## Overview The meeting focused on reviewing progress and deliverables for the reference architectures and critical use cases work streams within the Agentic AI Foundation. Yaron highlighted the importance of monthly reports and mentioned a potential policy against AI-generated contributions. Mario presented the current state of the reference architectures, including principles, patterns, and two architectures: single agent processing with human approval and bounded autonomous remediation. The group discussed the need to align terminology, such as replacing "task" with "activity" to match the accepted taxonomy. Jodee updated the group on the critical use cases work, outlining the schema and inventory, and seeking feedback on the approach and evidentiary standards. Sumanth provided an update on taxonomy integration and alignment with other work streams. Govindaraj raised the need for better cross-working group collaboration, particularly with security and privacy, and discussed the challenge of inconsistent meeting notes across groups. The conversation ended with a plan to review the discussed resources for upcoming votes. ## Key Topics ### Monthly Reports Status Update Yaron updated the team on monthly reports being written to the TC, explaining that these reports contain summaries of work done each month and are stored in the GitHub repository's reports directory. He noted that while the TC is reviewing these reports, both July and August reports are currently in PR stage. The meeting began with casual greetings and time zone discussions before transitioning to the report status update. ### Reference Architecture Review Discussion Yaron and Sumanth discussed reviewing a document, with Yaron requesting Sumanth's GitHub handle for review purposes. The team reviewed progress on reference architectures, including principles, contributing guidelines, and two current architectures: single agent processing with human approval and bounded autonomous remediation. Yaron informed the team about potential new policies regarding AI contributions that may restrict or limit automated contributions to the foundation's work. ### Reference Architecture Patterns Discussion Mario explained the reference architecture patterns, describing how they are composed of reusable solutions to specific problems, with examples including bounded convergence loop, deterministic acceptance gate, and human approval gate. Yaron questioned whether the term "task" should be replaced with "activity" throughout the documentation, which Mario agreed to review. Laxmikant sought clarification on how patterns differ from architecture folders, and Mario explained that patterns solve specific problems independently while reference architectures are compositions of patterns to address larger problem classes. ### Validation Records for Architectures Mario explained the purpose of validation records for reference architectures, which list use cases covered by each architecture to demonstrate practical effectiveness beyond theoretical documentation. He noted that Ivana and Daniel contributed two open pull requests with additional validation records and information about their testing processes. Yaron confirmed one of the pull requests had been merged and suggested focusing on the single human approval reference architecture as the first deliverable for voting. ### Use Case Voting Discussion Yaron announced that the team will vote on specific changes during the September meeting, with the second meeting serving as the milestone for this decision. Jodee presented work on cleaning up use case sources and proposed voting on three key aspects: the schema approach, the use case taxonomy verticals, and the evidentiary standards for use cases. The team discussed reviewing the use case inventory and schema documentation before calling for a vote, with Jodee agreeing to share the relevant links on Discord for team review. ### Resource Review Progress Update Yaron assigned Mario and Zane to review the patterns section and Sumanth to review taxonomy alignment. The team discussed progress on three resources requiring review: the use case inventory, landscape for critical use cases, and single agent approval gate for reference architectures. Jodee's team is cleaning up references in the inventory table to remove extraneous business value information and ensure only live links to actual deployments are included. Sumanth provided an update on taxonomy work, noting that their initial PR was sent to the main taxonomy workstream and will be polished with new definitions following agreed guidelines for scope and notes. ### Project Architecture and Use Cases Sumanth discussed reviewing definitions, architect reference architecture, and use cases for the project. Jodee requested feedback on use cases and additional references. Govindaraj mentioned a request for creating a security and privacy cross-working group to address interdependencies between workgroups. ### AI Foundation Collaboration Initiatives Govindaraj raised concerns about siloed approaches across working groups affecting deliverables and the need for consistent messaging as a combined AI Foundation. Yaron supported the initiative but emphasized that cross-working group collaboration should be established through formal chair synchronization meetings rather than ad hoc participation. They agreed to work with Alex and Manik to address the challenge of inconsistent meeting notes in GitHub by potentially automating a consolidated summary process. ## Next Steps - Sumanth: Review the new PR for taxonomy and provide comments; inform Yaron if it needs to be closed. - Mario: Go through the reference architecture documents to check and update the usage of "task" to "activity" as per the accepted taxonomy. - Mario and Zane: Review the patterns section of the use case schema to ensure alignment with the reference architectures. - Sumanth: Review the use case schema and inventory to ensure taxonomy alignment. - Jodee: Clean up references and remove extraneous business value information from the use case inventory table. - Yaron: Put the reference for the single agent approval reference architecture on Discord and in meeting notes for review and upcoming vote. - Yaron: Touch base with Alex (and possibly Manik) regarding consolidating meeting notes into a single source of truth in GitHub. - All attendees: Review the single agent approval reference architecture, use case inventory, and use case landscape for upcoming votes. - Sumanth: Comment on the new taxonomy PR with agreed-upon definition structure (one to two terms with scope and note). - Sumanth: Look into how taxonomy can be incorporated into reference architectures and use cases.55:T1684,## Overview This meeting was a regular working group session where Steven presented the Agent Action Capsule concept, a standards-based approach for creating immutable records of agentic actions that can be stored in append-only ledgers and verified by third parties. Steven demonstrated the open-source implementation and explained how it addresses the need for trustable evidence of agent behavior, particularly for compliance and governance purposes. Sean introduced his nonprofit's legal analysis engine (LexLint.org) which compiles global AI and privacy laws, suggesting potential integration with the capsule system to help organizations understand legal requirements across different jurisdictions. Neil expressed interest in the concept, noting similarities to a project called Notary, and discussed the importance of jurisdictional tracing for legal compliance. The group discussed how the capsule system could help bridge the gap between technical logging and legal requirements, with participants agreeing to follow up offline to explore potential collaborations and integration opportunities. ## Key Topics ### Informal Team Location Discussion The meeting began with casual conversation as participants joined, including Steven who was