iiinova — where your digital self keeps evolving

Nearly 100 AAIF chapters, with 40 events across 15 countries in Sept! A warm welcome to the chapters that made their debut this month in Colombo, Luxembourg, Singapore, Shenzhen, Bengaluru and Dallas. Thank you to the local organizers bringing these communities together. bit.ly/3TXO1YL

000

After AGNTCon + MCPCon EU, Obot (member company) shares three themes: AI control planes are taking shape, governance is becoming part of the development workflow, and the old API-and-app model is being stretched by agents. Read on: bit.ly/3TzrNMD AGNTCon + MCPCon NA: bit.ly/4vkntyD

WebMCP, MCP Apps, and A2UI are building the second layer of the web — made for agents, with humans still in the loop. Alex Salkever breaks it down: bit.ly/4zqAVTd Want to go deeper? Catch the conversation at AGNTCon + MCPCon North America, Oct. 22–23 in San Jose: bit.ly/4lT6ZJM

120

An OCI image records what an app contains and how it launches. It has no field for what the app may access once it runs. Docker calls this the agentic container gap. V3 of the Docker Sandbox Kit Spec, Apache 2.0, moves agent permissions inside the OCI image: bit.ly/4xVMmkv

OpenSSF @openssf.org · 29/09/2026

Reposted by Agentic AI Foundation

Angie Jones, VP at the Agentic AI Foundation, joins What's in the SOSS talking about what it actually takes to run AI agents in production. 🎧 openssf.org/podcast/2026...

112

Angie Jones, VP of the Agentic AI Foundation, and Jono Bacon are joining AGNTCon + MCPCon North America for two days of agents, MCP, open source, interoperability, and everything moving this ecosystem forward. October 22–23. Come build what’s next. bit.ly/4vkntyD

Every high-value agentic use case in banking, insurance, and payments runs on data that lives on a mainframe. Invert the pattern, move the reasoning not the records, and make the harness the audit boundary. bit.ly/4iOTefH

000

Agentic AI Foundation @aaif.io · 28/09/2026

Beyond Coding sat down with Dario Hamidi from Amp at AGNTCon + MCPCon Amsterdam to talk about what agentic engineering actually looks like in practice. bit.ly/4zneXjZ Come catch the next one at AGNTCon + MCPCon North America. Register: bit.ly/4vkntyD Code BEYONDCODING for 15% off.

000

Hear the whole thing: go.mlops.community/y07qhw Spotify: open.spotify.com/episode/0ExN...

go.mlops.community

AWS Has 16,000 APIs. Can MCP Handle It? - Video | Agentic AI Foundation

000

Agentic AI Foundation @aaif.io · 28/09/2026

AWS has ~16,000 APIs, so its MCP server runs in code mode. James Ward (AWS) on why agent-written Python is hard to review, and the restricted, schema-validated DSL he's testing instead.

110

Agentic AI Foundation @aaif.io · 28/09/2026

AAIF Ambassador Dakota Fabro built a possibility matrix to change the way humans make judgment calls with agents: instead of asking you to generate every possible answer, the agent lays out the options. Read on: bit.ly/4d3j4ZK

bit.ly

dakotaOS — Dakota Fabro

Android Engineer · Organizational Intelligence Engineer · Research Scientist. Building products that make humans more capable — and measurement instruments that prove it.

000

Agentic AI Foundation @aaif.io · 25/09/2026

From Tokyo to Amsterdam—and next, San Jose. AAIF Ambassador Lin Sun shares what she learned across two weeks of AGNTCon + MCPCon. Read her take: lnkd.in/e6TAG5RZ Then keep the conversation going at AGNTCon + MCPCon NA, October 22–23 in San Jose: lnkd.in/gSkG3yzZ

020

Agentic AI Foundation @aaif.io · 25/09/2026

MCP's spec says protect requestState integrity. It doesn't say where keys live, how replicas share them, or what enforces single-use. A follow-up on the three operational gaps the SDK leaves open, and how to close them with OpenBao. bit.ly/3VcS57V

130

Agentic AI Foundation @aaif.io · 25/09/2026

From Tokyo to Amsterdam—and next, San Jose. AAIF Ambassador Lin Sun shares what she learned across two weeks of AGNTCon + MCPCon. Read her take: bit.ly/4z0nc5f Then keep the conversation going at AGNTCon + MCPCon NA where Lin will be joining us again: bit.ly/4vkntyD

