Agentic AI Engineer

What you learn

The agent loop built by hand before any framework touches it
Tools designed as contracts and granted with least privilege
Memory engineered with a conscience about what it keeps
Autonomy that is traced, budgeted, gated, and stoppable

What you will understand

An agent is an actor with credentials whose behaviour is shaped by untrusted input
Limiting what an agent can do beats hoping it behaves
Instructions arriving inside content are data, never commands
The honest question of when a workflow beats an agent entirely

How you will practice

Build a working loop against a raw model API with no framework at all
Smuggle an instruction into a tool result and prove it changes nothing
Exercise a kill switch mid run and account for the state left behind

What you will build

A production agent with a blast radius map and human gates on consequential actions
An evaluation harness that catches a behaviour regression before a user does

The proof you build

Evidence that you can ship autonomy an organisation can hold accountable.

Your work is observed with your consent, scored for independence and assistance, and turned into proof that carries a confidence level. The career path can reach a high-assurance credential, anchored by a scored capstone.

See how proof works

Related capability paths

AI-Assisted DevelopmentPrerequisiteBuild software with generative tools without losing responsibility for what ships.
Agentic Coding WorkflowsPrerequisiteDelegate real engineering work to autonomous agents and stay accountable for what comes back.

What is not live yet

The desktop app, consent-based observation, scoring, and credentials are in development. Nothing here implies they are live yet.

Get early access