located on a boat in San Francisco Marina. The group discussed locations, with Adrian in DC and others joining from various places. The conversation was informal and social in nature, with participants sharing where they were located and discussing topics like remote work possibilities and Starlink technology. The meeting appeared to be starting late, with some uncertainty about who was running the show. ### AI Legal Compliance Tool Discussion The team discussed confusion about whether today's meeting was canceled, with several members including Thilaknath and Sean clarifying that this was the regular bi-weekly meeting. Sean introduced himself as a new participant and shared information about his US-based software nonprofit's legal analysis engine, which he had opened up as a free tool at lexlint.org to help developers identify potential legal compliance issues with their AI software. The discussion focused on how the tool helps address legal requirements across different jurisdictions, particularly for AI development and government transparency. ### Global AI Regulations Database Sean discussed his organization's work on collecting and analyzing global AI regulations, noting that their current database represents only about 1% of existing laws across approximately 600 jurisdictions. He offered to help the working group expand their legal coverage beyond the current spreadsheet approach. Steven then shifted the meeting focus to present Agent Action Capsule, an open-source solution for creating trustable records of agent actions that can be stored in a shared ledger. The solution aims to establish a standard format for recording what happened, what was verified, and what was blocked, with specifications donated to the IETF and demonstrated through interop testing at various hackathons including one in Vienna with seven different teams. ### MeshLLM Project Overview Steven presented MeshLLM, a project from Block that leverages unused compute resources for inference and includes trust-related components built with agents. He explained how capsules record agent actions and can be mapped to policy obligations in a ledger system, providing verifiable proof of proper agent behavior. The system uses SCIT RFC9943 standards and Merkle Mountain Range technology for secure checkpointing, with all code being open source and unencumbered. Neil expressed interest in the project, comparing it to work by Brandon Bennett, and John inquired about TDX certificates and the scope of the policy engine. ### AI Charter and GRC Challenges Steven discussed how the AIF charter aims to accelerate industry progress, particularly in GRC areas where evidence gathering can be challenging. Sean highlighted that much new AI law focuses on higher-level issues like disclosure rather than individual transactions, and raised questions about how to determine applicable jurisdictions when making agentic requests across different geographic locations. The discussion touched on the importance of logging agentic actions and the challenges of tracking what actions were taken and whether proper checks were performed before execution. ### Software Traceability Implementation Discussion The discussion focused on implementing jurisdictional traceability for software systems, particularly regarding user location tracking for regulatory compliance. Steven explained how his team developed an "Action Capsule" profile that works with RFC 9943 standards to create tamper-evident audit trails, which can be used to determine where users are located and ensure regulatory compliance. Neil expressed interest in connecting offline to discuss how this approach fits within broader governance systems and agent containment efforts, while John mentioned his work with a governance working group and expressed interest in potentially combining efforts with Steven's team. ## Next Steps - Steven: Follow up with Sean to check out lexlint.org and discuss how its database and solution can help with the spreadsheet of regulations and acts. - Neil: Review the Agent Action Capsule materials and follow up with Steven to discuss further. - John: Reach out to Steven to compare and discuss combining the governance protocol from Spellgarden with the Agent Action Capsule work. - Steven: Send a note to John with the IETF papers and working group information related to the Agent Action Capsule. - Steven: Follow up with Neil and John separately to discuss their respective interests and potential collaborations.56:T549,This is a focus group within the AAIF Observability & Traceability Working Group. This focus group covers the surfaces through which agent intent becomes an observable action: tools, skills, image ingestion, reasoning tokens and other agent-facing primitives exposed by models or frameworks. - Tool, skill, and primitive execution — structured representation of what the agent invoked, including tool use, skill execution, structured actions, and modality-specific operations that shape agent behavior. - Intent, outcome, and side-effect attribution — what the agent requested, what the execution surface returned, and what externally visible state changes resulted. - Retry and fallback. - Skill selection, provenance, and composition — which capability was chosen, what alternatives existed, where it came from, and how sub-skills, tools, or other primitives were composed into a larger behavior. - Skill-level impact — attributing outcomes, costs, and side effects to the composed capability level. Regarding protocols, protocols such as MCP, A2A, UCP, and others define how agents interact with tools, resources, and each other, and they create the boundaries across which trace data must remain coherent. This focus group covers both protocol-native observability and the conventions needed to make agent trace data portable across systems.57:T16c1,## Overview The team met to discuss progress on the Best Practices Guide and feature matrix documentation. Fernando reported that most comments on his pull request had been addressed, with only two remaining items related to adding links that would be handled as separate pull requests after official commitment to the repository. The group discussed a suggestion from an external contributor (Narco from Empire Labs in Australia) to add a fourth column to validate feature presence through self-declaration, observable operations, or external establishment, though concerns were raised about the practical implementation and timing of this addition. Erin presented her work on the feature matrix PR, which was generated using AI assistance and needs cleanup to remove duplicates before further review. The team also discussed linking to external resources in the documentation and the possibility of reducing future meeting times to 30 minutes while maintaining the current one-hour blocked time slot. ## Key Topics ### Best Practices Guide Update Fernando provided an update on the Best Practices Guide pull request, stating that most changes have been addressed and he is waiting for initial commenters to validate if any further modifications are needed. He indicated that if no additional comments are received within the next week, the changes will be committed and officially added to the repository. ### GitHub Pull Request Progress Update Fernando reported significant progress on reviewing GitHub pull request comments, addressing most suggestions except for two link-related items that will be handled after official repository commitment. He also mentioned starting work on the threat modeling document. Samantha agreed to complete her final review of the pull request, with both parties aligning to commit the current version first before addressing the remaining link references as separate pull requests. ### Time-Based Commit Strategy Discussion Matthew expressed excitement about transitioning to time-based commits instead of feature-based commits, noting this approach makes more sense after completing the