000

An enterprise runs the agent on its own infrastructure. The model provider needs its weights usable only by an approved serving application. A look at Weight Custody Manifest, an open pre-1.0 spec that keeps release conditions beside the model: bit.ly/3VdsXxK

020

How MoonPay brought MCP sprawl under control. 30M+ customers, 900+ partners, and employees connecting AI assistants directly to internal systems through MCP. With Speakeasy, they now have 200+ MCP servers under governance and 60K+ agent sessions covered. bit.ly/4hJXNqo

Why Skills And MCP Are Apples And Oranges? - Video | Agentic AI Foundation

Catch the episode: go.mlops.community/9l7v0d Listen here: open.spotify.com/episode/7Arl...

go.mlops.community

000

Agentic AI Foundation @aaif.io · 23/09/2026

A server that lists its skills and serves them as resources, so tools ship with instructions for using them. Ola Hungerford of Nordstrom on Skills over MCP, dependency gaps, and the context bloat that returns at hundreds of skills.

210

Paste a URL, and your agent picks up new capabilities. That "paste and go" experience is the core design intent of MCP: clients and servers that have never met can still connect. Enterprise closed-world setups get their own extensions. Read the full post: bit.ly/4yXf5X6

Agentic AI Foundation @aaif.io · 23/09/2026

Missed AGNTCon + MCPCon EU live? We’ve got you. Two days of agentic AI, MCP, open source, and what builders are actually working on right now — featuring voices from Anthropic, OpenAI, AWS, Google DeepMind, Hugging Face, and more. - Day 1: bit.ly/4yfpWeY - Day 2: bit.ly/4iDyNlL

000

Agentic AI Foundation @aaif.io · 23/09/2026

Angie Jones has officially entered her protocol girly era. 💅🤖 Her take on WebMCP: what if tool calling lived directly in the web page itself? Hear more from Angie and other builders pushing Agentic AI forward at AGNTCon + MCPCon NA: bit.ly/4vkntyD 📍 San Jose 📅 Oct. 22–23

000

Agentic AI Foundation @aaif.io · 22/09/2026

x402 v2 and MCP 2026-07-28 both moved routing and payment data into HTTP headers, so gateways can meter paid tool calls without parsing JSON-RPC. The harder part is deciding whose budget is being spent, and where that decision lives. New technical writeup: bit.ly/4y5U5gD

000

Agentic AI Foundation @aaif.io · 22/09/2026

An 8B model can beat a 70B model at tool routing, without fine-tuning. The difference? Don’t make the smaller model search a 443-tool catalog. Retrieve the relevant tools first, then bind a shortlist of about 10. Blog by AAIF Ambassador Vincent Caldeira: bit.ly/4hEK18s

bit.ly

Improve SLM tool calling without post-training

An 8B model with a short tool list can beat a 70B model on the full catalog. Keep the cheaper model; filter what it sees — no fine-tune required.

Agentic AI Foundation @aaif.io · 22/09/2026

AAIF is proud to join the Gates Foundation and 60+ organizations around a shared five-year goal: help an estimated 3.4 billion people whose languages are underrepresented in today’s AI models use AI tools in their own language and voice. - gates.ly/4xExA1u - gates.ly/4y4ZGnp

Agentic AI Foundation @aaif.io · 21/09/2026

Japan understood the assignment. 🔥 North America, your turn. AGNTCon + MCPCon NA is hitting San Jose Oct. 22–23 for two days of agents, MCP, open protocols, production stories, and plenty of “wait, you built WHAT?” moments. Register now: bit.ly/4vkntyD

110

Recap from the AAIF meetup in Shanghai on Sept 5, ahead of AGNTCon + MCPCon China. Talks on cross-domain agent collaboration, safety checks for mobile GUI agents, RocketMQ lessons for agent messaging, and models + harnesses + tools developing together. bit.ly/3Tg6QWS

Walking Tokyo Talking Agent Protocols - Video | Agentic AI Foundation

Full conversation here: go.mlops.community/sy8ks7 Spotify: open.spotify.com/episode/5nZJ...

go.mlops.community

WebMCP moves tool calling into the web page, scoped to the page you are on and running in your logged-in session, instead of an agent screenshotting and searching the accessibility tree. Angie Jones on why that matters and what still blocks adoption.

