Peraton CTO Palmer Says AI Agents Dismantle Enterprise IT Gatekeeping

Living Platform: How AI is Transforming the Information Technology Model With Peraton — Ai4

Ai4

Natural language interfaces and autonomous agents eliminate the intermediary developer layer by enabling domain experts to build and adapt enterprise workflows in real time.

AI-assisted summary of Living Platform: How AI is Transforming the Information Technology Model With Peraton — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published on YouTube by the Ai4 channel on August 14, 2026, this presentation features Christine Palmer, Chief Technology Officer for the Federal Civilian Sector at Peraton. The session explores how frontier AI models and natural language interfaces are eliminating legacy software bottlenecks to transform enterprise information technology from a static assembly of tools into a dynamic 'living platform' that adapts alongside its users in real time.

## TL;DR Tracing the historical arc of knowledge distribution from the printing press and early mainframes to modern interactive AI, the talk illustrates how enterprise IT is experiencing a democratization shift akin to consumer crowdsourcing models like Uber and Airbnb. Natural language now acts as the primary interface, allowing domain specialists to execute ideas directly without translating requirements through traditional development queues. Realizing this living platform requires replacing static ticketing and hardcoded systems with three core building blocks: autonomous agents, context engineering, and embedded safety guardrails. Ultimately, modernizing the IT control plane with semantic networks and contextual databases demands overcoming cultural inertia and adopting an interactive, co-creative mindset with machines.

## Atmosphere & style The presentation carries a polished, authoritative, and forward-looking keynote tone driven by deep technical and strategic leadership experience. Grounding complex infrastructure concepts with historical parables and architectural frameworks, the talk moves with clarity and deliberate momentum. It feels thought-provoking and urgent, challenging enterprise technologists to let go of legacy procedural delays and step into an era of real-time, iterative human-AI collaboration.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=sxrg06XQWFI&t=4s) Historical resistance to automation, such as the Ottoman scribes delaying the adoption of the printing press, demonstrates the significant risk of falling behind during major technological transitions. - [00:04](https://www.youtube.com/watch?v=sxrg06XQWFI&t=4s) Autonomous AI agents are compressing centuries of computing and communication evolution into rapid, real-time loops of contextual decision-making. - [01:58](https://www.youtube.com/watch?v=sxrg06XQWFI&t=118s) Enterprise computing has transitioned from isolated 1960s mainframes to mobile and cloud platforms, culminating in interactive AI that turns one-way computer instructions into two-way collaborative intelligence. - [04:03](https://www.youtube.com/watch?v=sxrg06XQWFI&t=243s) Enterprise IT is undergoing a democratization shift analogous to consumer crowdsourcing disruptions seen in retail, transit, and media. - [04:03](https://www.youtube.com/watch?v=sxrg06XQWFI&t=243s) Removing technical intermediaries enables domain experts to directly create and iterate on applications in real time, preventing the loss of institutional nuance. - [07:12](https://www.youtube.com/watch?v=sxrg06XQWFI&t=432s) Natural language serves as the universal user interface, allowing users to express pure domain intent that enterprise orchestration layers translate into functional technology. - [07:12](https://www.youtube.com/watch?v=sxrg06XQWFI&t=432s) Traditional IT mechanisms are being replaced by three core architectural pillars: autonomous agents as units of work, context engineering, and automated guardrails. - [07:12](https://www.youtube.com/watch?v=sxrg06XQWFI&t=432s) AI safety currently faces an implementation gap between high-level governance policies and low-level, automated guardrails embedded directly into the IT stack. - [10:11](https://www.youtube.com/watch?v=sxrg06XQWFI&t=611s) The modern IT control plane is moving toward semantic networks that dynamically discover optimal execution paths based on user intent rather than static routing rules. - [10:11](https://www.youtube.com/watch?v=sxrg06XQWFI&t=611s) Enterprise observability is advancing from passive telemetry dashboards to active, closed-loop agentic control planes capable of immediate automated remediation. - [12:44](https://www.youtube.com/watch?v=sxrg06XQWFI&t=764s) The most formidable barrier to enterprise and federal AI adoption is cultural inertia and legacy ticketing workflows rather than technical limitations. - [12:44](https://www.youtube.com/watch?v=sxrg06XQWFI&t=764s) Information technology is evolving from rigid, top-down infrastructure into an adaptive 'living platform' that grows continuously alongside its users.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=sxrg06XQWFI&t=4s) The Historical Arc of Communication and the Rise of Living Platforms

Information technology is undergoing a fundamental transformation into what can be described as a "living platform." The way artificial intelligence—and autonomous agents in particular—are experiencing and interacting with the world mirrors the historical progression of human knowledge sharing. The earliest method for broadcasting information across geographic and national borders was the physical book. A compelling historical example illustrates the risk of hesitation: while the movable-type printing press was created in Europe, the Ottoman Empire had early opportunities to adopt similar printing innovations. They chose not to automate because of an entrenched guild of scribes who manually copied texts to preserve existing social and employment structures. Consequently, they fell behind, while printed books reshaped global communication.

That historical trajectory has continuously repeated across the history of computing. Society progressed from the earliest computing mainframes to local enterprise networks, then to the global internet, and eventually to real-time social platforms. Now, software agents are tracing that same evolutionary path, but with unprecedented velocity—ingesting vast streams of information, contextualizing data, and making autonomous decisions far faster than human operators ever could.

- **Historical Parallel**: Reluctance to embrace automation (as seen with Ottoman scribes and the printing press) historically leads to falling behind technologically. - **The Agentic Acceleration**: AI agents are compressing centuries of human communication and computing evolution into rapid, real-time loops of contextual decision-making.

---

[01:58](https://www.youtube.com/watch?v=sxrg06XQWFI&t=118s) The Evolution of Enterprise Compute: From Mainframes to Interactive AI

The modern information technology paradigm began in earnest during the 1960s with the introduction of the first enterprise mainframes. These systems were massive, costly, and strictly confined to large institutions, government organizations, and elite scientific research facilities that possessed both the physical space and the budget to run heavy compute workloads. Information access was static and isolated. Over subsequent decades, computing steadily migrated closer to the end user:

``` +-------------------------------------------------------------------+ | 1960s Mainframes (Isolated, expensive, research institutions) | | | | | 1980s-90s Desktops & PCs (Office access, local workgroup sharing)| | | | | 2000s Internet & Web (Global distribution of information) | | | | | 2010s Mobile & Cloud (Compute in pocket, massive remote datasets)| | | | | Present: Interactive AI (Two-way intelligence & living platforms)| +-------------------------------------------------------------------+ ```

The transition from desktop personal computers to the internet expanded distribution across wider audiences, while mobile devices moved processing power directly into people's pockets. The advent of cloud computing eliminated the requirement for localized, on-premise data centers to process enterprise-scale datasets, providing the underlying compute engine that now powers large language models (LLMs). The defining shift with generative AI is that interaction is no longer one-directional—users no longer simply write code or feed rigid instructions into a machine; systems now actively analyze inputs, respond with real-time insights, and suggest optimized ways to communicate and build.

---

[04:03](https://www.youtube.com/watch?v=sxrg06XQWFI&t=243s) The Crowdsourcing Paradigm Shift in Enterprise IT

Across multiple major consumer industries, centralized and gatekept business models have been dismantled through crowdsourcing and democratization: - **Retail**: Amazon enabled individual sellers to reach global markets without traditional brick-and-mortar storefronts. - **Transportation**: Uber bypassed the medallion taxi system, allowing anyone with a vehicle to provide transit. - **Hospitality**: Airbnb transformed private residences into decentralized accommodations. - **Media**: YouTube turned everyday consumers into global content creators.

Enterprise information technology is approaching the exact same inflection point. Historically, enterprise IT has operated as a closed, top-down model: an employee conceives an idea, submits a formal IT ticket or software request, and waits for a specialized development team to construct a bespoke application. In past roles as a data scientist, building custom dashboards for non-technical colleagues was standard practice because stakeholders lacked the technical means to manipulate raw data. Today, that intermediary layer is dissolving. Domain experts who possess the nuanced operational knowledge can directly query, shape, and visualize information on demand—and modify those views dynamically from one day to the next without technical intermediaries.

- **Direct Execution**: Ideas can be realized and run on the very same day they are conceived by the individuals who understand the operational problem best. - **Preserving Context**: Building directly in natural language avoids the loss of institutional nuance that typically occurs when business requirements are translated across multiple technical teams.

---

[07:12](https://www.youtube.com/watch?v=sxrg06XQWFI&t=432s) Architecture and the Three New Building Blocks: Agents, Context, and Guardrails

The foundation enabling this democratization is that natural language has become the primary user interface. Eliminating the barrier of having to learn a programming language allows users to express pure intent, which the underlying system then translates into functional technology. However, realizing this across an enterprise requires more than an isolated natural language query box; it demands a unified orchestration layer that binds disparate enterprise tools together, much like network protocols and the web unified isolated mainframe computers in prior decades.