first pass. The team discussed potentially applying this concept to future PRs with one-month windows for each phase. Fernando mentioned he couldn't attend the Identity and Trust meeting to discuss Best Practices, and asked Matthew if he had any discussion with that team on the topic. ### Agent Action Captive Protocol Discussion The group discussed a recent meeting where Matthew Khouzam had attended presentations, including one by Stephen Mee about the Agent Action Captive protocol. There were technical difficulties with Erin's audio connection during the discussion. The conversation ended with Erin mentioning she wanted to discuss PR work she had done on the feature matrix, though this topic was not fully covered due to the meeting's conclusion. ### Document Feedback and Updates Erin discussed feedback from a colleague at Empire Labs regarding a document they are working on. The feedback suggested adding a fourth column to validate the presence of features, with examples of self-declared implementation versus observable operation. Fernando clarified that the numbering system should be maintained even when practices are deprecated, with a potential additional column to track superseding categories. Erin agreed to clean up the document before implementing the suggested changes. ### Feature Implementation Status Tracking The team discussed adding an additional column to track feature implementation status, with options including self-declared, observable in operation, or established outside of deployment. Erin expressed some concerns about the scope and potential overlap between different tracking methods, while Fernando and Matthew Khouzam suggested breaking this into separate PRs to manage scope more effectively. The team agreed to move the discussion to Discord to better understand the intended purpose and implementation goals of the additional column. ### Documentation and PR Review Updates The team discussed progress on documentation and PR reviews. Erin and Fernando agreed to wait a week before reviewing Aaron's PR to avoid duplicate feedback, and Fernando suggested running a new conversation with the AI to incorporate additional HOA content. The team also discussed adding links and tooltips to documentation, with Samantha providing guidance on implementing markdown links. Fernando proposed reducing future meeting time to 30 minutes but decided to keep the current hour-long format after team input. ## Next Steps - Fernando: Wait for final validation from initial commenters on the Best Practices Guide PR (#9) this week; if no additional comments, merge the PR into the repo. - Fernando: After merging PR #9, address the remaining two link-related comments via subsequent pull requests. - Fernando: Begin working on the threat modeling document (PR #9) and incorporate comments. - Samantha: Perform a final review of Fernando's Best Practices Guide PR (#9) today. - Erin: Clean up duplicates and clarify the feature matrix PR, then respond to Narco's comments (especially regarding the proposed fourth column) in GitHub. - Erin: Reach out to Narco (Empire Labs) via GitHub/Discord to discuss the proposed fourth column and clarify intent. - Erin: Update the feature matrix to link to specific sections of Fernando's Best Practices Guide and consider adding tooltips for clarification. - Erin: Rerun the AI-generated content for the feature matrix, incorporating Fernando's newly added section on HOA outside of MCP. - Bar: Review Fernando's Best Practices Guide PR (#9) and Erin's feature matrix PR this week. - Fernando: Decide on scheduling for a separate threat modeling meeting (either reduce this meeting to 30 mins or find another slot).58:T1c59,## Overview The meeting focused on reviewing and merging PRs for taxonomy terms, discussing the process for approvals, and handling contributions. Julianna updated the group on the identity and trust terms PR, and the group agreed that two approvals from recognized delegates were sufficient for merging, with a preference for asynchronous GitHub reviews over in-meeting votes. They discussed whether to have one PR per term or multiple terms per PR, leaning towards one PR per definition for clarity but allowing batches for non-contentious terms. The group merged Joe's "Meta Harness" PR after addressing comments. They decided to punt on the Landscape work to focus on Taxonomy, noting the complexity and potential for noise from unsolicited AI-generated PRs. The group expressed concerns about contributions from unknown contributors, especially AI-generated ones, and discussed the need for clearer contributor guidelines and expectations for engagement. Julianna created a "ready for review" milestone to manage and filter PRs. The meeting also touched on the need to address terms marked as "debate" and the volume of new terms in the spreadsheet. ## Key Topics ### Web Server Progress Update Matthew shared that he created a web server to help with JavaScript editing and content generation, similar to an MCP approach. Junjie noted that several PRs were submitted since the last meeting, indicating good progress. Gala mentioned they couldn't attend the chair-co-chair meeting due to scheduling conflicts. ### PR Approval Process Updates The team discussed the process for reviewing and approving PRs for term definitions. They agreed that having 2-3 approvals from recognized committers should be sufficient to merge a PR, rather than requiring a formal vote during meetings. The group also debated whether to create one PR for multiple terms or separate PRs for individual terms, deciding to test both approaches with specific examples. Junjie agreed to follow up with Christina about creating a GitHub delegates group to streamline the approval process. ### PR Review Process Discussion The team discussed PR review processes and approval requirements. They agreed that comments should be addressed before merging, with Julianna committing to resolve existing comments on PR 40. For PR 41, which defines multiple taxonomy terms including some from security and privacy, Julianna decided to seek additional review from relevant team delegates before proceeding. The group also discussed whether to use one commit per definition versus multiple definitions per PR, ultimately agreeing that user judgment would be appropriate for handling batches of terms. ### LGTN Merge Discussion Meeting The team discussed merging LGTN and confirmed there were no outstanding issues, with all comments addressed. They agreed to proceed with merging the pull request, with Junjie having the merge button. Matthew expressed a preference for using rebase-only strategy for future projects, though acknowledged this wasn't critical for the current deployment which doesn't affect running services. Julianna mentioned there were additional PRs for terms that needed review, including security and privacy terms, and noted an outstanding question about landscape. ### Landscape Contributions Definition Discussion Julianna raised questions about updating and defining the purpose of landscape contributions, noting challenges in tracking contributions from external contributors like Narco. Gala and Leonardo highlighted that landscape work has been postponed due to uncertainty about desired granularity and selection criteria. Junjie explained that the landscape work should focus on high-level market availability of Agentic AI products, similar to the taxonomy work, and suggested starting with basic content to demonstrate the portal's functionality while remaining open for community contributions. ### Landscape Project Postponement Decision The team discussed the Landscape project and decided to punt on it for now, focusing instead on completing the Taxonomy work first. They identified concerns about governance and scope, noting that CNCF's landscape contains many more entries than their current four or six projects. The group agreed to communicate this decision to the technical committee and keep