Evidence index

Signals become useful when their sources remain visible.

A growing set of research notes connecting companies, protocols, standards and regulation to the evidence fields Manfred tracks across artificial intelligence, health, energy and robotics.

Artificial intelligence · Protocol watch

Why MCP is becoming the context layer for enterprise agents

The question is no longer whether a model can call a tool. It is whether a company can expose the right data, actions and permissions through a controlled, portable interface.

Artificial intelligence · Operational control

Agent evaluation is moving from benchmark to operating control

Once an agent can act inside a workflow, evaluation stops being a scorecard. It becomes a continuing system for checking permissions, tool use, outcomes and recovery.

Artificial intelligence · Governance watch

Enterprise agents create a new permissions problem

The critical question is not whether an agent is useful. It is what it can access, change, trigger or escalate inside a real organisation.

Health & medtech · Company & model signal

MedGemma shows where the clinical multimodal stack is forming

Models that can work across medical text and images make it easier to prototype a more continuous evidence environment. Clinical validity still has to be earned case by case.

Health & medtech · Regulation watch

Medical AI is becoming a lifecycle evidence problem

The relevant question is not whether a model was impressive at launch. It is whether its safety, performance and intended use can remain visible as data, workflows and populations change.

Health & medtech · Performance watch

Real-world performance is becoming the clinical AI frontier

The next barrier is not only building an AI-enabled device. It is detecting when performance changes across sites, populations, inputs and time.

Energy & industry · Infrastructure watch

Digital twins are moving from grid visualisation to grid operation

A digital twin becomes strategically important when it supports coordinated planning, resilience and real-time operational decisions across a physical system.

Energy & industry · Policy watch

AI is entering the energy system through digital infrastructure

The relevant shift is not AI as a separate vertical. It is AI becoming part of how grids, flexibility, planning and energy data are coordinated.

Energy & industry · System dynamics

Grid flexibility is becoming a control problem

Storage and demand response matter because they change the operating range of the system. Their value depends on coordination, signals and reliable execution.

Robotics & strategic systems · Embodied AI watch

Vision-language-action models are changing the robotics stack

Robots are beginning to receive a higher-level layer for perception, spatial reasoning, planning and adaptation. The remaining difficulty is making that layer dependable in the physical world.

Robotics & strategic systems · Deployment watch

On-device models matter because latency is a safety and capability question

Robotics cannot always depend on a distant cloud service. Local inference changes how quickly a machine can perceive, decide and respond inside a physical environment.

Robotics & strategic systems · Regulation watch

The EU Machinery Regulation puts physical AI on a compliance horizon

Embodied intelligence changes what a machine can do. It also makes the safety case, documentation and conformity pathway more central to product design.

Editorial policy

We distinguish evidence from interpretation.

Each note identifies the external records used for context and states a working view rather than a prediction. Company announcements are treated as market signals, not independent validation. Internal Manfred evidence is described separately from public-source evidence.