``` +------------------------------------------------------------------+ | Natural Language Interface | | (Express intent in plain domain-specific words) | +---------------------------------+--------------------------------+ | +---------------------------------v--------------------------------+ | Enterprise Orchestration Layer | | (Connects tools, systems, pipelines, and workflows) | +---------------------------------+--------------------------------+ | +---------------------------------v--------------------------------+ | Data & Context Engineering Layer | | (Provides institutional knowledge, business logic & memory) | +---------------------------------+--------------------------------+ | +---------------------------------v--------------------------------+ | Execution & Safety Layer | | [ Autonomous Agents ] [ Enforced Guardrails ] | +------------------------------------------------------------------+ ```

To make this architecture operational, three core building blocks must replace traditional IT mechanisms:

1. **Agents as the Unit of Work**: Autonomous agents are superseding static tickets, brittle scripts, and scheduled cron jobs, dynamically managing tasks across connected systems. 2. **Context Engineering**: Instead of engineering a rigid, static application that executes only hardcoded commands, context engineering supplies systems with rich background knowledge so they can evaluate scenarios and take the most effective action within safe operational parameters. 3. **Guardrails and Safety Implementation**: Similar to how cybersecurity lagged significantly behind the initial rollout of the internet, modern AI safety currently suffers from an implementation gap. While high-level governance policies exist, automated, low-level technical guardrails embedded directly into the IT stack are still nascent and must be built out rapidly.

---

[10:11](https://www.youtube.com/watch?v=sxrg06XQWFI&t=611s) Reimagining the IT Control Plane: Semantic Networks and Contextual Databases

The traditional IT control plane is being restructured across four primary architectural layers:

- **Semantic Networks**: Networks are moving away from hardcoded routing paths toward intent-based execution. Rather than requiring human engineers to anticipate and configure hundreds of possible workflow combinations, semantic networks utilize agents to interpret human intent and automatically discover the optimal path forward. - **Contextual Databases**: While human working memory can typically hold only three to five concepts simultaneously, AI models can maintain significantly larger working context windows. Combining human domain direction with an agentic AI's ability to aggregate disparate context transforms the database from a static lookup table into an intelligent, contextual retrieval engine. - **Dynamic Identity and Authorization**: Identity management has historically been a static compliance check. In an agent-driven ecosystem, it must evolve into a real-time framework capable of tracing authority, delegation, and granular actions exchanged between humans and autonomous agents across complex IT interfaces. - **Actionable Observability**: Enterprise observability is advancing past passive telemetry and monitoring dashboards ("seeing what is happening on the network"). Integrated with agents, observability becomes a closed-loop control plane that detects an anomaly, understands the surrounding context, and immediately executes corrective remediation.

---

[12:44](https://www.youtube.com/watch?v=sxrg06XQWFI&t=764s) Overcoming the Cultural Mindset and Building the Living Platform

For decades, technology was engineered through a one-directional paradigm: humans strictly defined hardware constraints, software specifications, and communication channels. Over a decade of deploying AI capabilities to technical teams and federal customers demonstrates that the most formidable barrier to adoption is cultural rather than technical. The legacy expectation—submitting a specification to an external team and passively waiting for a finished deliverable—is deeply ingrained.

Working alongside agentic AI requires an interactive, iterative mindset. Rather than issuing a one-off command and receiving a single static output, users enter a continuous creative loop with the machine, discovering insights, learning from the system, and refining their objectives in real time. Because AI capabilities are now being embedded across every layer of the infrastructure stack, IT is shifting away from being a rigid system that must be expensively engineered from the top down. It is becoming an open, living platform where every person across an organization can directly imagine, build, and adapt technology at the speed of their ideas.

## Notable quotes > Every generation of computing has asked the same thing of us. Adapt yourself to the tool, and someone else will decide what the tool can do. — Christine Palmer

> Frontier models have turned language into an interface. Anyone who can describe an idea can now put it in motion, and the systems built on top can reason across whatever scenario is handed to them, without demanding a new ontology or data model each time. — Christine Palmer

> It stops being something we assemble and becomes something that grows in real time, with the people using it. — Christine Palmer

> [01:58](https://www.youtube.com/watch?v=sxrg06XQWFI&t=118s) The defining shift with generative AI is that interaction is no longer one-directional—users no longer simply write code or feed rigid instructions into a machine; systems now actively analyze inputs, respond with real-time insights, and suggest optimized ways to communicate and build. — Christine Palmer

> [04:03](https://www.youtube.com/watch?v=sxrg06XQWFI&t=243s) Building directly in natural language avoids the loss of institutional nuance that typically occurs when business requirements are translated across multiple technical teams. — Christine Palmer

> [07:12](https://www.youtube.com/watch?v=sxrg06XQWFI&t=432s) Eliminating the barrier of having to learn a programming language allows users to express pure intent, which the underlying system then translates into functional technology. — Christine Palmer

> [07:12](https://www.youtube.com/watch?v=sxrg06XQWFI&t=432s) Instead of engineering a rigid, static application that executes only hardcoded commands, context engineering supplies systems with rich background knowledge so they can evaluate scenarios and take the most effective action within safe operational parameters. — Christine Palmer

> [10:11](https://www.youtube.com/watch?v=sxrg06XQWFI&t=611s) Rather than requiring human engineers to anticipate and configure hundreds of possible workflow combinations, semantic networks utilize agents to interpret human intent and automatically discover the optimal path forward. — Christine Palmer

> [12:44](https://www.youtube.com/watch?v=sxrg06XQWFI&t=764s) Over a decade of deploying AI capabilities to technical teams and federal customers demonstrates that the most formidable barrier to adoption is cultural rather than technical. — Christine Palmer

> [12:44](https://www.youtube.com/watch?v=sxrg06XQWFI&t=764s) It is becoming an open, living platform where every person across an organization can directly imagine, build, and adapt technology at the speed of their ideas. — Christine Palmer

## Mentioned Christine Palmer, Peraton, Ai4, Amazon, Uber, Airbnb, YouTube, Ottoman Empire

## Video description Christine Palmer - Chief Technology Officer, Federal Civilian Sector, Peraton

Do you want to experience the future of information? Every generation of computing has asked the same thing of us. Adapt yourself to the tool, and someone else will decide what the tool can do. Agentic AI is already starting down that same path, but this time we don't have to wait. Frontier models have turned language into an interface. Anyone who can describe an idea can now put it in motion, and the systems built on top can reason across whatever scenario is handed to them, without demanding a new ontology or data model each time. That changes what information technology fundamentally is. It stops being something we assemble and becomes something that grows in real time, with the people using it. This session looks at the shape of work on the other side of that shift - how people think and create when the wait is gone, how you build when the tool moves at the speed of the idea, and how the act of building together is what turns information technology into a living platform.