Landscape on the roadmap as a future initiative that will be addressed once Taxonomy is more established. ### PR Review and Term Definitions The team discussed progress on PR reviews and term definitions, with Julianna noting 29 terms awaiting review in the spreadsheet. Leonardo reported alignment with Ilya from Magento Commerce on Agentic Commerce terms and created a draft PR, though they decided to wait for an upcoming meeting before finalizing the review. The group agreed to address debate terms by establishing clear processes for resolution, with Junjie suggesting simple voting methods for less controversial terms, while acknowledging that complex terms like "agent" would require more discussion. Julianna mentioned she was working on a paper about defining agents for the Identity and Trust group, which would be published next month. ### PR Management and Contributor Guidelines The team discussed managing PRs and addressing issues with AI-generated contributions and uninvolved contributors. They decided to create a "ready for review" milestone to filter and track relevant PRs, and agreed that contributors should introduce themselves or engage asynchronously before their PRs are reviewed. Matthew volunteered to reach out to an uninvolved contributor to understand their goals. The group also noted the need to update contributor guidelines to address AI-generated content and ensure human curation of PRs. ## Next Steps - Junjie: Check with Christina about creating a GitHub group of delegates and configuring GitHub to automatically request reviews from delegates for PRs. - Julianna: Double-check and resolve comments in PR #40, then request re-approval in Discord if needed. - Julianna: Hold off on merging PR #41 until members from other relevant working groups have reviewed it, and post in Discord to request those reviews. - Julianna: Create a PR to update the onboarding/contributing documentation to clarify that Landscape is on the roadmap but not currently being worked on, and to discourage new Landscape PRs for now. - Matthew: Reach out to Narco4u to understand his goals and discuss his AI-generated PRs, coordinating with the Security working group if needed. - Junjie: Check with Chris/Christina about gaining more repo permissions to better control PRs and contributor access. - Julianna: Maintain and update the "ready for review" milestone in GitHub, adding PRs that have gone through the process and are ready for delegate review. - Junjie: Communicate to the chairs/TC that the Taxonomy group is focusing on Taxonomy first and will revisit Landscape later, and provide feedback on the impact of the repo being public. - Leonardo: Wait until after the upcoming Agentic Commerce meeting to finalize and review the terms in the draft PR for Agentic Commerce. - All delegates: Continue reviewing open PRs, especially those in the "ready for review" milestone, and leave comments, approvals, or objections.59:T1b49,## Overview The meeting focused on updates regarding cross-working group dependencies and overlaps within the AIF (Artificial Intelligence Foundation) project. Govindaraj provided updates on various working groups, including changes in the identity and trust team's focus on taxonomy reference architecture and primitives, the A2A project's integration with the AIF foster project, and the agent e-commerce team's discussion of delegated purchasing authority models. The group discussed challenges around scope creep and the need for better coordination between working groups, particularly regarding the agent router donation to AIF and security and privacy considerations. Matthew demonstrated an AI-powered tool he created for defining taxonomy terms, while the team also discussed the importance of using GitHub for serious work after initial brainstorming in Google Docs, with plans to share their working methods with other chairs. ## Key Topics ### Identity Primitives Taxonomy Updates Matthew and Govindaraj discussed updates on taxonomy and identity primitives work, with Govindaraj explaining that the identity and trust team has been reviewing reference architecture and external designs from various organizations. Govindaraj mentioned that A2A is part of the AIF foster project and noted discussions about version agent identity schema and accuracy reliability. Matthew commented that it was positive that different teams were not staying within strict boundaries in their work. ### Identity and Observability Security Discussion The team discussed identity and observability aspects affecting security, with specific focus on Agent Action Capsule (AAC) and related protocols. Govindaraj noted that the identity team has become active in working on primitives and other components, and emphasized the need to communicate these developments to other teams like workflow and process integration to avoid overlaps. The discussion also touched on Steven's work on observability and tracing, and mentioned that the best practices team is progressing well. ### Identity and Trust Initiatives Update Govindaraj discussed updates on identity and trust initiatives, including Fernando's involvement in connecting with Alper, Julian, and Grant to work on best practices. He mentioned developments in AgentAuth specifications and their potential to complement OpenID for identity and trust, as well as delegated purchasing authority models. Govindaraj also noted that the agent router has been donated to AAF and discussed the emergence of private agent registries from cloud service providers, raising a question about whether the agent router should serve as a trusted agent registry for identity and trust. ### Agent Security and Privacy Considerations Govindaraj discussed security and privacy considerations for agent registration and access control, emphasizing the need for proper authentication, authorization, and observability in the agent router. He mentioned that upcoming security and privacy group meetings would address questions related to A2A and Agent Router projects. Govindaraj stressed the importance of treating new additions as incremental updates rather than causing scope creep to existing deliverables. ### Perfect vs Progress Discussion The group discussed the importance of not letting perfection block progress, with everyone agreeing that "perfect is the enemy of good." Govindaraj mentioned not receiving feedback on a previously shared document and noted that Empire Labs Private Limited was reviewing an issue they had opened up. Samantha clarified that the Narco4U person at Empire is involved in reviewing AIF PRs, though the exact number of people involved remains unclear. ### AI-Enhanced Taxonomy Web Server Matthew demonstrated a web server program he created that adds AI functionality to the taxonomy working group's JSON, including authentication warnings for AI-generated content. The team discussed accessing GitHub issues, particularly regarding define workload and narco for you feedback. Samantha asked about Govindaraj's preferred name pronunciation, and Govindaraj confirmed either "Govi" or "Govin" was acceptable. ### Observability and Traceability Working Group The group discussed a workgroup participant's involvement in observability and traceability, with Matthew planning to meet him on Wednesday. Govindaraj explained that in the observability and tracing working group, which he leads along with Matthew, Steven, and Pavan, they focus on security through hash-based systems and Merkle trees rather than using entire data sets. Govindaraj noted that some information posted about their workgroup specifications may be inaccurate, as Steven is working on sharing updated specifications. ### Security and Governance Working Updates Govindaraj discussed ongoing work on security and privacy issues, including gap analysis and threat modeling, with Moushumi mentioned as a participant working on best practices and design patterns. Govindaraj reported limited progress from the governance