The challenge

On 2 October 2026, Anthropic announced a $100 million investment in the Claude Frontier Academy, with the goal of training 10,000 Frontier Deployed Engineers by the end of 2027.

Anthropic describes the central enterprise AI problem not simply as access to better models, but as a shortage of people capable of taking those models from an idea to a system running in production.

That is an interesting proposition.

It also creates an interesting experiment.

How much of that capability can an experienced domain engineer build independently?

This project is my attempt to find out.

The experiment

Over twelve weeks I will combine structured learning with the design and construction of a real AI system in a domain I already know well: industrial and operational technology cybersecurity.

This is not an attempt to reproduce Anthropic's Frontier Academy, nor does completion confer Anthropic's Frontier Deployed Engineer credential.

Instead, Anthropic's description of the FDE provides the benchmark.

The objective is to move beyond simply using an LLM and develop the engineering capability required to turn one into a useful, controlled and measurable enterprise system.

The proving ground

The project will centre on an AI-assisted OT cybersecurity assurance system.

Rather than allowing an AI system to make security decisions autonomously, the initial objective is to investigate how an agentic system can assemble and correlate assurance evidence, identify missing or contradictory information, reason against defined requirements and present its conclusions to a human reviewer with clear provenance.

That creates a useful test of several problems that become unavoidable when AI leaves the laboratory and enters an enterprise environment:

• controlled access to data and tools
• evidence provenance
• retrieval and context management
• uncertainty and hallucination
• permissions and least privilege
• human approval boundaries
• evaluation and testing
• observability and auditability
• security of the AI system itself
• measurable improvement over the existing process

Learning by building

The educational programme runs alongside the engineering work rather than preceding it.

The syllabus covers agentic AI, tool use, Model Context Protocol (MCP), retrieval-augmented generation, agent orchestration, evaluation and LLMOps.

Each stage must result in an observable improvement to the system.

The aim is therefore not to accumulate AI certificates.

It is to build something.

The question

At the end of the twelve weeks I should be able to answer a much more useful question than whether I completed a collection of courses:

Can an experienced industrial cybersecurity engineer acquire the practical capabilities required to take frontier AI from an idea to a defensible enterprise system?

The successes, failures, architectural decisions and results will be documented here as the project develops.