[Source](https://www.youtube.com/watch?v=sxrg06XQWFI)

中文简报

Peraton首席技术官Palmer称AI智能体正在瓦解企业IT把关壁垒 自然语言界面与自主智能体使领域专家能够实时构建与调整企业工作流，从而消除了中间开发者层。 回顾从印刷机、早期大型机到现代交互式AI的知识传播历史轨迹，演讲阐明了企业IT正在经历一场类似于Uber和Airbnb等消费端众包模式的民主化转型。自然语言如今成为核心界面，让领域专家无需通过传统的开发排队流程转化需求，便能直接将想法付诸实践。要实现这一动态生命力平台，需要用自主智能体、上下文工程和嵌入式安全护栏这三大核心基石，取代静态工单与硬编码系统。最终，借助语义网络和上下文数据库实现IT控制平面的现代化，需要克服文化惯性，并树立人机互动共创的思维模式。 回顾奥斯曼抄写员抵制印刷机的历史教训，技术迟疑会导致严重落后；如今自主 AI Agent 正在将数百年的计算演化压缩为实时的情境决策闭环。 (0:04) 企业 IT 正在经历类似于零售和交通行业的去中心化变革，自然语言成为通用界面，使业务专家无需技术团队中介即可在当天直接构建并迭代应用。 (4:03) 新型 IT 架构正被三大核心支柱重构：作为工作基本单元的自主 Agent、提供丰富业务记忆的上下文工程，以及直接嵌入技术底座的自动化安全护栏。 (7:12) 拥抱 Agentic AI 的最大挑战是传统单向交付的思维定势，未来的企业信息技术将不再是自上而下的僵化系统，而是随用户创意实时生长的“生命平台”。 (12:44)

Reading summary

Enterprise information technology is transitioning from a rigid, centralized service model into an adaptive "living platform" driven by autonomous agents and natural language interfaces. Drawing on historical computing cycles and enterprise deployments, Peraton argues that domain experts can increasingly bypass technical intermediaries to design, query, and execute software directly. Democratizing Enterprise IT Through Consumer Platform Models (4:03) The presentation compares the emerging enterprise IT shift to consumer platform disruptions across retail (Amazon), transit (Uber), lodging (Airbnb), and media (YouTube), which bypassed traditional gatekeepers to crowdsource capabilities. Historically, enterprise IT operated as a closed, top-down structure where employees submitted formal tickets and waited for technical specialists or data scientists to construct bespoke applications and dashboards. Today, natural language enables operational experts to query, shape, and modify workflows directly on the day an idea is conceived, preserving institutional nuances that are typically lost across multi-team handoffs. This transition follows an evolutionary computing path that began with isolated 1960s mainframes and progressed through desktops, the internet, and cloud computing. The speaker references the Ottoman Empire's historical reluctance to adopt the movable-type printing press—driven by scribes protecting their employment and social status—as a warning that institutional resistance to automation often results in technological obsolescence while accelerated communication paradigms reshape society. Core Building Blocks: Agents, Context, and Guardrails (7:12) Democratizing enterprise technology requires an enterprise orchestration layer that connects disparate tools under a natural language interface, supported by three primary architectural building blocks: autonomous agents, context engineering, and low-level guardrails. Autonomous agents act as the dynamic unit of work, replacing static IT tickets, brittle scripts, and scheduled cron jobs across enterprise systems. Concurrently, context engineering replaces rigid, hardcoded programming by feeding models institutional memory and operational parameters, enabling dynamic scenario evaluation. The speaker warns that AI safety currently suffers from an implementation gap reminiscent of early internet cybersecurity, noting that while high-level governance policies exist, the automated technical guardrails directly embedded into the execution stack remain nascent and must be built out rapidly. Re-engineering the Enterprise IT Control Plane (10:11) The IT control plane is being restructured across four fundamental layers: semantic networks, contextual databases, dynamic identity management, and actionable observability. Semantic networks move away from manually hardcoded routing paths, allowing agents to interpret intent and discover optimal workflow paths automatically. Contextual databases augment static lookup tables by leveraging large AI context windows, overcoming the natural constraint of human working memory, which typically manages only three to five concepts at once. Security and system monitoring are undergoing a parallel redesign. Identity management must evolve from static compliance checks into real-time authorization systems that verify delegated actions between humans and autonomous agents. Meanwhile, enterprise observability is shifting from passive monitoring dashboards to closed-loop control planes that not only flag network anomalies, but also interpret operational context and execute automated remediations. Overcoming Cultural Inertia in Agentic Environments (12:44) Drawing from over a decade of deploying AI solutions to technical and federal customers, the speaker asserts that the primary obstacle to modernization is cultural rather than technical. Decades of computing have entrenched a passive mindset where users define static requirements and wait for external teams to build deliverables. Adopting living platforms requires an interactive, iterative feedback loop between human operators and software agents. By embedding AI capabilities directly into every layer of the infrastructure, organizations can transition away from expensive top-down IT pipelines and allow non-technical personnel to adapt systems in real time at the speed of their ideas. While agentic orchestration and natural language interfaces offer significant operational agility, enterprise realization depends on bridging the current implementation gap in technical guardrails and dismantling legacy, ticket-based corporate cultures.

精炼阅读摘要

演讲者认为，企业信息技术正从由专门技术团队把关、自上而下的刚性模式，演进为由自主智能体与自然语言驱动的“生命化平台”（Living Platform）。这种范式转变消除了传统的IT工单中介，使一线业务专家能够依托上下文工程与统一编排层，直接通过自然语言构建并动态调整企业级应用。 历史演进与智能体加速机制 (0:04) 人类信息传播与计算架构遵循着一致的历史演化轨迹。历史上，欧洲发明活字印刷术后，奥斯曼帝国的抄写员工会因固守既有社会地位与雇佣结构而拒绝引入自动化印刷，最终导致整体技术发展滞后；而在计算领域，基础设施从1960年代大型机的高成本与物理隔离，逐步演进至个人电脑、互联网以及移动云计算，算力持续向终端用户渗透。演讲者指出，当前软件智能体正在以人类历史上未曾见过的速度重现这一演化周期，生成式AI使人机交互从单向输入硬编码指令转变为双向智能协作，智能体能够快速消化海量信息并在情境中自主完成实时决策。 IT大众化与业务专家的直接构建 (4:03) 消费领域中，亚马逊、优步、Airbnb与YouTube通过众包与平民化模式，打破了传统集中把关的行业壁垒；企业IT当前也迎来了相同的转折点。以往企业IT属于封闭模式，业务人员产生想法后必须提交正式IT工单，等待专门的数据或开发团队制作定制化应用或仪表板，流程漫长且容易因多层转述而丢失业务制度语境。演讲者认为，自然语言成为核心界面后正在瓦解这一中介层，具备一线深厚经验的领域专家能够在提出构想的当天直接查询、构建并可视化所需数据，且支持日后随业务需求实时动态调整。 架构支撑与三大核心技术支柱 (7:12) 要支撑全员以自然语言表达意图，底层需要一套横跨孤立工具的企业编排层、上下文工程层以及执行安全层。该架构由三大核心要素取代传统IT机制：首先是以自主智能体作为基本工作单元，取代静态工单、脆性脚本和定时任务；其次是依托上下文工程为系统注入组织背景知识与记忆，使系统能在安全边界内评估场景并采取最优行动，而非运行死板的硬编码程序；最后是强化技术层面的安全防线。演讲者强调，正如早期互联网的网络安全建设显著滞后一样，当前AI安全也存在落地脱节问题，行业虽有宏观治理政策，但真正内嵌于IT底层的自动化技术防线仍处于萌芽阶段。 IT控制平面的四层技术重构 (10:11) 传统IT控制平面正在四个关键架构维度发生重组：在网络层面，语义网络以意图为导向驱动执行，由智能体自主探测最优路径，免去工程师手动配置海量工作流的负担；在存储层面，鉴于人类工作记忆通常仅能维持3至5个概念，AI大上下文窗口能力使数据库从静态查表升级为具备上下文聚合能力的智能检索系统；在权限层面，身份认证从静态合规转变为支持人机动态授权与细粒度追踪的实时框架；在监控层面，企业可观测性从单纯呈现指标的被动看板，演变为能够识别异常、结合情境理解并自主执行修复的闭环控制系统。 演讲者结合十余年向技术团队及联邦政府客户部署AI的实践指出，普及生命化平台的最大障碍并非技术局限，而是根深蒂固的组织文化惯性。习惯于“提交需求并被动等待交付”的单向思维必须转变为人机实时互动的持续探索循环，才能使IT摆脱自上而下的高成本工程束缚，成为支持全员即时实现创意的开放平台。