groups on reviewing issue number 18 regarding governance working group scope and overlaps, as well as from the Agentic Commerce working group. Fernando acknowledged the shared document about AAF cross working group dependencies and overlap and indicated he would review it later. ### GitHub Workflow Standardization Discussion The team discussed establishing a standard workflow where everyone uses GitHub for serious work after initial brainstorming in Google Docs. Matthew explained that Google Docs serves as a low-barrier entry for initial brainstorming before moving to GitHub, and noted that managers may need GitHub content converted to PDF or Word documents. The group agreed to share their working methods with other chairs, particularly Alex, and discussed using Python scripts for document conversion and exploring MeshLLM for AI-related work. ## Next Steps - Fernando: Connect with Alper, Julian, and Grant to discuss what can be done from the Identity and Trust content perspective for the best practices guide. - Fernando: Review the AAF cross working group dependencies and overlaps document shared by Govindaraj. - Matthew: Meet with Narco4U (from Empire Labs) on Wednesday. - Govindaraj: Follow up with Bepsy on the status of the gap analysis for security and privacy working groups (issue #16). - Govindaraj: Follow up with the Governance group regarding the nomination for reviewing issue #18 (Governance working group scope and overlaps). - Govindaraj: Reach out to the Agentic Commerce working group for traction on their participation. - Fernando and Matthew: Discuss with Alex (chair) about establishing a standard workflow (e.g., using GitHub) and share it with other chairs. - Govindaraj and Matthew: Talk to Steven about using MeshLLM for AIF-related work. - All participants: Review the document and project board links shared by Govindaraj, and volunteer to take up tasks via GitHub, Discord, or email.5a:T549,This is a focus group within the AAIF Observability & Traceability Working Group. This focus group covers the surfaces through which agent intent becomes an observable action: tools, skills, image ingestion, reasoning tokens and other agent-facing primitives exposed by models or frameworks. - Tool, skill, and primitive execution — structured representation of what the agent invoked, including tool use, skill execution, structured actions, and modality-specific operations that shape agent behavior. - Intent, outcome, and side-effect attribution — what the agent requested, what the execution surface returned, and what externally visible state changes resulted. - Retry and fallback. - Skill selection, provenance, and composition — which capability was chosen, what alternatives existed, where it came from, and how sub-skills, tools, or other primitives were composed into a larger behavior. - Skill-level impact — attributing outcomes, costs, and side effects to the composed capability level. Regarding protocols, protocols such as MCP, A2A, UCP, and others define how agents interact with tools, resources, and each other, and they create the boundaries across which trace data must remain coherent. This focus group covers both protocol-native observability and the conventions needed to make agent trace data portable across systems.5b:T1977,## Overview This meeting focused on the Tracing and Skills Focus Group, where participants discussed tooling, skills, and observability in the Agentic ecosystem. Matthew raised concerns about a user named Narco4U/Eddie who made 400 GitHub contributions in two weeks using AI assistance, and the group discussed the need for guidelines around AI contributions. The team explored the distinction between skills and tools, with Manu explaining from an academic perspective that skills were designed to be precision-focused like "precision medicine" rather than general tools, emphasizing end-to-end problem solving and observability. Matthew shared his work on instrumenting MCP hooks in Theia IDE to analyze observability patterns, and the group discussed connections between this work and the Agent Communication Protocol (ACP). The conversation also touched on potential applications in financial sectors and governance, with Manu suggesting integration with security and policy frameworks could help address trust issues in deterministic verification cases. ## Key Topics ### AI Note-Taker Discussion Meeting Matthew discussed his long day of meetings with AIF people, noting it was enjoyable to work with intelligent colleagues toward shared goals. The conversation then shifted to comparing AI note-takers, with Matthew asking Manu about the effectiveness of AI note-takers versus LFX platform ones. Manu explained that his AI note-taker was automatically activated due to back-to-back meetings and agreed to remove it. The discussion ended with Matthew raising a personal issue about Narco4U, requesting Pavan's input on the matter. ### AI Contribution Guidelines Discussion Matthew reported that Eddie made approximately 400 GitHub contributions in the last two weeks, primarily using AI tools like ChatGPT to scrape and address issues. Pavan suggested that Eddie's contributions appear AI-driven and not easily comprehensible, leading to a discussion about developing guidelines for AI contributions. Matthew offered to help by potentially syncing with Eddie once a week at midnight to better understand his motivations and assist with contributions. ### Landscape Analysis Discussion The team discussed the focus group's work on tool skills and execution, specifically regarding a landscape analysis. Pavan confirmed they should take inspiration from the CNCF landscape, which contains over 300 definitions, though Matthew noted their project has only five projects compared to CNCF's extensive list. The group agreed to create a more logical and appropriate landscape analysis for their specific context rather than replicating CNCF's extensive framework. Matthew also mentioned plans to work on best practices for intent outcome and side effects attribution, though he noted he had not had time to update the related PR. ### Non-L7 Layer Capabilities Discussion Pavan proposed exploring non-L7 layer capabilities for the focus group, with Matthew expressing support for this idea. They discussed the potential for hardware-level accelerators and how these might interoperate with the existing Agentic stack at the L7 runtime level. Matthew then shifted the discussion to skills versus tools, questioning what makes skills special compared to tools and ensuring they stayed on track with the Weekly Sync objectives. ### AI Skills vs Tooling Distinction Manu shared insights from academic work at MIT's CSAIL department about the distinction between skills, tooling, and utilities in AI development. He explained that skills were originally designed to be like precision medicine with end-to-end problem-solving focus, observability, and adaptability, while tooling and utilities function more like general medicine with composability features. Matthew mentioned adding instrumentation to MCP hooks in Theia IDE to improve observability, and asked if others were working on similar efforts. ### MCP Hooks and Protocol Standards Matthew explained his work on MCP hooks in VS Code, comparing it to a hypothetical scenario where large companies like Ericsson and SAP influenced Microsoft to develop cloud-based IDE support. Pavan asked about the differences between MCP and ACP (Agent-Client Protocol), which aims to standardize communication between IDEs and coding agents. Matthew clarified that MCP and ACP are not overlapping but noted that MCP could potentially become a model-to-tool standard in the future, similar to A2A's integration with AIF. ### ACP and MCP Discussion Summary Matthew and Pavan discussed the differences between ACP (Agent Communication Protocol) and MCP, clarifying that ACP connects coding agents to editors while MCP connects tools to agents. Manu asked for elaboration on