AI InnerSource Usurper or Superpower #InnerSourceSummit2025

AI: InnerSource Usurper or Superpower? #InnerSourceSummit2025 🔗 **Source:** [AI: InnerSource Usurper or Superpower? #InnerSourceSummit2025](https://www.youtube.com/watch?v=CtWGVX700J4) > [!note]+ Document Notes > tags: [inner-source, ai-development, developer-experience] > author: InnerSource Commons > source: YouTube > reference: https://www.youtube.com/watch?v=CtWGVX700J4 > date created: 2024-12-19 14:30 > > # 🏷️ Title: AI: InnerSource Usurper or Superpower? #InnerSourceSummit2025 >  ## 🧰 Action Inbox - [H1] Evaluate current inner source documentation for AI agent consumption — 00:00-03:45 #transparency - [H2] Consider MCP server integration for internal libraries — 03:45-07:30 #discovery - [H3] Apply inner source principles to AI agent development — 07:30-11:15 #community ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [InnerSource Commons](https://innersourcecommons.org/) - [AWS CDK Constructs](https://docs.aws.amazon.com/cdk/latest/guide/constructs.html) ## 📝 Summary - Lizzie Celita from Booz Allen explores whether AI threatens inner source or enhances it - Inner source principles (transparency, community development, collaboration) remain valuable in AI era - AI can accelerate inner source adoption through better discovery, contribution tooling, and cross-functional handoffs - Booz Allen examples include MCP servers for AWS constructs, agent studio marketplace, and design-to-code automation - Bottom line: AI complements rather than replaces inner source, with governance becoming even more critical ## 📖 Main Content The presentation addresses a fundamental question facing organizations embracing both AI and inner source practices: does AI make inner source obsolete by allowing developers to generate custom solutions instantly? Celita argues that while AI changes the development landscape, the core principles of inner source become more important, not less. She identifies three key inner source principles that remain valuable: transparency (for discovery and adoption), community-driven development (for shared improvements), and collaboration (for knowledge sharing). Each principle not only benefits from AI tooling but also helps organizations develop better AI solutions. The transparency principle ensures that organizational knowledge is documented and discoverable - exactly what's needed for AI agents to understand company-specific solutions. Community-driven development concentrates security and compliance efforts on reusable solutions rather than scattered AI-generated code. Collaboration builds the human connections that AI cannot replace while enabling better cross-functional workflows. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/CtWGVX700J4?t=0) — [H1] The changing inner source equation - Key idea: AI appears to offer faster custom solutions than traditional inner source adoption - Evidence: "I can just ask AI to build exactly what I need... seems a lot faster and easier" #### ⏱️ [01:15](https://youtu.be/CtWGVX700J4?t=75) — [H2] Risk of returning to silos - Key idea: AI-generated solutions may recreate the isolation problems inner source solved - Evidence: "We may be spending less money, time, and resources, but we risk developing those same silos" #### ⏱️ [02:00](https://youtu.be/CtWGVX700J4?t=120) — [H3] Speaker introduction and context - Key idea: Booz Allen builds AI solutions for federal clients where efficiency is critical - Evidence: "Building advanced technology and AI solutions for federal government clients" #### ⏱️ [02:45](https://youtu.be/CtWGVX700J4?t=165) — [H1] Three principles framework - Key idea: Inner source principles enable organizational innovation and resilience - Evidence: "Transparency powers discovery... Community-driven development sharpens solutions... Collaboration underpins the ecosystem" #### ⏱️ [03:30](https://youtu.be/CtWGVX700J4?t=210) — [H2] Transparency enables AI organizational context - Key idea: Well-documented inner source code provides the foundation for AI agents to learn from - Evidence: "That same level of high-quality documentation and accessibility is critical for building up organizational context for LLMs" #### ⏱️ [04:15](https://youtu.be/CtWGVX700J4?t=255) — [H3] AWS CDK constructs library example - Key idea: Internal libraries need special AI integration since they're not publicly available - Evidence: "This library is made up of internal Booz Allen IP. So an AI assisted coding tool like cursor wouldn't know about it" #### ⏱️ [05:00](https://youtu.be/CtWGVX700J4?t=300) — [H2] MCP server solution for discovery - Key idea: Model Context Protocol servers enable AI tools to access internal documentation - Evidence: "The team added an MCP server for the construct