Detailed notes — zh

## Background 本内容由 Ai4 于 2026 年 8 月 14 日发布在 YouTube 平台，主题为 Peraton 联邦民用部门首席技术官 Christine Palmer 的演讲《活体平台：AI 如何重塑 Peraton 的信息技术模型》。内容探讨了前沿模型如何将自然语言转化为统一接口，消除系统构建等待期，从而将传统组装式 IT 彻底转变为与用户实时共同生长的“活体平台”（Living Platform）。 ## TL;DR 演讲回顾了从大型主机、PC、互联网、移动计算到云计算的 IT 演进历程，指出传统单向指令流正被生成式与智能体 AI 驱动的双向交互所颠覆。通过将自然语言作为通用接口，并依托编排层、上下文工程与动态护栏，企业能够实现类似消费互联网的去中心化众包创新，让业务人员直接落地构想。随着控制平面在语义网络、上下文数据库、动态身份溯源与闭环可观测性上的重塑，企业必须打破传统工单交付与工程思维，转向人机持续共创与实时进化的活体 IT 架构。 ## Atmosphere & style 整场演讲呈现出高屋建瓴的技术远见与深厚的行业积淀，语气沉稳笃定、逻辑层层递进。讲者以宏大的历史纵深感切入，将奥斯曼印刷术等历史教训与现代企业数字化转型无缝衔接，带来兼具哲思与技术洞见的叙事质感。现场氛围专业而具前瞻启发性，尤其在剖析组织文化惯性与思维重塑时，传递出打破旧有 IT 桎梏、拥抱人机协同进化的紧迫感与从容自信。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=sxrg06XQWFI&t=4s) 信息技术正在演变成实时生长的“活体平台”，智能体正在以指数级的速度重演人类历史上从印刷术、大型主机到互联网的信息通信演进路径。 - [00:04](https://www.youtube.com/watch?v=sxrg06XQWFI&t=4s) 奥斯曼帝国因维护抄写员利益而推迟普及印刷术的历史案例表明，对自动化与信息去中心化的抗拒会导致技术落后，这对当今 AI 变革依然具有警示意义。 - [01:58](https://www.youtube.com/watch?v=sxrg06XQWFI&t=118s) 计算技术演进从大型主机、PC、互联网到云计算，始终是人类单向输入指令；而智能体 AI 彻底打破了这一模式，开启了人机双向深度交互与系统反哺洞察的新阶段。 - [04:03](https://www.youtube.com/watch?v=sxrg06XQWFI&t=243s) 消费互联网已通过众包去中心化颠覆了零售和出行，但企业级 IT 仍深陷“提交工单、等待专家排期定制”的中心化旧模式，亟待向去中心化众包模式转型。 - [06:30](https://www.youtube.com/watch?v=sxrg06XQWFI&t=390s) “自然语言即接口”打破了业务与技术之间的壁垒，业务人员可直接用业务语言将想法转化为实际运行的系统，避免了多层需求转译造成的信息损失。 - [06:30](https://www.youtube.com/watch?v=sxrg06XQWFI&t=390s) 实现活体平台需要在交互层（自然语言）、编排层（多智能体互联调度）以及数据上下文层（动态注入组织规则）三大架构层面实现协同重构。 - [08:18](https://www.youtube.com/watch?v=sxrg06XQWFI&t=498s) 智能体（Agent）正成为替代传统工单（Tickets）、脚本及定时任务（Cron jobs）的全新工作单元，重塑了端到端的工作流转方式。 - [08:18](https://www.youtube.com/watch?v=sxrg06XQWFI&t=498s) 上下文工程（Context Engineering）是摆脱传统静态软件的关键，它使系统能够依据充足的背景信息在规则边界内自主判断最优解。 - [10:11](https://www.youtube.com/watch?v=sxrg06XQWFI&t=611s) IT 控制平面正在发生深刻重塑：网络升级为意图驱动的语义网络，数据库演化为上下文数据库，身份鉴权走向动态行为溯源，监控进化为具备自动修复能力的闭环可观测性。 - [12:44](https://www.youtube.com/watch?v=sxrg06XQWFI&t=764s) 企业应用 AI 的最大瓶颈在于组织文化与传统单向工程思维的惯性；企业必须从“请求-交付单一结果”转向“人机共创、持续迭代”的活体生长模式。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=sxrg06XQWFI&t=4s) 活体平台与信息传递的历史演进 信息技术正在演变成一个**“活体平台”（Living Platform）**。从某种意义上说，人工智能（尤其是智能体 Agent）正在以一种类似“灵魂出窍”（out-of-body）的方式，重演人类历史上经历过的信息演化路径。回顾历史，人类最早跨越地域和国界传递信息的主要媒介是书籍。关于印刷术的发展，奥斯曼帝国原本有机会在欧洲之前率先普及这项技术，但当时由于帝国境内庞大的抄写员（scribes）群体负责人工抄写和传播信息，统治阶层不愿引入更多自动化技术来推动信息的全球化扩散，最终导致了技术的落后。如今，印刷出版依然存在，而这种演进模式也贯穿了人类与技术沟通的整个历程——从早期的计算机、大型主机（Mainframes），到依托底层技术催生出的社交媒体与实时通信网络。 人类社会长期以来所经历的信息通信变迁，如今正在智能体身上重现，但智能体的发展速度极其迅猛：它们吸收的信息量呈指数级增长，并且能够凭借远超人类的上下文理解能力更快速地做出决策。人机之间信息交互模式的变迁，正推动着技术体系从单向工具走向活体生命般的智能协作。 --- ### [01:58](https://www.youtube.com/watch?v=sxrg06XQWFI&t=118s) 从大型主机到交互式 AI：IT 范式的代际演变 计算机与信息技术的发展始于 20 世纪 60 年代的大型主机时代。那时的算力主要服务于需要庞大计算资源的大型科研机构和组织，系统极其昂贵且难以获取，普通人根本无法接触。随着技术迭代，计算能力逐渐走向桌面和办公电脑，人们在工作场所获得了更多访问权限，信息也从大型主机上的静态存储转变为办公室内同事间的本地共享。紧接着，互联网与万维网的爆发彻底打破了物理边界，实现了更广泛受众之间的互联互通。随后，桌面电脑被装进了人们的口袋——移动设备让计算随时随地发生，彻底摆脱了办公桌的束缚。 云计算的出现则让访问海量数据集与超大数据库变得轻而易举，这直接奠定了当前大语言模型（LLM）以及高算力需求 AI 技术的基础。在以往，只有资金雄厚、自建本地数据中心的大型组织才能运行此类系统。AI 带来的并非简单的渐进式迭代，而是对整个 IT 模型的底层重构：以往的技术演进本质上是“单向的”（One-directional）——人类输入指令与软件硬件构建规则；而现在的 AI 带来了双向的深度交互，系统不仅能接收指令，更能主动反哺洞察，指导人类如何更好地沟通与构建系统。 ```mermaid flowchart LR accTitle: IT Interaction Paradigm Evolution subgraph S1["第一阶段：集中与静态"] A["60年代 大型主机<br/>(机构专享/静态受限)"] --> B["桌面与办公PC<br/>(局域协同)"] end subgraph S2["第二阶段：网络与分布式"] B --> C["互联网 / 移动端<br/>(随时随地触达)"] --> D["云计算<br/>(算力普惠/海量数据)"] end subgraph S3["第三阶段：交互式活体平台"] D --> E["生成式与智能体 AI<br/>(双向交互/意图驱动)"] end ``` --- ### [04:03](https://www.youtube.com/watch?v=sxrg06XQWFI&t=243s) 众包革命与企业 IT 的去中心化契机 消费互联网领域已经通过“众包”（Crowdsourcing）模式彻底颠覆了多个传统行业： - **亚马逊（Amazon）**：打破了传统实体店销售壁垒，让普通人都能以极其高效流线化的方式销售商品和服务。 - **优步（Uber）**：省去了昂贵的年度出租车牌照和中介调度系统，让大众能够自由提供出行服务。 - **爱彼迎（Airbnb）**与 **YouTube**：分别将全球普通人的房屋变成了旅宿，让每个人都成为了内容创作者。 然而，企业级 IT 至今仍然被困在旧有的中心化框架中。尽管 AI 能力已经渗透到整个基础设施技术栈，但企业构建软件的方式依然遵循老套路：业务人员提出需求 $\rightarrow$ 提交工单（Ticket） $\rightarrow$ 由专业技术团队定制开发。企业管理层习惯性地为 AI 设定类似传统软件交付的条条框框与安全边界，但这种思维极大地限制了生产力。 企业 IT 的未来在于实现真正的去中心化众包：“如果企业内部任何人的任何想法，都能直接由提出想法的人亲手构建出来，而不是提交需求等待专家定制，世界会变成什么样？”过去的业务专家由于缺乏编码技能，不得不依赖数据科学家为其制作可视化仪表盘（Dashboard）；如今借助 AI，业务人员不仅能直接调取所需信息，还能根据当天或明天的不同需求随时变换呈现形式，彻底抹平了从业务痛点到软件实现之间的漫长鸿沟。 --- ### [06:30](https://www.youtube.com/watch?v=sxrg06XQWFI&t=390s) 自然语言即接口与三大基础设施构件 让业务专家亲手构建系统的核心支撑，在于**“自然语言成为了通用接口”**（Language is the interface）。过去在将业务需求翻译为技术规格说明书的过程中，大量信息和业务洞察会在层层转译中丢失；而现在，员工可以用自己最熟悉的业务语言清晰表达意图（Intent），并将其直接转化为运行中的系统。这种模式不仅让业务构想在产生的当天即可落地运行，还能与企业的各项组织规则和合规底线深度锚定。 要在企业级技术栈全面实现这一愿景，并非仅靠输入自然语言提示词就能达成，它依赖于三个关键层面的协同重构： | 构建层级 | 传统模式 | 智能体活体平台模式 | |:--- |:--- |:--- | | **交互层（Interface Layer）** | 代码、表单、硬编码 UI | 纯自然语言交互，直观表达业务意图 | | **编排层（Orchestration Layer）** | 单一工具套件、人工胶水代码 | 跨技术栈全域互联，协同调用多智能体 | | **数据与上下文层（Data & Context）** | 静态数据库、预定义数据模型 | 动态注入业务上下文与组织规则 | 这如同当年人类从大型主机跨越到个人电脑、从局域网构建到万维网一样，当前的当务之急是构建起打通全企业系统的强大编排层。 --- ### [08:18](https://www.youtube.com/watch?v=sxrg06XQWFI&t=498s) 智能体、上下文工程与护栏机制 要让这一全新体系运转，企业 IT 正在引入三大新型构建基块： ```mermaid flowchart TD accTitle: Three New IT Building Blocks subgraph BB["三大新型基础设施基块"] direction TB B1["智能体 Agent<br/>(替代工单、脚本与 Cron 作业的新工作单元)"] B2["上下文工程 Context Engineering<br/>(赋予系统自主判断与决策能力)"] B3["动态护栏 Guardrails<br/>(保障安全、合规与行为边界)"] end B1 --> B2 B2 --> B3 ``` 1. **智能体（Agents）作为全新工作单元**：智能体正在全面取代传统的工单（Tickets）、自动化脚本和定时任务（Cron jobs）。过去的单向流转闭环被智能体网络打破，当智能体通过编排层互联时，大部分机械式的流转工作被彻底消除。 2. **上下文工程（Context Engineering）**：这是摆脱“静态软件”的关键。传统软件只能死板执行预先编写好的规则，而上下文工程赋予了系统充足的信息背景，使系统不再只是被动按指令行事，而是能在规则范围内“自主判断何为最佳解决方案”，推动人机协作走向真正的双向共创循环。 3. **技术级安全护栏（Guardrails）**：目前护栏机制面临的挑战正如互联网初期网络安全（Cybersecurity）严重滞后于网络普及一样，行业普遍意识到了风险，但大多停留在政策规范（Policy level）层面，缺乏行之有效的底层技术落地解决方案。 --- ### [10:11](https://www.youtube.com/watch?v=sxrg06XQWFI&t=611s) 控制平面的重塑：语义网络、上下文数据库、身份与可观测性 随着 IT 运行模式的演进，整个系统的控制平面（Control Plane）正在四个核心维度发生深刻重塑： - **网络走向语义化（Semantic Network）**：网络不再仅仅关注数据包路由，而是理解用户的意图。用户无需掌握达成目标的上百种技术路径组合，智能体与 AI 会根据意图自动寻优并构建最佳执行链路。 - **数据库走向上下文驱动（Contextual Database）**：人类大脑的工作记忆（Working memory）在同一时刻通常只能容纳 3 到 5 个概念，而 AI 的容量和调动能力远超人类极限。通过人机接口，人类负责把握核心业务方向与特定领域的上下文，AI 则负责调集全域多维数据进行上下文增强，二者优势互补。 - **身份认证与权限溯源（Identity & Traceability）**：身份过去只是一个静态的勾选框（Check box），但在人机与多智能体高度交织的复杂环境中，系统必须能够实时追踪人与智能体之间的行为边界、授权流转与操作权限，静态鉴权正被迫转向高度流动的动态追溯体系。 - **闭环可观测性（Actionable Observability）**：传统的监控仅停留在“了解网络和企业内部正在发生什么”，而与智能体结合后的新一代可观测性，不仅能实时感知问题，还具备完整的上下文修复能力，直接形成“监测 $\rightarrow$ 诊断 $\rightarrow$ 自动修复”的可操作控制闭环。 --- ### [12:44](https://www.youtube.com/watch?v=sxrg06XQWFI&t=764s) 颠覆单向工程思维：迈向共创式“活体平台” 数十年来，信息技术的演进始终被困在单向思维中：人类定义功能、定义访问方式、定义信息流转路径，然后向系统索取结果。即便如今部署智能体 AI，很多组织依然在不自觉地复制这种传统工程流程。 > “在过去的十多年里，我一直在向团队和客户推广 AI 能力。生成式 AI 在过去几年里带来了翻天覆地的技术变革，但其中最大的挑战——也是持续至今的挑战——始终是**文化和思维模式的转变**。” 企业必须彻底打破“提出请求 $\rightarrow$ 获取单一静态结果”的思维定势。人机与智能体的协作本质上是一个**持续迭代的探索过程**：人类与 AI 并肩工作，从 AI 提供的增量信息中汲取灵感，持续进行反馈循环，从而创造出超越以往极限的成果。如今，AI 能力已经嵌入到 IT 架构的几乎每一个工具节点中，只要企业敢于打破旧有审批与开发框架，全员由下而上的想象力就能直接转化为生产力，使整个企业 IT 真正演变为一个实时生长的“活体平台”。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=sxrg06XQWFI&t=4s) 人类社会长期以来所经历的信息通信变迁，如今正在智能体身上重现，但智能体的发展速度极其迅猛：它们吸收的信息量呈指数级增长，并且能够凭借远超人类的上下文理解能力更快速地做出决策。 — Christine Palmer > [01:58](https://www.youtube.com/watch?v=sxrg06XQWFI&t=118s) 以往的技术演进本质上是“单向的”（One-directional）——人类输入指令与软件硬件构建规则；而现在的 AI 带来了双向的深度交互，系统不仅能接收指令，更能主动反哺洞察，指导人类如何更好地沟通与构建系统。 — Christine Palmer > [04:03](https://www.youtube.com/watch?v=sxrg06XQWFI&t=243s) 如果企业内部任何人的任何想法，都能直接由提出想法的人亲手构建出来，而不是提交需求等待专家定制，世界会变成什么样？ — Christine Palmer > [06:30](https://www.youtube.com/watch?v=sxrg06XQWFI&t=390s) 过去在将业务需求翻译为技术规格说明书的过程中，大量信息和业务洞察会在层层转译中丢失；而现在，员工可以用自己最熟悉的业务语言清晰表达意图（Intent），并将其直接转化为运行中的系统。 — Christine Palmer > [08:18](https://www.youtube.com/watch?v=sxrg06XQWFI&t=498s) 智能体正在全面取代传统的工单（Tickets）、自动化脚本和定时任务（Cron jobs）。过去的单向流转闭环被智能体网络打破，当智能体通过编排层互联时，大部分机械式的流转工作被彻底消除。 — Christine Palmer > [08:18](https://www.youtube.com/watch?v=sxrg06XQWFI&t=498s) 传统软件只能死板执行预先编写好的规则，而上下文工程赋予了系统充足的信息背景，使系统不再只是被动按指令行事，而是能在规则范围内“自主判断何为最佳解决方案”，推动人机协作走向真正的双向共创循环。 — Christine Palmer > [10:11](https://www.youtube.com/watch?v=sxrg06XQWFI&t=611s) 网络不再仅仅关注数据包路由，而是理解用户的意图。用户无需掌握达成目标的上百种技术路径组合，智能体与 AI 会根据意图自动寻优并构建最佳执行链路。 — Christine Palmer > [10:11](https://www.youtube.com/watch?v=sxrg06XQWFI&t=611s) 身份过去只是一个静态的勾选框（Check box），但在人机与多智能体高度交织的复杂环境中，系统必须能够实时追踪人与智能体之间的行为边界、授权流转与操作权限，静态鉴权正被迫转向高度流动的动态追溯体系。 — Christine Palmer > [12:44](https://www.youtube.com/watch?v=sxrg06XQWFI&t=764s) 在过去的十多年里，我一直在向团队和客户推广 AI 能力。生成式 AI 在过去几年里带来了翻天覆地的技术变革，但其中最大的挑战——也是持续至今的挑战——始终是文化和思维模式的转变。 — Christine Palmer > [12:44](https://www.youtube.com/watch?v=sxrg06XQWFI&t=764s) 人机与智能体的协作本质上是一个持续迭代的探索过程：人类与 AI 并肩工作，从 AI 提供的增量信息中汲取灵感，持续进行反馈循环，从而创造出超越以往极限的成果。 — Christine Palmer ## Mentioned Christine Palmer, Peraton, Ai4, 亚马逊 (Amazon), 优步 (Uber), 爱彼迎 (Airbnb), YouTube, 大型主机 (Mainframes), 云计算 (Cloud Computing), 大语言模型 (LLM), 智能体 (Agentic AI / Agents), 上下文工程 (Context Engineering), 控制平面 (Control Plane), 语义网络 (Semantic Network) ## Video description Christine Palmer - Chief Technology Officer, Federal Civilian Sector, Peraton Do you want to experience the future of information? Every generation of computing has asked the same thing of us. Adapt yourself to the tool, and someone else will decide what the tool can do. Agentic AI is already starting down that same path, but this time we don't have to wait. Frontier models have turned language into an interface. Anyone who can describe an idea can now put it in motion, and the systems built on top can reason across whatever scenario is handed to them, without demanding a new ontology or data model each time. That changes what information technology fundamentally is. It stops being something we assemble and becomes something that grows in real time, with the people using it. This session looks at the shape of work on the other side of that shift - how people think and create when the wait is gone, how you build when the tool moves at the speed of the idea, and how the act of building together is what turns information technology into a living platform. [Source](https://www.youtube.com/watch?v=sxrg06XQWFI)