100

Agentic AI Foundation @aaif.io · 17/09/2026

Session 25 of Coding Agents Lunch & Learn explores verification and recovery with Tanmay Sah and Dolly Sah, covering The Verifier Tax and EvoUndo, and what it takes to build coding agents that can check their work and recover from failures. Today from 9–10 AM PDT: bit.ly/4r31eMc

000

AAIF Community Munich brings builders together to unpack what’s working, what’s breaking, and what still hasn’t been figured out. September 22 in Munich, with talks on AI Landing Zones and harness architecture in AWS. Register: luma.com/fkwvz7p3

000

Agentic AI Foundation @aaif.io · 16/09/2026

Think you know who's taking the keynote stage? 👀 Co-founder of the AI Disclosures Project. Helped launch "open source" as a movement at a pivotal 1998 summit. 🔎 The mystery speaker is Tim O'Reilly. Catch him at AGNTCon + MCPCon North America, 10/22–23. Register now: bit.ly/4xTLw8i

000

New on the AAIF blog: moving from a supervised shell to a purpose-built MCP server for agent tool access. Worked example is gpu-mcp-server on a Tesla T4 in Colab. Four named GPU tools, invalid input rejected, unregistered shell_exec refused, live CUDA workload validated. bit.ly/3UPqPMA

151

Context Inspector: Measuring Context Quality Before the LLM Call

Member company American Express explores a practical question for RAG systems: what if context quality could be evaluated before the LLM call? Context Inspector measures coverage, freshness, source starvation, contradiction density, and more. Read on: bit.ly/4yrvS4h

bit.ly

A practical framework for identifying gaps, conflicts, and quality risks in retrieved context before they reach the model.

000

Live from AGNTCon + MCPCon Europe. September 17–18, the Agentic AI Foundation is bringing keynotes, technical talks, speaker interviews, Expo Hall coverage, demos, podcasts, and behind-the-scenes conversations from RAI Amsterdam to YouTube. Register free to watch live: luma.com/aaif-afd7

011

The Gates Foundation is joining AAIF as our first philanthropic member, with Matt Vasey taking a board seat. Collaboration kicks off at MCP Dev Summit Nairobi, where the Gates Foundation is a Diamond Sponsor and funding developers from across Africa to attend and contribute. bit.ly/4h6UgRN

000

Solo.io Open-Sourced Desktop Agent Controls. Solo.io’s Apache-2.0 Agentdesktop inventories agents, servers, and skills, applies policy, and connects devices through existing identity infrastructure. Telemetry is opt-in: agentdesktop.dev/docs

000

"As a company we are spending $2 per million token. What does it mean? It doesn't mean anything." Kuntal Patel and Abhinav Lad of Palo Alto Networks on why per-token cost is a weak KPI, and what they measure once agents start looping on their own.

110

On Sept 16, join the AAIF for a virtual event exploring voice agents, real-time conversational AI and what comes next for voice-powered AI. Featuring Luke Miller and Nicola Croon from SLNG on the launch of UNMUTE. ⏰ 8:30–10:00 AM PDT 💻 Virtual Register: bit.ly/3UrMudB

000

Agentic AI Foundation @aaif.io · 11/09/2026

Introducing the MCPA, the first official certification for the Model Context Protocol, from the Agentic AI Foundation. 120-min proctored exam covering fundamentals, architecture, execution, security, and ecosystem. Aligned to spec 2026-07-28. Enroll now: bit.ly/4rgbi4E

000

Member company Block built an AI agent that fixes bugs. Moneybot reproduces, diagnoses, fixes, and verifies issues before opening patch and regression-test PRs. More than 40 fixes have been verified, with humans approving the merges: bit.ly/3UXsbon

000

Agentic AI Foundation @aaif.io · 11/09/2026

As skills move from individual workflows to team-wide and company-wide systems, evaluations become critical. How do teams know which version works best? Continue the conversation with keynote speaker Dexter Horthy at AGNTCon + MCPCon EU in Amsterdam, September 17–18: bit.ly/4xvWpNq

020

Agentic AI Foundation @aaif.io · 10/09/2026

Dex Horthy has fresh evidence that unattended coding agents still turn healthy codebases into radioactive spaghetti. HumanLayer's own software factory hit 30k+ lines before they archived it. There Is No Software Factory Without Better Verifiers at AGNTCon + MCPCon NA, 10/22. bit.ly/4h7gtiG