library so that developers can install it in their AI assisted coding tool" #### ⏱️ [05:45](https://youtu.be/CtWGVX700J4?t=345) — [H3] Auto-generated documentation approach - Key idea: Documentation automation ensures AI context stays current with code changes - Evidence: "Documentation that powers the MCP server is autogenerated based on comments which are enforced by a linter" #### ⏱️ [06:30](https://youtu.be/CtWGVX700J4?t=390) — [H1] Community-driven development importance - Key idea: AI solutions benefit from collective organizational feedback and contributions - Evidence: "Code underlying AI solutions will better reflect the collective ingenuity of your organization when it's sharpened by feedback" #### ⏱️ [07:15](https://youtu.be/CtWGVX700J4?t=435) — [H2] AI as contribution accelerator - Key idea: AI tools can handle routine contribution tasks, freeing developers for creative work - Evidence: "AI agents helping with some of the drudgery... helping to write unit tests, generate documentation or fix linting errors" #### ⏱️ [08:00](https://youtu.be/CtWGVX700J4?t=480) — [H3] Agent studio marketplace concept - Key idea: Applying inner source principles to agent development through shared templates and examples - Evidence: "Agent marketplace where you'll find templates and boilerplate code for how to start building agents" #### ⏱️ [09:15](https://youtu.be/CtWGVX700J4?t=555) — [H1] Collaboration enhancement through AI - Key idea: AI elevates human collaboration, especially for cross-functional team handoffs - Evidence: "AI can elevate collaboration especially when it comes to handoffs between cross-functional teams" #### ⏱️ [10:00](https://youtu.be/CtWGVX700J4?t=600) — [H2] Design-to-code automation example - Key idea: Inner sourced design systems with AI integration automate designer-developer handoffs - Evidence: "Developers can now point to a design and based on Copilot's knowledge of the component library, it can autogenerate front-end code" #### ⏱️ [10:45](https://youtu.be/CtWGVX700J4?t=645) — [H3] Multiplication effect across teams - Key idea: Inner sourced AI acceleration benefits scale across multiple autonomous teams - Evidence: "The acceleration that AI is providing here gets multiplied across all of those teams" #### ⏱️ [11:30](https://youtu.be/CtWGVX700J4?t=690) — [H1] Conclusion: AI complements inner source - Key idea: AI and inner source together combat non-deterministic behavior in mission-critical systems - Evidence: "The combination of AI tools and governed proven inner source solutions helps to combat some of the non-deterministic behavior of AI" ## 📚 References - [InnerSource Commons](https://innersourcecommons.org/) - [AWS CDK Documentation](https://docs.aws.amazon.com/cdk/latest/guide/) - [Model Context Protocol](https://modelcontextprotocol.io/) ## 🗣️ Paraphrase - AI doesn't replace inner source but changes how we approach organizational code reuse and collaboration - The same documentation and governance that make inner source successful also enable AI agents to understand company-specific solutions - Organizations can apply inner source principles to AI development itself, creating marketplaces for agents and shared development patterns ## 📰 Newsletter Exciting developments at the intersection of AI and inner source! Lizzie Celita from Booz Allen shared compelling examples of how AI can supercharge inner source practices rather than replace them. Key innovations include MCP servers that give AI coding assistants access to internal libraries, agent marketplaces that apply open source principles to AI development, and automated design-to-code workflows that scale across teams. For WordPress developers, this suggests opportunities to enhance plugin and theme development through better documentation for AI tools, shared agent templates for common WordPress tasks, and improved handoffs between designers and developers. Resources: [InnerSource Commons](https://innersourcecommons.org/), [Model Context Protocol](https://modelcontextprotocol.io/), [WordPress Developer Resources](https://developer.wordpress.org/) ## ✅ Action Items #### DevRel - Developer advocacy: Showcase AI-enhanced inner source workflows at conferences and meetups - Education: Create workshops on integrating AI tools with internal documentation and libraries - Developer experience: Evaluate MCP server opportunities for internal WordPress development tools - Developer marketing: Content series on AI + inner source best practices and case studies - Community management: Facilitate discussions on AI ethics and governance in developer communities #### OSPO - Policy & Governance: Develop policies for AI-generated code contributions and licensing compliance - License Compliance: Update scanning tools to detect