Source quotes

[0:04] Christine Palmer: 人类社会长期以来所经历的信息通信变迁，如今正在智能体身上重现，但智能体的发展速度极其迅猛：它们吸收的信息量呈指数级增长，并且能够凭借远超人类的上下文理解能力更快速地做出决策。 [1:58] Christine Palmer: 以往的技术演进本质上是“单向的”（One-directional）——人类输入指令与软件硬件构建规则；而现在的 AI 带来了双向的深度交互，系统不仅能接收指令，更能主动反哺洞察，指导人类如何更好地沟通与构建系统。 [4:03] Christine Palmer: 如果企业内部任何人的任何想法，都能直接由提出想法的人亲手构建出来，而不是提交需求等待专家定制，世界会变成什么样？ [6:30] Christine Palmer: 过去在将业务需求翻译为技术规格说明书的过程中，大量信息和业务洞察会在层层转译中丢失；而现在，员工可以用自己最熟悉的业务语言清晰表达意图（Intent），并将其直接转化为运行中的系统。 [8:18] Christine Palmer: 智能体正在全面取代传统的工单（Tickets）、自动化脚本和定时任务（Cron jobs）。过去的单向流转闭环被智能体网络打破，当智能体通过编排层互联时，大部分机械式的流转工作被彻底消除。 [8:18] Christine Palmer: 传统软件只能死板执行预先编写好的规则，而上下文工程赋予了系统充足的信息背景，使系统不再只是被动按指令行事，而是能在规则范围内“自主判断何为最佳解决方案”，推动人机协作走向真正的双向共创循环。 [10:11] Christine Palmer: 网络不再仅仅关注数据包路由，而是理解用户的意图。用户无需掌握达成目标的上百种技术路径组合，智能体与 AI 会根据意图自动寻优并构建最佳执行链路。 [10:11] Christine Palmer: 身份过去只是一个静态的勾选框（Check box），但在人机与多智能体高度交织的复杂环境中，系统必须能够实时追踪人与智能体之间的行为边界、授权流转与操作权限，静态鉴权正被迫转向高度流动的动态追溯体系。 [12:44] Christine Palmer: 在过去的十多年里，我一直在向团队和客户推广 AI 能力。生成式 AI 在过去几年里带来了翻天覆地的技术变革，但其中最大的挑战——也是持续至今的挑战——始终是文化和思维模式的转变。 [12:44] Christine Palmer: 人机与智能体的协作本质上是一个持续迭代的探索过程：人类与 AI 并肩工作，从 AI 提供的增量信息中汲取灵感，持续进行反馈循环，从而创造出超越以往极限的成果。

