What an agentic AI intern actually builds in twelve weeks
From a client problem to a production agent, with the client's compliance officer satisfied.
"Internship" usually means fetching coffee and formatting slides. Ours means shipping. Over twelve weeks an intern on our Agentic AI Internship Programme takes a real client problem and turns it into a production agent that a business licenses and pays for.
Week one is not a model
It starts with the problem, not the technology. The intern sits with a client, finds the process that is actually costing them time, and scopes something that can be delivered. Most of the skill is here: learning to tell a nice-to-have from the thing that is genuinely eating the week.
The build
Then the engineering — prompt and tool design, retrieval, evaluation, guardrails, and the monitoring that lets someone trust an agent in production. Alongside a senior engineer, not in a sandbox. The agent runs against real data under real governance.
The test that matters
The programme is not finished when the demo works. It is finished when the client's compliance officer is satisfied — when the audit trail, the data handling and the documentation stand up to the person whose job is to say no.
Seventy-five percent of the licence revenue from a completed agent goes to the intern who built it. Over the past year, 75% of candidates were placed with a business inside the twelve weeks. That is the point: a first real job in tech, earned on real work.