the current status of the tool framework, and Matthew explained that Eclipse Theia is an IDE with an integrated agent called Thea AI that can work with various servers and MCPs. Matthew planned to instrument both Theia and Goose to compare and contrast patterns for potential universality in their implementation. ### Agentic Adoption in Financial Sector Manu discussed challenges with Agentic adoption in the financial sector, particularly around observability, instrumentation, governance, and policy implementation. He suggested integrating security features and tool invocation with MCP (MCP server), and mentioned the potential differentiator of Eclipse Enclave and nono.sh approaches. Matthew offered to arrange demos from the Thea team for these technologies, though the discussion ended without concrete decisions or next steps being established. ## Next Steps - Matthew: Remove Manu's Fireflies AI auto-note-taker from the meetings. - Matthew: Sync up with Eddie (Narco4U) from Empire Labs once a week at midnight (Australian time) to discuss his GitHub contributions and motivations. - Matthew: Share the link to the Discord channel for the focus group. - Matthew: Share the link to the Theia IDE instrumentation project with Pavan. - Matthew: Invite the Eclipse Theia team to give demos of their IDE and integrated agent to the focus group. - Matthew: Update the PR for the tracing landscape to reflect a more focused approach. - Matthew: Develop analyses on the MCP hook instrumentation in Theia IDE to identify patterns and anti-patterns. - Matthew: Investigate the overlap and differences between MCP instrumentation and the Agent Client Protocol (ACP). - Pavan: Create a Slack channel for the focus group if needed. - Manu: Run Matthew's Theia IDE instrumentation project after the call.5c:T181e,## Overview The meeting focused on updates and discussions related to the Accuracy and Reliability Working Group's deliverables, particularly the taxonomy terminology and the gap analysis survey. Matthew presented a tool to facilitate proposing and editing terms, which tracks AI-generated content and aids in reviewing contributions. Sanjeev Suresh provided an update on the survey, which was nearly finalized and ready for merging, and the group discussed strategies to increase participation. They also talked about the need to establish communication with other working groups, mentioning the Circleback meeting as a potential forum. The group agreed to start contributing to the terminology deliverable, focusing on key terms like accuracy and reliability. Additionally, Matthew highlighted some relevant technologies from the Eclipse Foundation. The conversation ended with a plan to review the progress of the deliverables and encourage further contributions. ## Key Topics ### Taxonomy Tool and Deliverables Discussion The team discussed Matthew's tool for proposing and editing taxonomy terms, which tracks AI contribution percentages and uses PR-based reviews. They briefly discussed upcoming conferences, with some members planning to attend AgentCon in Amsterdam and the US version in San Jose. The conversation ended with updates on taxonomy deliverables, including plans to handle PRs and merge iterations on a biweekly basis, with Jordan inviting team members to express interest in bootstrapping over-deliverables on specific topics. ### Survey Promotion Strategy Discussion Sanjeev Suresh reported that he addressed feedback and polished the survey, adding welcome and thank you screens, with the text nearly final. The team discussed strategies to promote the survey, including creating a presentation slide and sending an email to increase participation. Jordan raised a question about whether people could complete the survey for multiple projects, and Sanjeev Suresh clarified that there was no restriction, as it was not a paid survey. They also considered offering early access to results as an incentive and discussed potentially organizing a virtual summit in late October to further promote the survey. ### Gap Analysis Progress Discussion The team discussed advancing the Gap analysis work while the survey is in progress, with Sanjeev Suresh suggesting they can start collecting thoughts and information without waiting for the survey results. Jordan proposed structuring the document around the lifecycle of agentic applications, including development, research, deployment, and user satisfaction frameworks. The team agreed to merge an existing PR that had been open for a long time, with Sanjeev Suresh planning to share survey edit access with Jordan and Matthew. ### Repository Access and Deliverables Review The team discussed merge access to a repository, with Jordan offering to grant Sanjeev Suresh access if needed. They reviewed deliverable one and agreed to prioritize work on accuracy and reliability definitions using Matthew's tool. The group decided to delay establishing liaisons with other working groups until more preparation work is completed on reliability and accuracy dimensions. ### Circleback Security Meeting Planning The team discussed attending a Circleback Security meeting on Monday, with Matthew offering to represent the group while wearing an observability hat. Jordan noted that attendance has been challenging and suggested having multiple representatives, including Mathieu as a potential volunteer. The discussion also covered ongoing deliverable tracks, including work on taxonomy terminology and gap analysis, with the taxonomy track needing more feedback while the gap analysis is at an intermediate stage. ### AI Contribution Guidelines Discussion Matthew raised concerns about AI-generated contributions in the Taxonomy Working Group, specifically noting that a non-member contributor appears to have used ChatGPT to generate code comments, making review difficult due to self-referential sentences and potential lack of understanding. The group discussed the challenge of reviewing such contributions and agreed that establishing guidelines for AI usage in contributions would be beneficial. Jordan added this to action points and suggested using circle back meetings to better understand working group interfaces and review deliverables. ### Technology Tools and Deliverables Review Matthew introduced two technology tools for the working group: Eclipse Enclave for more isolated VM-based sandboxes and Eclipse Theia AI with an integrated agent harness. Jordan proposed focusing on terminology contributions over the next two weeks and requested team members to suggest terms for review. The group agreed to check the status of their two deliverables at the next meeting, including integrating feedback and iterating on versions, with Sanjeev Suresh suggesting potential promotion at AsianCon to attract more participants. ## Next Steps - Matthew: Share the tool for proposing and editing taxonomy terms, and encourage others to use it for contributions. - Sanjeev Suresh: Merge the PR for the survey after addressing feedback, and share edit access to the survey with Jordan, Matthew, and others interested. - Jordan: Check and possibly grant merge access to Sanjeev Suresh for the repository, or ask Christina for help if needed. - Jordan: Draft a slide and email draft to advertise the survey, and check for any template or style guidelines from the foundation. - Jordan: Communicate about the Circleback meeting and find volunteers to attend and represent the working group. - Jordan: Highlight the terms to focus on for the taxonomy terminology deliverable and encourage contributions. - All attendees: Review the different content of deliverables and contribute feedback. - Matthew: Attend the Circleback meeting and represent the working group, wearing both the observability and accuracy hats. - Sanjeev Suresh: Prepare a slide with the survey link for possible inclusion in materials. - Working Group: Consider establishing guidelines for AI-generated contributions to ensure quality.5d:T11f1,## Overview

