Agentic AI
570 documents · 93% importance
The strongest AI historical winner in the 12-month reference view, represented across three source types.
Focus / ARTIFICIAL INTELLIGENCE
From models
to operating systems.
The centre of gravity is moving beyond model access. The durable layer is the system that gives models context, tools, permissions, evaluation and a place inside a real workflow.
ContinueRead the thesisThe thesis
MANFRED / 02
The durable advantage is moving beyond model access: toward systems in which intelligence can act, remember, be checked and be trusted.
The public signals now sit around the infrastructure of agency: enterprise control planes, open protocols, tool access and the governance required to operate them.
What Manfred is seeing
The pattern is already visible in the evidence field.
Manfred internal metrics · not market-size forecasts
Snapshot dated 27 June 2026. AI domain tag: 7,710 documents and 21 final clusters. Active observation window: 28 April to 27 June 2026. Historical-winner comparison uses a 12-month reference window.
RAG & knowledge retrieval
345 documents · 66% importance
A large cross-source reference theme whose current matches repeatedly converge with orchestration and vector-database patterns.
Vector database / multi-agent orchestration
88.5 emergence · 79% opportunity
The highest visible orchestration cluster in the snapshot, aligned with the historical Agentic AI reference pattern.
LLM agent / coding agent
84.2 emergence · 63% opportunity
A fast-moving execution layer in which agents are applied to bounded, testable work rather than generic conversation.
Interpretation: The strongest common motif is not a single model family. It is the surrounding architecture that turns model output into work: retrieval, context layers, handoffs, tool use, agent harnesses, evaluation and controlled deployment.
What is changing now
The field is acquiring institutions, standards and operating constraints.
Public moves in products, standards and policy give a current external context to the Manfred signal field. They are selected because they relate directly to the clusters above.
Public events are used here as tests of the Manfred reading: they do not prove the thesis, but show whether the field is beginning to organize itself around the same technical and institutional constraints.
Company signal
The agent control plane becomes a product.
OpenAI’s Frontier is presented as a platform to build, deploy and manage agents with shared context, feedback, permissions and boundaries. The product shape itself validates that the bottleneck has moved beyond the model call.
Open standardInteroperability crosses the production threshold.
The Linux Foundation reports A2A support from more than 150 organizations, with integrations across Google, Microsoft and AWS. Agent-to-agent coordination is starting to look like infrastructure rather than a vendor feature.
Protocol watchTool access is becoming common infrastructure.
Anthropic donated the Model Context Protocol to the Linux Foundation’s Agentic AI Foundation. The signal is not that one protocol has won; it is that data and tool connection is being treated as a shared layer.
Regulation watchModel governance reaches enforcement.
The Commission states that GPAI provider obligations have applied since August 2025 and that its enforcement powers begin on 2 August 2026. Procurement now has to account for documentation, risk and governance, not capability alone.
Company announcements are treated as market signals, not independent performance validation. The observable pattern matters: model providers, cloud platforms and standards bodies are all investing in the layer around models.
What the evidence may suggest
The question is no longer only what a model knows. It is what the system lets it do.
Context becomes infrastructure
Retrieval, memory, identity and permissions become durable organizational layers. Models may be substituted more readily than the contextual system around them.
Evaluation moves into the product
Once systems can take actions, evaluation cannot remain a benchmark exercise. It becomes continuous verification of tools, workflows, failures, handoffs and recovery cost.
Agency creates a new control surface
The critical governance problem is increasingly what an agent may access, change, trigger or escalate inside a real operating environment.
What could challenge this view
A thesis becomes useful when it can be tested.
This view weakens if model capability continues to dominate outcomes while open standards, tool reliability and organisational integration fail to mature. It also weakens if most agent systems remain narrow productivity features rather than reaching repeatable, accountable production workflows.
Evidence register
The source layer stays visible.
The interpretation on this page is Manfred’s. The records below separate the internal evidence pattern from external sources used to test and contextualise it.
Manfred
Build
AI field snapshot
Captured 27 June 20267,710 AI-tagged documents and 21 final clusters. Selected reference themes: Agentic AI, RAG & knowledge retrieval, software engineering agents, evaluation and governance. Internal evidence shows what the system retrieved; it is not a market-size measure.
●OpenAI · Frontier
5 February 2026Company announcement for an enterprise platform focused on shared context, permissions, onboarding, deployment and management of AI agents. Used here as a product-architecture signal.
StandardLinux Foundation · Agent2Agent Protocol
9 April 2026The Foundation reports cross-platform production use and support from more than 150 organizations. Used as evidence of growing interoperability infrastructure, not a claim that interoperability is solved.
ProtocolAnthropic · Model Context Protocol donation
9 December 2025Anthropic announced the donation of MCP to the Linux Foundation’s Agentic AI Foundation. Used as a standardisation signal for agent-to-tool connections.
RegulationEuropean Commission · GPAI provider guidelines
Current guidanceCommission guidance explains GPAI model provider obligations and the 2026 enforcement transition. It does not, by itself, classify every agentic application.