Source quotes

Christine Palmer: Every generation of computing has asked the same thing of us. Adapt yourself to the tool, and someone else will decide what the tool can do. Christine Palmer: Frontier models have turned language into an interface. Anyone who can describe an idea can now put it in motion, and the systems built on top can reason across whatever scenario is handed to them, without demanding a new ontology or data model each time. Christine Palmer: It stops being something we assemble and becomes something that grows in real time, with the people using it. [1:58] Christine Palmer: The defining shift with generative AI is that interaction is no longer one-directional—users no longer simply write code or feed rigid instructions into a machine; systems now actively analyze inputs, respond with real-time insights, and suggest optimized ways to communicate and build. [4:03] Christine Palmer: Building directly in natural language avoids the loss of institutional nuance that typically occurs when business requirements are translated across multiple technical teams. [7:12] Christine Palmer: Eliminating the barrier of having to learn a programming language allows users to express pure intent, which the underlying system then translates into functional technology. [7:12] Christine Palmer: Instead of engineering a rigid, static application that executes only hardcoded commands, context engineering supplies systems with rich background knowledge so they can evaluate scenarios and take the most effective action within safe operational parameters. [10:11] Christine Palmer: Rather than requiring human engineers to anticipate and configure hundreds of possible workflow combinations, semantic networks utilize agents to interpret human intent and automatically discover the optimal path forward. [12:44] Christine Palmer: Over a decade of deploying AI capabilities to technical teams and federal customers demonstrates that the most formidable barrier to adoption is cultural rather than technical. [12:44] Christine Palmer: It is becoming an open, living platform where every person across an organization can directly imagine, build, and adapt technology at the speed of their ideas.