The meeting focused on updates and discussions from various working groups within the AIF Security and Privacy Working Group. Matthew introduced a new taxonomy browser tool to help manage AI-generated definitions and PRs. The group discussed challenges with unsolicited PRs and comments, and considered strategies to manage contributions, including potentially restricting PRs to core team members. Fernando provided updates on the threat modeling and best practices guides, noting ongoing review processes and the challenge of managing numerous comments. Govindaraj raised issues about cross-working group collaboration and the need for better representation from other groups. The group also shared news updates, including mentions of new European standardization efforts related to AI cybersecurity and potential connections with other organizations.

## Key Topics ### Project Updates and Workflow Improvements

Project Updates and Workflow Improvements

The team discussed updates across multiple workstreams. Matthew presented a new taxonomy browser tool that helps track AI-generated definitions and their modifications. The group addressed challenges with uncontrolled PR contributions, with Bar suggesting implementing restrictions similar to other open-source projects. Govindaraj reported issues with assigning tasks in the project management system that need resolution. Fernando provided updates on the best practices guide, noting it's in review phase with many comments that need addressing before merging. The team also discussed scheduling challenges for APAC team participation, with Junjie mentioning Alex's efforts to set up alternative meeting times that better accommodate different time zones.

PR Process and Collaboration Updates

The team discussed PR review processes and agreed to implement a comment deadline policy, with Fernando planning to address Bar's comments and create issues for more complex items. Govindaraj reported challenges with cross-working group collaboration, particularly noting limited representation from accuracy, workflow, governance, and Agentic Commerce working groups, and requested better delegation to address overlaps with projects like Agent Router becoming a Linux Foundation project. The meeting also covered updates on AI safety initiatives from major tech companies, European standardization efforts through ETSI, and upcoming events including a Dallas chapter AIF community event on September 26th where Govindaraj sought presentation materials about AIF projects.

Team Conference Updates and Plans

Hsiao announced plans to attend a conference in Amsterdam later this week and invited the team to connect through the Discord channel for a potential lunch gathering on Thursday. Junjie mentioned he would be attending the San Jose conference next month. Govindaraj indicated he would send a Discord message to gather more information about ARD (Agent Resource Discovery) and clarified that it is not currently an AIF project.

## Next Steps -

Fernando: Address the remaining comments (especially from Bar) on the Agentic AI Best Practices PR, then merge the PR and move to the Threat Modeling document.
Fernando: Try to contact Alon (co-lead of Threat Modeling WG) via email to get his involvement.
Matthew: Share the Eclipse Enclave sandbox for AI agents with the group and, if there is interest, arrange for the maintainers to present.
Junjie: Follow up with Christina/Dana regarding GitHub project board permissions, especially the ability to assign issues to specific people.
Junjie: Bring up the need for delegates from other working groups (accuracy/reliability, workflow, governance, Agentic Commerce) in the chairs meeting to improve cross-WG collaboration.
Fernando: For the Best Practices PR, convert unresolved comments into GitHub issues and reference them, so further work can be tracked via issues.
Matthew: Enable or confirm the use of code suggestions in GitHub comments to reduce burden on PR authors.
Fernando: Add a comment deadline/agreement to the CONTRIBUTING.md file regarding PR review timelines.
Govindaraj: Send a Discord message to clarify the status and alignment of Agent Resource Discovery (ARD) within AIF.
Govindaraj: Reach out to attendees/offline for the AIF Dallas chapter event on September 26th, and collaborate on presentation content about AIF projects and open standards.