AI-generated code patterns and attribution - Security Management: Establish guidelines for AI tool access to proprietary code and documentation - Software Inventory: Include AI-generated components and agent dependencies in SBOM tracking - Developer Support: Training on secure AI-assisted development and inner source contribution workflows - Community Engagement: Fund research on AI impact on open source sustainability and maintainer burden - Strategic Alignment: Define KPIs for AI-enhanced productivity while maintaining code quality and security #### Community - Maintainers: Develop guidelines for accepting AI-assisted contributions and ensuring human oversight - Contributors: Create onboarding materials that explain how to leverage AI tools while following project standards - Sponsors: Transparent reporting on AI tool costs and productivity impacts from funded development - Ecosystem: Cross-project coordination on AI governance standards and shared agent templates #### WordPress - Core: Evaluate AI-assisted development workflows for WordPress core contribution processes - Core Performance: Develop AI agents that can suggest performance optimizations based on WordPress best practices - Core AI: Create governance frameworks for AI feature integration and user privacy protection - Documentation: Enhance DevHub and HelpHub documentation to be more accessible to AI agents and tools - Community: Develop guidelines for AI use in WordCamp presentations and workshop content - Training: Create learning paths that teach AI-assisted WordPress development alongside traditional skills - Plugins: Establish review criteria for plugins that incorporate AI functionality or were developed with AI assistance - Themes: Guidelines for AI-generated theme code and design asset attribution requirements #### The WPCC - Programs: Fund fellowships focused on AI ethics and governance in WordPress development - Projects: Support documentation sprints that make WordPress knowledge more accessible to AI tools - Funding: Matched grants for projects that develop responsible AI integration patterns for the WordPress ecosystem #### FAIR - Technical independence: Ensure AI development tools don't create vendor lock-in or proprietary dependencies - Policies: Develop funding transparency requirements for AI-assisted project development - Governance artifacts: Create templates for AI governance policies that communities can adapt ## 🧪 Hypotheses - Organizations with strong inner source practices will adapt to AI development tools more successfully than those without - AI-assisted development will increase the importance of code documentation and governance rather than diminish it - Companies that apply inner source principles to AI agent development will achieve better organizational AI adoption - The combination of AI tools and inner source practices will reduce time-to-market while improving code quality and security ### 📊 Methodology - Study design to test hypothesis 1: Survey organizations with varying inner source maturity levels about AI adoption success metrics - Data sources and analysis for hypothesis 2: Analyze documentation quality and governance overhead before/after AI tool adoption - Experiments to validate hypothesis 3: Compare AI agent adoption rates between organizations with vs without agent marketplaces ### ❓ Questions - How can organizations measure the ROI of AI-enhanced inner source practices versus traditional approaches? - What governance frameworks are needed to ensure AI-generated contributions maintain the quality and security standards of inner source projects? - How do we balance the speed benefits of AI-generated code with the collaborative learning benefits of traditional inner source contribution? ### 🔮 Future Research - Long-term impact of AI-assisted development on developer skill development and career progression - Optimal integration patterns between AI coding assistants and internal developer platforms - Sustainability models for maintaining AI agent marketplaces and MCP server catalogs ### 💡 Implications - Practice: Development teams need new workflows that combine AI assistance with peer review and organizational standards - Policy: Organizations require governance frameworks for AI tool access to proprietary code and contribution guidelines - Security: Enhanced focus on AI-generated code review and supply chain security for AI development dependencies - Education: Developer training must evolve to include responsible AI use alongside traditional programming and collaboration skills - Funding: Budget allocation needed for AI tooling, agent development platforms, and enhanced documentation infrastructure ### 🧠 Learning Styles Assessment Reading level estimate: Flesch-Kincaid Grade Level 12 #### 👀 Visual learners - Present: Slide screenshots and framework diagrams referenced in talk - Improve: Architecture diagrams showing MCP server integration, agent marketplace workflows, design-to-code pipelines #### 🔊 Aural learners - Present: Original video presentation with clear narration - Improve: Podcast-style discussion with Q&A about implementation challenges #### 🗣️ Verbal learners - Present: Clear definitions of MCP servers, agent studios, and inner source principles - Improve: More analogies comparing traditional vs AI-enhanced development workflows #### 🧩 Kinesthetic learners - Present: Specific implementation examples from Booz Allen - Improve: Step-by-step setup guides for MCP servers, hands-on agent development labs #### 🧠 Logical learners - Present: Three-principle framework with clear cause-and-effect relationships - Improve: Decision trees for when to use AI vs inner source, metrics for measuring success #### 🧑‍🤝‍🧑 Social learners - Present: Conference presentation format with community discussion opportunities - Improve: Facilitated workshops, peer implementation study groups #### 🧘‍♀️ Solitary learners - Present: Individual reflection prompts on AI vs inner source trade-offs - Improve: Self-assessment tools for organizational readiness, personal learning paths #### 🌿 Naturalistic learners - Present: Ecosystem thinking about organizational development practices - Improve: Sustainability considerations for AI tool adoption, environmental impact analysis ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear heading structure in notes, descriptive timestamp links - Improve: Enhanced alt text for any referenced diagrams, ARIA landmarks for content sections #### Keyboard navigation - Present: All timestamp links keyboard accessible - Improve: Skip links to major sections, keyboard shortcuts for video navigation #### Contrast and readability - Present: Standard text meets 4.5:1 contrast requirements - Improve: Ensure any embedded visuals meet contrast standards #### Text resizing and reflow - Present: Content reflows appropriately at 200% zoom - Improve: Verify all embedded content remains accessible when resized #### Media and captions - Present: Video source available with automated captions - Improve: Professional captions, downloadable transcript, chapter markers #### Forms and interactions - Present: Standard link interactions follow accessibility guidelines - Improve: Enhanced focus indicators, accessible video controls #### Standards reference - --- ## Full Transcript [00:06] So, it's kind of a strange time to be an [00:08] inner sourcing enthusiast, right? [00:11] Because in the past, the inner source [00:14] journey has looked something like this. [00:16] I have to go find a solution that works [00:20] for my use case, read the documentation [00:23] and figure out how I can leverage it, [00:26] likely contribute back to that solution [00:28] so that it meets my needs. Probably talk [00:31] to the maintainers along the way and [00:34] then stay up to date with any changes to [00:36] the solution that I adopted. [00:39] But now the equation has changed. I can [00:42] just ask AI to build exactly what I [00:45] need. Give it my requirements, get some [00:48] real-time suggestions, [00:50] and get a solution that I maintain full [00:53] control over. That seems a lot faster [00:56] and easier than the journey on the left. [01:01] So, has AI usurped inner source? [01:05] Or are the problems that inner sourcing [01:07] solves exacerbated by AI? [01:11] In the past, that inner sourcing journey [01:12] was worth it because I was saving time [01:16] by reusing something that another team [01:18] had built. They had done the heavy [01:20] lifting and I wasn't wasting time, [01:23] money, or resources reinventing the [01:25] wheel. With AI assisted coding, [01:29] we may be spending less money, time, and [01:32] resources, but we risk developing those [01:35] same silos. [01:38] The answers to these questions are [01:41] paramount at Booze Allen, where we're [01:43] building advanced technology and AI [01:46] solutions for federal government [01:48] clients. And in the current climate, [01:51] efficiency and cost savings are key. I'm [01:54] Lizzie Celita. I'm a senior software [01:57] engineer and developer experience [01:59] strategist. I'm currently in a customer [02:02] success engineering role for Booze [02:04] Allen's platform engineering product [02:06] team and I've also worked on our [02:09] internal developer portal and some of [02:12] our early inner sourced projects in the [02:14] past. [02:17] So yes, sometimes it will be faster and [02:20] easier to ask an LLM to build exactly [02:22] what you need, [02:24] but each step of the inner sourcing [02:27] journey