [3:23] triage it, log into all the systems governed and  controlled. We don't just let it go uncontrolled [3:29] into the environment, and actually come back with  a root cause of the problem with the idea of oh, [3:36] can I solve that problem for you?  Right? So, how do we do that? [3:42] Um we've built a platform around this. Uh and the  way the platform works today is it takes events, [3:48] it takes different uh entry points. It  can be tickets, it can be Teams chats, [3:52] it can be phone calls, it can be um your data  dog integration, it could be a pager duty alarm, [3:59] whatever it is. It just takes a an event comes  in. The first thing it does is it it'll look [4:03] at that event across your entire landscape, it  understands um what you have deployed in Azure, [4:08] what you have deployed in AWS. It has a full  dependency map around that environment, and it [4:13] starts to triage that event. So, alert correlation  is a big deal. You know, how many times do you [4:19] get an alarm that's happened, you know, five  other times throughout the day, but then no one [4:23] looks at. No one knows. No one understands  that there's a problem in the environment. [4:27] Well, the agent can see that now. We give it  that context. Um we identify the problem, we [4:32] correlate that event. Then, we try to figure out  the next step. The agent looks at it, understands [4:38] what tools are involved or what systems are  involved in the back end through your CMDB. [4:42] It's fully integrated into that, so it understands  all of the different components. That turns into [4:48] um an actual bunch of commands that it needs  to run. So, it logs into these systems. Um [4:54] what we've done though is we've built governance  around this, too. So, um [snorts] think about [4:59] we don't want the agent just to execute commands  in your environment. The agent has a set of [5:04] tools that we've given it, or if it doesn't have  a tool, it creates that tool on the fly, sends [5:10] that to the to the the human to actually look at  the tool and say, "Is this good for me to run?" [5:15] That goes into its catalog, and then next time  it has the problem, it can execute against that. [5:20] So, we're building self-healing IT services around  this agent. So, it's not just an agent that can, [5:27] you know, solve basic problems, but it can  actually build its tools. It can get smarter. [5:31] It can inject into the process all of the, you  know, the different context and issues that have [5:35] happened. Hey, this problem's happened four times  over the last 3 days. Um here's my resolution [5:40] how to fix it. Are you good with me to fix it?  Cool, yes. If it needs a human, it reaches out, [5:46] makes a phone call. So, if it is 2:00 in the  morning and it can't solve the problem, the [5:49] agent can reach out and actually talk to the human  with full context, tell them what's going on, [5:54] then the human can grab it, solve the problem very  much more efficiently than, you know, in the past. [6:00] Uh we're doing this through runbooks, is what  we're calling them, but at the end of the day, [6:05] it's a set of tooling and procedures that the  agent crafts on its own. So, think about you if [6:12] you want to let's go for a simple one. Let's run  the the monitoring and triage side, but let's say [6:17] HR puts in a in a a ticket that they need to take  the user and we need to de-decon them at 5:00. So, [6:24] we're we're getting rid of someone. Um at 5:00,  that execution happens. It will go through, [6:29] find the manager, route its email um his email  to the manager, and actually disable that user, [6:36] remove the licensing, all those things. This is  all done through natural language. So, we have a [6:40] an interface that you go to, talk to, it creates  these runbooks. These runbooks get approved, [6:44] they get executed into um the agent's tool set. Um  so, it's it's always about how do you bring more [6:52] context into the agent? How do you give them more  tools? Otherwise, these agents aren't very good, [6:57] right? And so, that applies to pretty much all  industries, but really in the IT industry as well. [7:03] Um the auto triage agent is kind of what  we've been talking about a little bit here. [7:07] Um the idea is how do you take that event and turn  it in from uh you know, a potential outage that's, [7:13] you know, could take the entire application down  to actually solving the problem before it actually [7:18] goes down. So, the agent can do all  this as part of it autonomously. [7:22] Um we've built tools around this. Uh we're really,  really excited. Um this is kind of our coming out, [7:28] I would say, from a from a platform perspective.  We've we've done a few things around this. We [7:32] have an ITSM product as well. We integrate with  all the existing tools that are out there today, [7:37] ServiceNow, Freshservice, Zendesk, Zoho. Um and  then we layer it on our ITSM on top of that. So, [7:43] you know, kind of from from that perspective we  have, you know, all those integration points. Uh, [7:48] we bought 150 other integration points and if we  don't have it, we can create those integrations on [7:52] the fly using the agent as well. So, there's a a  lot of things that we've built into this platform. [7:58] Uh, we have a booth here. Um, if anyone's  interested in talking more about this, [8:02] uh, we're we're happy to.  We're in Len what booth is it? [8:07] K519. Sorry, I should know that, but I  don't. Um, but other than that, you know, [8:11] I appreciate everyone's time today. A- Any  questions in the audience? Anything we can answer? [8:19] No? Awesome. Well, thank you. Thank you so much.