5e:T1462,## Overview The meeting focused on a presentation by Steve Kaliski and Brendan from Stripe about the Machine Payments Protocol (MPP), an internet-native, open protocol for agentic headless payments designed to support AI agents interacting with payment systems. They explained how MPP differs from X402 by being rail-agnostic, supporting multiple payment methods including cards, stablecoins, and traditional payment processors, and operating through HTTP requests rather than browser-based forms. The presenters demonstrated how MPP works through examples of payment negotiation and discussed the protocol's governance model, which allows individual payment methods to be added independently while maintaining a core specification managed through the IETF. They also addressed questions about how MPP handles authorization and mandate information, explaining that this lives in separate layers rather than being defined in the core protocol. The discussion concluded with insights about future developments, including plans to focus on composing with identity and risk tools, and increasing deployment to learn more about real-world usage patterns. ## Key Topics ### Agentic Commerce Taxonomy Initiative Ilya welcomed new participants to the Agentic Commerce meeting and introduced the taxonomy work being conducted across multiple working groups. Leonardo explained that the Taxonomy and Landscape working stream aims to define key terms that are ambiguous or have different interpretations in the context of Agentic Commerce, starting with a list of 10 initial items focused on commerce protocols. Leonardo requested attendees to review the terms on Discord and provide feedback on whether they are fair starting points or if anything should be added or removed. ### Machine Payments Protocol Overview Brendan and Steve presented an overview of the Machine Payments Protocol (MPP), an open protocol for Agentic headless payments designed to address the evolving needs of AI agents in the digital economy. They explained that as AI agents become more capable and serve as economic actors, traditional payment interfaces may become inefficient and unreliable, leading to the development of MPP. The protocol aims to streamline payments by leveraging internet-native solutions, focusing on scalability and reducing friction compared to existing methods. ### Machine Payment Protocol Overview Brendan explained MPP (Machine Payment Protocol), a headless HTTP-based payment protocol designed to support 150+ payment methods like Stripe, specifically optimized for AI agents. He demonstrated how MPP can handle payments within a single request-response flow without requiring existing accounts, and emphasized its rail-agnostic design that works across different payment methods including cards, Klarna, and stablecoins. Brendan noted that MPP is used in production by various AI native services and has seen adoption in unexpected use cases like online butcher shops and wine sales, while also clarifying that it's best suited for high-frequency data access rather than physical commerce transactions. ### Payment Method Negotiation Examples Ilya requested an example of how to handle payment method negotiation when receiving a 402 response, particularly regarding how agents can discover available payment options like Klarna. Brendan and Steve Kaliski demonstrated an MPP (Multi-Method Payment) request and response example, explaining that MPP runs over the authentication scheme similar to HTTP 401, allowing agents to request payment authentication with various methods supported by the server. The discussion highlighted the evolving landscape of payment protocols, noting that while WebMCP is currently popular due to browser-based agent activity, there are ongoing considerations around PCI compliance and security when implementing payment methods. ### MPP Payment Protocol Design Discussion Brendan and Steve presented MPP (Payment Method Protocol) and discussed its design decisions compared to other payment standards like 402. They explained that MPP was intentionally designed to be crypto-agnostic, flexible, and compliant with web standards, allowing individual payment methods to plug in and extend the protocol independently. The discussion covered how new payment methods can be added to MPP through either dynamic implementation or custom SDK packages, with governance maintained through a bottoms-up approach where anyone can contribute but core changes require high standards. The presenters indicated that the core protocol is nearly complete, with current focus on integrating with identity and risk tools, and expanding merchant adoption rather than adding new payment flows. ## Next Steps - Leonardo: Share the links to the first batch of taxonomy terms on the chat and Discord for the group to review. - All attendees: Review the list of taxonomy terms shared by Leonardo and provide feedback on missing or unnecessary terms. - Brendan: Share the governance page and documentation for adding new payment methods to MPP in the chat. - Aziz: Consider creating and publishing a new payment method SDK for MPP independently, optionally submitting a PR to the MPP GitHub for discoverability.5f:T20a4,## Overview The meeting focused on updates and coordination for the Observability and Tracing working group within the Agentic AI Foundation. Joe and Matthew discussed the taxonomy workstream, emphasizing the need to submit PRs for terms by the end of September and the use of a new tool to track progress. Pavan and Matt shared updates from the chairs meeting, including the creation of a shared repo for tooling and the proposal to add a monthly meeting time for APAC members, which was approved. The group also discussed the Agent Behavior Trace Model deliverable and the need for a white paper on Agentic Observability best practices. Sal raised questions about the appropriate place for reference architecture contributions, and Manu suggested targeting an IETF submission. The conversation also touched on potential collaboration with projects like ARD and the importance of addressing identity and observability challenges. ## Key Topics ### Taxonomy Workstream Progress Update The team discussed updates from the taxonomy workstream, focusing on submitting PRs for terms marked as "keep" in a shared spreadsheet. They agreed to target submitting PRs by the end of September for inclusion in their monthly report to the TC. Matthew presented a Discord tool he created to help track which terms have been submitted, and Joe provided feedback on improving the tool's functionality, including adding metadata fields and identifying PR submitters. The group confirmed GitHub as the central source of truth for tracking their work, and discussed potentially contributing Matthew's tool to the taxonomy repo. ### Taxonomy Guidelines and Updates The team discussed guidelines for submitting taxonomy terms, with Joe confirming that while one PR per term is preferred for better feedback, there's flexibility to group simple terms together. Matthew reported syntax errors in the taxonomy JavaScript on GitHub that need fixing before merging. Pavan shared updates from a chairs meeting, including the creation of a Git repository for automating monthly reports and discussions about adjusting meeting times to better accommodate team members in APJC/Australia. ### APAC Region Synchronization Meeting The team agreed to implement a monthly 30-minute synchronization meeting for the APAC/APJ regions, with timing to be determined through Discord coordination. Matt confirmed that while the current tooling is specific to taxonomy, the group welcome proposals for more general tooling that could benefit the AAI group as a whole. The team also discussed ongoing work with LF to potentially introduce an MCP interface for accessing AI recordings and transcripts from the LF platform, though no timeline was provided for this feature. ### Repository Organization and Community Updates The team discussed repository organization and community structure, with Matt explaining that working group repositories can temporarily be used while a clearer hierarchy is established. Sal provided feedback about Discord usage, preferring GitHub for professional communication due to existing platform fatigue. The group reviewed updates from focus groups, including taxonomy work experiencing increased participation and agent state/context/reasoning progress led by Govindaraj, with collaboration ongoing between multiple team members on AIC and tracing enhancements. ### Trace Behavior Model Progress Update Matt announced that initial tasks for the trace behavior model have been created in Github with a goal to complete an initial reference implementation by end of September. Pavan suggested including external contributors in future discussions rather than having them comment directly on issues. The team discussed an opportunity to submit a proposal to the ARD project community to address operational gaps in resource discovery and federation across different services, potentially using W3C Trace Context and OpenTelemetry specifications to create a unified trace across registries and agents. ### AIIF ARD Project Integration Discussion Govindaraj discussed bringing the Agent Resource Discovery (ARD) project under the AIIF umbrella, noting its alignment with existing AIIF projects and potential benefits for Agent Router and Agent Gateway. Pavan proposed creating a vendor-neutral white paper on agentic observability best practices, covering topics like trace verifiability and attestation, while emphasizing the use of existing standards. Govindaraj mentioned progress on IETF drafts and the development of extension profiles for different needs, suggesting these could inform the best practices document by focusing on currently available solutions. ### Vendor-Neutral Observability Best Practices The group discussed developing vendor-neutral best practices for observability, with Govindaraj suggesting collaboration with companies like Splunk, Datadog, AWS, and Google. Manu shared insights from AWS's customer council, noting that while these companies are working on observability at different levels, they are still developing implementation approaches rather than documentation. Manu proposed creating a design aesthetic paper that could be submitted to IETF as an RFC, with a deadline around October 31st or November 3rd, suggesting this could be a significant contribution given the current lack of standardization in the field. ### Agent Framework Security Best Practices The team discussed the need to incorporate security and privacy best practices into agent frameworks, with Manu offering to contribute to a draft white paper for potential submission to an IETF conference. Manu announced that their organization is now an associate member of LF and offered to share IETF presentation codes with interested team members. The discussion also covered the importance of observability across hyperscalers like GCP, AWS, and OpenStack, with Manu identifying four key areas for taxonomy development: security and compliance, cost management, network architecture, and agent design. Sal emphasized the need for stateful and idempotent guarantees in the framework to ensure cross-platform interoperability. ### Observability and Identity Collaboration Sal requested feedback on a PR by the end of the week to help address gaps before discussing it with maintainers, planning to return with integrations in two weeks. The group discussed ongoing work in observability and identity, with Manu expressing interest in developing best practices and standards in these areas. Govindaraj shared insights about AOT (Agent-Oriented Trust) and its potential to address identity and authorization challenges, particularly in cross-domain scenarios. The conversation ended with discussions about potential collaboration opportunities, including a planned Friday meeting between Manu and Govindaraj to discuss agent context and reasoning, and an introduction to Subramania's open source project Observable.io focused on agent resource discovery. ## Next Steps - Matthew: Fix the Discord-based taxonomy tool to address the error reported by Joe and add functionality to scan PRs and show the PR owner. - Matthew: Consider contributing the taxonomy tool to the taxonomy repo after addressing current issues. - Joe: Use the taxonomy tool to update the spreadsheet with the latest PR status and metadata. - Matthew: Add a feature to the taxonomy tool to show who submitted each PR. - All Working Group Members: Volunteer to submit PRs for the taxonomy terms marked as "keep" in the spreadsheet, aiming to have PRs submitted by the end of September. - Matthew and Pavan: Coordinate via Discord to determine a suitable time for a monthly 30-minute meeting to include APAC/APJ members, then schedule the meeting. - Matt: Follow up with LF folks regarding the MCP interface for AI recordings/transcripts. - Sal: Share the link to the reference architecture PR in the chat and request feedback asynchronously. - Pavan: Forward the invite to the Protocols and Primitives focus group meeting to Sal. - Matthew: Share the link to Sal's PR on Discord. - Manu: Share a free pass code for the upcoming IETF conference with interested participants. - Manu: Send a meeting invite to Govindaraj for a discussion on Friday at noon Central time regarding agent context and reasoning. - Manu: Share a template for the Observability Best Practices white paper with the group.60:T155f,

Explore Working Groups Events

Subscribe to the AAIF Briefing

Explore Working Groups Events

Subscribe to the AAIF Briefing