relates to a principle that's [02:30] healthy for our organizations. And if we [02:33] give up on inner source, we risk losing [02:36] ground on the things that make our [02:37] organizations innovative and resilient. [02:42] Transparency powers the discovery and [02:45] adoption phase. [02:48] Communitydriven development sharpens [02:51] those reusable solutions through [02:52] contributions. [02:55] And the collaboration between adopters, [02:58] maintainer and maintainers underpins the [03:01] ecosystem. [03:04] Today we'll look at each of these three [03:06] areas and consider one how these [03:09] principles can actually help in the [03:13] development of AI solutions themselves [03:17] and two how AI can actually be a [03:19] superpower and accelerate the inner [03:22] sourcing journey. We'll also look at [03:24] examples of this at Boo Allen where [03:27] we're embracing the intersection between [03:29] AI and inner source. [03:33] In the transparency, adoption and [03:35] discovery phase, [03:38] well doumented and discoverable code is [03:41] the norm in inner source. You may even [03:45] have a catalog that's pointing to vetted [03:48] and governed solutions. And that same [03:52] level of highquality documentation and [03:54] accessibility is critical for building [03:57] up a organizational context for LLMs to [04:02] be able to learn from it and agents to [04:05] interact with it. If like me you dream [04:08] of an a secure all- knowing chatbot that [04:12] understands how your company builds [04:14] software, what libraries are available [04:16] and how to integrate with them. the same [04:20] transparency and accessibility [04:23] uh is necessary to have that knowledge [04:26] written down so that agents can interact [04:29] with it and LLM can learn from it. On [04:32] the flip side, how can AI help with the [04:36] discovery and adoption of those reusable [04:39] solutions? [04:41] One example of this at Booze Allen is [04:43] our platform engineering team has an AWS [04:47] cloud development kit constructs [04:49] library. [04:50] You can think of the AWS CDK constructs [04:54] as Lego blocks for cloud infrastructure. [04:57] And this library is inner sourced [04:59] because it's intended to be used by [05:01] multiple teams including the core [05:04] platform engineering team for multiple [05:07] instantiations of the platform and also [05:10] for tenants of the platform itself. [05:14] And this library is made up of internal [05:18] booze Allen IP. So an AI assisted coding [05:21] tool like cursor wouldn't know about it [05:24] because you can't find information about [05:26] this on say Google or Stack Overflow. [05:29] So the team added an MCP server for the [05:33] construct library so that developers can [05:37] install it in their AI assisted coding [05:39] tool and then cursor or whatever their [05:43] tool of choice is has access to what [05:46] constructs are available and how to use [05:48] them. [05:50] A couple of things to note about how [05:52] this works. The documentation that [05:55] powers the MCP server is autogenerated [05:58] based on comments which are enforced by [06:00] a llinter. So that ensures that the [06:02] documentation stays up to date when new [06:05] constructs are added or existing ones [06:07] are modified. And we also chose to [06:10] create an MCP server rather than just [06:13] giving AI assisted coding tools access [06:16] to the documentation itself because the [06:18] documentation is already over 10,000 [06:21] lines of code long. [06:23] Another caveat is that a developer would [06:26] have to know that the constructs library [06:29] and this MCP server exists in the first [06:32] place in order to install it in their [06:34] coding agent and take advantage of it. [06:37] Which leads me to our next example. [06:42] The second principle is communitydriven [06:44] development. And I believe that the code [06:48] underlying AI solutions will better [06:51] reflect the collective ingenuity of your [06:54] organization when it's sharpened by [06:57] feedback and contributions from [07:00] developers outside just the team that's [07:02] developing it. [07:04] We want to reduce risk by concentrating [07:08] our efforts around security and [07:10] compliance on a reusable solution so [07:14] that we're not having to go and update [07:18] uh generic AI generated solutions in [07:20] multiple places. So I believe that the [07:23] principles of communitydriven [07:26] development are even more important in [07:28] an AI enabled world. [07:31] How can AI be a superpower for [07:34] communitydriven development? You can [07:37] think about AI agents and tools helping [07:39] with some of the drudgery and [07:41] non-creative elements of