Original transcript

[0:04] Information technology is becoming a living  platform. And when you think about it, [0:09] it's kind of an out-of-body experience the way  that AI and agents in particular are basically [0:15] experiencing the world the same way that we  have over all, you know, all throughout history. [0:21] Um and the way that that's kind of reflected, you  know, when you think about it, the first way that [0:25] we communicated information across boundaries,  across countries really was through books. [0:31] And one of the fav- one of my favorite stories  from that, I read an article a few weeks ago, [0:35] um that the printing press was originally  created in Europe, but it was actually the [0:40] Ottoman Empire that had the opportunity to create  it first. And the reason they didn't was because [0:45] they had scribes creating information that was  shared with people. And they didn't want to, [0:50] you know, bring in more automation to make that  something that could be spread across the world. [0:55] And they fell behind. So, today we still have the  printing press, books are still being printed, [0:59] it's still being shared. And you kind of  see that trend across the way that we've [1:04] continued to evolve how we communicate  with each other and then eventually with [1:07] technology. [snorts] You know, we see that  through the creation of the first computer, [1:12] the mainframes, how we, you know, evolved  all of that from a technology perspective, [1:17] and then how we actually leverage that foundation  to do more things like social media and actually [1:21] new ways of communicating with each other more  in real time. So, it's an interesting trend [1:27] and all of this is what we as a society have,  you know, observed and seen and experienced. [1:32] And now we're seeing agents do the same thing, but  they're doing it much faster, they're getting more [1:36] information, they're able to make decisions much,  you know, much quicker and with more context than [1:41] a human ever could. So, I'm going to walk  through a little bit of what that looks like, [1:47] what we're kind of seeing now, and, you know,  how how that's reflected in the way that we've [1:51] been doing commun- we how we've been communicating  information between humans and machines over time. [1:58] So, for computers and information technology in  particular, what really started was around the [2:05] '60s when the first mainframes came out. Now,  that was really targeted at large institutions, [2:10] organizations, scientific research that needed a  lot of computing power, and that's really what it [2:16] was focused on. It was very hard to access, not  a lot of people had it except for researchers or, [2:21] you know, large institutions. That kept evolving,  so we had uh more progress towards desktops and [2:27] office computers. So, if you were at work, you  usually had a little bit more access to that. [2:32] It still really wasn't that common, but it was  still that theme of, you know, we started with [2:36] being able to access information statically in a  mainframe, starting to be able to access it more [2:41] between coworkers on site at your offices, and  then evolving further into the internet and web, [2:47] and now we're able to communicate information  across a broader array a broader audience. [2:53] And that keeps evolving, so desktops started  moving into your pocket. Now, we have mobile [2:57] devices. You can carry it with you everywhere.  You used to have to be tied to your desk to be [3:01] able to do that. So, you keep seeing the iteration  evolving over time. Cloud makes it even easier to [3:07] access larger databases and larger data sets  that we weren't able to do before, which is a [3:12] foundational way that we're able to access large  language models now and other AI capabilities that [3:17] require a lot of compute. Uh used to be you had  to have, you know, an on-site data center to be [3:22] able to do that, and again, that was restricted to  large organizations that had the money to do that, [3:26] not really the average users. And now with AI,  what we're really seeing is kind of a tran- [3:34] really a transformation of that entire model that  was very iteratively going down the same path, [3:39] bringing information closer and closer to us,  but now it's not so one-directional. It's more [3:44] interactive. We're able to not just feed  information to, you know, different systems [3:49] of how we want to build software and hardware.  We can actually interact and get more insights [3:52] back into what we should be doing and what we  could be doing better to communicate information. [3:59] So, when you think about what that means, [4:03] we have an opening here that has been addressed in  a lot of other industries today, where, you know, [4:09] different models have been crowdsourced  that were really individual areas where, [4:15] you know, you had individual companies doing  certain, um, you know, for Amazon, for example, [4:20] different brick-and-mortar companies that would  sell things. Amazon broke that model and made [4:24] it easier for the average person to be able to  sell their products and their capabilities in [4:29] a very streamlined sort of fashion. Same with  Uber, made it easier for people to be able to [4:34] give rides. You didn't have to go through a taxi  service. You didn't have to have a, uh, you know, [4:39] an annual license that you paid for. Opens it up  to society and the population. Airbnb, same thing [4:44] with homes. YouTube also opened it up for everyone  to be a creator. And now, with enterprise IT, [4:52] this is the same opportunity that we're starting  to see on a gradual basis, where we have a lot of [4:57] AI capabilities across the entire infrastructure  stack. And the last part that we haven't really [5:02] gotten to yet is being able to crowdsource that.  So, we're still operating on the same model of, [5:08] you know, if we want a new software built,  they're still you go and request it to be built. [5:13] A lot of times, organizations are still putting  boundaries around the way that we do this with AI, [5:18] the same way that we've done software. We need to  put guardrails around it, we need to deliver it in [5:23] the same way that we've done software, and that's  not really how we have to do it if you think about [5:26] it differently. There are different ways to  do it now, and it could be transformational. [5:33] So, just to get into that a little bit more, what  if any idea that anyone had in the enterprise [5:39] could be built by the person who had it, rather  than the standard model before, where you had to [5:44] put in a request, you had to have an expert build  something bespoke for you? How much of that do we [5:48] actually need now? How much of it can we actually  move closer to the person who needs to do the [5:53] work? They are the ones with the expertise.  Move it closer to them to be able to do that. [5:58] Um you know, I've been a data scientist for a  while and a lot of what I did in those roles was [6:02] creating dashboards for people who, you know,  didn't have the technical ability to do that. [6:07] That's completely changed now. That a lot of  times you can access the information you need, [6:11] present it in the way that you want to see it, and  maybe you want to see it differently tomorrow, and [6:14] you can do that, too. So, it's really shortening  that gap between the information you need, [6:19] how you want to see it communicated, and  how you want to present it to others, [6:22] and opening that up to the enterprise is how  you get more towards that crowdsource model. [6:30] And what that's really built on is being able  to deliver it to both the workforce, users, [6:36] customers, people in a way that it can actually  be built by the people who have that expertise. [6:41] They can describe it in the language that  they speak instead of losing information in [6:45] translation as you go between different  requirements. Um it can actually be run [6:50] the same day that the idea has arrived, uh and  then grounded in the different organizational [6:56] rules that need to be accounted for. So, even just  from a productivity and efficiency perspective, [7:00] being able to shorten that time frame because  you're really putting these capabilities in the [7:04] hands of people who can run it best themselves,  and really grow that capability from there. [7:12] And the way this is actually going to be possible  and is possible in pockets today and in silos [7:19] is that language is the interface, and that makes  it accessible to everyone. You don't have to learn [7:23] a coding language to be able to do that. There  are things that need to be connected across the [7:28] enterprise to be able to make that work, but  ultimately getting more towards a model where [7:33] you can express the intent of what you want to do  and have that automatically translated into things [7:38] that you want to build is what will transform  the information the information technology model. [7:44] There's a few layers that you have to go  through to do that. It's not just about [7:48] being able to query it in plain language,  although that is one, that interface layer. [7:52] Uh it does require a lot more orchestration and  usually that's not a single tool at an enterprise [7:58] level or for the entire information technology  stack. That's a lot of connectivity that has to be [8:03] done, but we did it from mainframes to computers,  we did it when we built network capabilities, [8:07] we did it when we built the web. This is what  needs to be built now is that orchestration layer. [8:12] The data layer is the next most important one,  which we'll also cover as part of the context. So, [8:18] there's three new building blocks that we'll  have to do to actually make this work. Agents [8:23] are already out there, we're already using that in  a lot of different ways. Uh it's a it's a new unit [8:28] of the way that we actually do work now. Um you  know, replacing tickets and scripts and cron jobs, [8:33] for example, those are you the standard way that  you would build the one directional request uh [8:38] and you know, create that closed loop of what  people need to actually do their job or to get [8:43] access to the software and the information that  they need. Um agents kind of eliminate a lot more [8:49] of that when they're actually connected via that  orchestration layer as well. Now, context is the [8:54] really big one here and that's context engineering  is really the the way that we get away from the [9:01] idea of build a static piece of software that does  what I specified it needs to do and now it has all [9:06] the information to not just do what I told it to  do, but to do what it thinks is best, you know, [9:11] obviously with guardrails and parameters around  that, but the closer you get to being able to [9:16] move towards our true human machine interface  where you can work together and it's less of you [9:21] directing the more you the more you actually get  not just from the system itself, but also for your [9:28] own productivity and really creating that whole  creative loop between the human machine interface. [9:34] Obviously, guardrails is a huge part of that and I  hear this a lot in a lot of different places. That [9:40] is really an area that I think needs to be  built out a lot more, and there are some [9:44] places doing it, but really from a technical  perspective for information technology, um [9:50] the challenge there is like when the internet  first came out and then cybersecurity came along [9:55] quite a bit after that. I think everyone's aware  of that issue now, but there's not a lot of great [9:58] solutions for how to do that beyond a policy level  to actually implement it. So, that obviously is a [10:04] very big one that needs to be addressed and  is being addressed in small pockets today. [10:11] So, when you think about how information  technology is actually modeled and controlled [10:16] in the way that we actually operate it today, you  can really break it out as a control plane and how [10:21] it's actually taking shape across many different  areas. So, the network itself is really becoming [10:26] more semantic, and that goes back to the ability  to express the intent of what you want to do [10:32] and let AI and agents help you figure out  the best way to do that. So, it doesn't just [10:36] require you to know every possible 100  combinations of the best path to take. [10:41] Really leveraging that information um to find  the best option to build what you want to build. [10:48] The database is is becoming more contextual,  and this is an important one as well because [10:53] when you think about it, the human brain  can hold about maybe three to five concepts [10:58] in working memory at any time. AI also has  limits to that. Its limits are much greater. [11:03] And when you have a genetic AI that really pulls  resources in a way that no single human can do. [11:08] And so, this is where it goes back  to that human-machine interface. You [11:11] have that direction, you have the targeted  domain context that you want to get towards, [11:17] and being able to pull in the rest of what you can  do with AI for additional context as your database [11:23] is really what's going to get you towards that  that ultimate transformation of the IT model. [11:29] This does introduce uh you know, more issues.  Identity is one that's becoming, you know, [11:34] it's it used to be the layer that you would add  on just to make sure everything was a check box [11:39] check box, it's good to go. And now we're getting  more into a scenario of how do you actually trace [11:43] authority and activities and actions that are  being taken between human and agents across [11:48] the entire IT model. There's a lot of interfaces  now and that traceability is difficult to do, uh, [11:55] because it was really more of a static interface  before and now it's becoming much more fluid, it's [11:59] changing much more quickly and it really is an  emerging area that needs to be addressed as well. [12:05] And then observability is, uh, a significant area  that also needs to be addressed and is starting, [12:12] you know, we're starting to move away  from I just want to see what's happening [12:15] on my network and I just want to see  what's happening in my enterprise. [12:19] The next layer of that from the information  technology perspective, the next layer of that [12:24] is being able to take combine that with agentic  activities and really close that loop between, [12:30] okay, I know this is what's happening, I  have the full context of how to fix it and [12:33] how to make it better and how do I actually pull  that together into an actionable control plane. [12:44] So for decades, we have engineered technology.  It's been very one directional again. [12:51] We have defined what it should look like, we've  defined how we need to access information and [12:56] how we want to communicate it to each other, uh,  and to across the information technology stack. [13:03] Every iteration of advancements that we've  had from the mainframe all the way up to [13:08] now and even the way that agentic AI is being  deployed is still following that same model. [13:13] So this is the time to really take a step back  and think, is this really, do I really need to [13:17] replicate the same way? And I'll even, uh, give  some examples. So I've I've been rolling out [13:23] AI capabilities to my staff and to customers  my team for about over a decade now. And the [13:29] transformation even just in the last couple of  years has been significant with generative AI. [13:35] The cultural shift I think has been the biggest  challenge and is continuing to be the biggest [13:39] challenge because that mindset of I need to  I want to request something to be made for me [13:46] and you know that's that's the result that I will  get. That is a hard one to unravel. And when you [13:50] get into more of the human agentic AI interface  that's the mindset that we have to get past and [13:57] it can be challenging because it's really not you  requesting something and getting one result back. [14:01] It really is an iterative process and it really  is hand-in-hand seeing what more information it [14:07] can give you, what you can learn from the AI  system and how you keep continuing with that [14:12] loop to build even more than what you're able  to do before. And so that that shift has been um [14:18] significant but I think it's really  what we need to do to get more to [14:23] really a new sort of model of how we can operate  this both from a crowdsource perspective, the [14:28] you know, how how we actually communicate  information across the entire model. [14:33] So now we are in a position where we have  AI capabilities in nearly every single tool [14:38] across the entire information technology stack.  We're enabling it, we're turning everything on, [14:43] we're connecting everything and it's getting  more towards that idea that we can actually [14:47] just imagine what we want and then really  start building it ourselves. Every person, [14:52] every every organization at all levels. And so  we really can get beyond where we just had to [14:57] specify the engineering component to it and it  can actually be imagined by all of us every day. [15:05] So that's all I had for today. I thought  I had a last slide but probably not. Um [15:11] so if you have any questions uh you  can reach out to me. We have you can [15:16] uh look us up pertonics.com/x and I'll be  happy to talk to any of you later. Thank you.