The agentic stack crossed a line in Amsterdam last week.
Two days — September 17-18, 2026 — at RAI Amsterdam and the conversation shifted decisively past demos and into production. The question on every engineer’s lips was no longer “can a model do X?” but “how do we run agents that are reliable, observable, and secure when they hit real infrastructure?”
What the data says
- 81% of teams now have agents in production — not prototypes, not experiments, live systems handling real workloads
- 89% MCP adoption among agent frameworks — the protocol is no longer emerging, it’s default
- 60k+ repositories now ship an
AGENTS.mdfile — the human-readable README-for-agents standard is exploding - goose (Block’s Rust-based agent) is seeing real adoption beyond coding — data analysis, research workflows, automation
The three shifts that defined the floor
1. From single-agent demos to multi-agent systems
Google’s A2A protocol was everywhere — not in keynotes, but in the hands-on demos. Teams are running fleets of specialized agents that discover each other’s capabilities, negotiate interaction modalities, and collaborate without sharing memory or tools. The security model is explicit: agents expose only what they need to, nothing more.
2. MCP is now infrastructure, not experimentation
Clare Liguori (AWS) and David Soria Parra (Anthropic) both emphasized the same trend: MCP has become the integration layer that teams default to. The July 28 specification changes — stateless core, extensions framework, authorization updates — are solving the production pain points: cost attribution, auth boundaries, and server lifecycle management.
The Agent Gateway pattern (securing and routing MCP + LLM traffic) moved from “interesting project” to “table stakes” in six months. Teams are standardizing on centralized gateways for observability, rate limiting, and cost control.
3. AGENTS.md is how you ship agents that respect your codebase
The message was consistent: an agent that doesn’t know your repo’s build commands, test conventions, and code style is a liability. AGENTS.md solves this with a single Markdown file at the repo root — no JSON schema, no config dance. Projects adopting it report 40% faster agent onboarding and fewer broken builds.
Key sessions that mattered
- David Soria Parra (Anthropic): MCP production patterns — stateful vs stateless servers, authorization flows, and the migration path from early MCP adopters
- Clare Liguori (AWS): MCP at scale — how AWS is running thousands of MCP servers in production and what broke getting there
- Angie Jones (Agentic AI Foundation): Agent evaluation frameworks — the shift from LLM benchmarks to agentic integration tests
- Dexter Horthy (HumanLayer): Context engineering — not prompt engineering — as the discipline that determines agent success in production
- Idit Levine (solo.io): Agent Gateway reference architecture — authn/authz, rate limiting, and observability patterns for agentic systems
The real problem: context, not compute
The dominant theme across workshops and hallway conversations: most agent failures in production are context problems, not capability problems. Agents lose track of what they’ve done, call the wrong tools, drift from intent, or silently fail.
Solutions that gained real traction:
- Persistent agent memory across sessions — not just conversation history, but learned preferences and discovered workflows
- Agent Router patterns — routing each request to the optimal model+tool combination based on cost, latency, and capability, not a static config
- Evals as release gates — running agent regression suites before deploy, not after
Where this goes next
The free livestream captured all of it. Recordings are rolling out on the Agentic AI Foundation YouTube channel.
The next AGNTCon is scheduled for early 2027. If the Amsterdam energy is any signal, agentic systems aren’t coming — they’re already here, and the conversation has moved to making them work reliably at scale.
This is a media-partner recap. My preview article with the full agenda, speaker lineup, and track breakdown is here.