contributing. [07:44] Doing things like helping to write unit [07:47] tests, generate documentation or fix [07:49] linting errors, the things that aren't [07:52] core to the value of a contribution, but [07:55] can take hours of maintainers and [07:57] contributors time. [07:59] So, where do these agents and other [08:02] agents that your company may be [08:04] developing come from? [08:07] One thing we're developing at Boo Allen [08:09] is an agent studio. I love this example [08:12] because we're applying the concepts of [08:14] inner source to the development of [08:17] agents themselves. There's an agent [08:20] marketplace where you'll find templates [08:22] and boilerplate code for how to start [08:25] building agents and some examples that [08:28] are already pre-trained for specific [08:30] purposes. [08:32] Similar to a micros service or a UI [08:34] component, we're enabling the discovery, [08:38] adoption, and contribution to different [08:40] elements of agent orchestration [08:44] and single agent construction. In many [08:47] cases, multiple agents will be necessary [08:49] for a use case. So, it's helpful to have [08:52] standardized documentation [08:54] and some baselines around how agents can [08:58] talk to each other. [09:00] Going back to our previous example, one [09:02] of the elements is an MCP server [09:05] catalog. So this is where a developer [09:07] could find and discover the m uh AWS [09:12] construct library MCP server and start [09:15] building an agent that could integrate [09:16] with it. [09:18] The agent studio is helping us to scale [09:22] knowledge in the rapidly evolving field [09:25] of agentic AI and upskill our engineers. [09:31] The final principle I'll cover today is [09:34] collaboration. [09:36] and developers talking to each other [09:39] builds trust helps improve job [09:42] satisfaction [09:44] and create a sense of belonging that AI [09:47] cannot replace. However, I do believe [09:49] that AI can elevate collaboration [09:54] especially when it comes to handoffs [09:56] between crossf functional teams. [10:00] One program that we work on at Booze [10:02] Allen has an inner sourced design kit [10:05] and corresponding UI component library. [10:09] These were created to help facilitate [10:11] the handoff between designers and [10:14] developers uh and make that handoff [10:16] process more seamless. And the team that [10:19] maintains the inner source component [10:22] library added two elements recently. Um, [10:26] one is a Figma MCP server which allows [10:29] developers to pull uh Figma design [10:32] elements directly into their code editor [10:35] and another is a set of specific GitHub [10:38] copilot instructions which added the [10:41] component libraries TypeScript [10:44] TypeScript definitions to the context in [10:47] C-pilot. By doing this, developers can [10:50] now point to a design and based on [10:55] Copilot's knowledge of the component [10:57] library, it can autogenerate front-end [11:00] code um based on the components and that [11:03] design it was given, making that handoff [11:05] basically automatic. [11:08] And this would be really cool and [11:10] provide a lot of acceleration for just [11:12] one development team. But the fact that [11:15] these libraries were inner sourced and [11:18] used across lots of smaller autonomous [11:20] development teams on the program means [11:23] that the acceleration that AI is [11:25] providing here gets multiplied across [11:29] all of those teams. [11:31] So as we've seen in the examples, the [11:34] combination of AI tools and governed [11:39] proven inner source solutions helps to [11:42] combat some of the non-deterministic [11:45] behavior of AI which is critical when [11:49] developing missionritical systems. [11:54] So back to the original question [11:57] is AI usurping inner source? I hope I've [12:00] convinced you the answer is no. I know [12:03] I'm preaching to the choir a little bit [12:05] here and I couldn't resist uh using AI [12:08] to generate our inner source hero uh [12:12] which you'll see on the screen. [12:15] But I just would like to say in [12:18] conclusion, inner source, it encourages [12:20] us to stay curious about what other [12:23] teams are building, to contribute back [12:26] to what other teams are building, to [12:28] continually make it better, and to build [12:31] bridges and connections between [12:33] development teams and individuals. On [12:37] that note, I'd like to thank uh some of [12:40] my teammates who provided the examples [12:43] that you saw today. Steven Toronto, [12:45] Jennifer Shepard, and Connor Youngland [12:48] and their teams. And thank you, and I [12:51] look forward to connecting throughout [12:52] the rest of the summit.