Digital Jersey’s AI Project Lab moves from training to prototypes

Digital Jersey’s Let’s Build AI Project Lab is designed to move local organisations beyond executive AI training and into tightly scoped working prototypes. The six-session programme requires business and technical participants to bring a defined use case, available data and employer backing, then work towards a prototype, risk-and-governance canvas and practical deployment plan.

The six-session lab asks Jersey teams to bring a tightly scoped use case, a business-case summary, an understanding of their available data and employer backing. In return, it promises a working prototype, a risk-and-governance canvas, a practical deployment plan and a 90-day commitment to continue the work. Applications closed on 19 August 2026; Digital Jersey says the programme begins on 9 September 2026. (digital.je)

What is Digital Jersey’s Let’s Build AI Project Lab?

Project Lab is the practical follow-on to Digital Jersey’s AI Executive Leadership Programme. The earlier programme is positioned as strategic education; the lab is intended to help organisations build and unblock a defined project. Teams work with two facilitators across six two-hour evening sessions, using tools that may include Claude Code, Cursor, VS Code, Jupyter, GitHub, Hugging Face, PyTorch and standard Python and machine-learning libraries. (digital.je)

That is deliberately a broad tool list, not a prescribed stack. A trust company’s internal document-assistance pilot, a professional-services workflow and a data-science model may need different models, hosting arrangements and coding environments. What matters is the format: architecture and data decisions are made against a real use case, not in a vendor demonstration.

Digital Jersey requires applications in pairs, covering business and technical perspectives. That may sound administrative, but it is one of the programme’s better design choices. AI projects often fail because a technically interesting demo has no accountable process owner, or because a business sponsor underestimates the integration, access-control and support work needed to make it dependable. A paired team doesn’t solve those problems. It does make them harder to ignore.

What will a participating organisation actually produce?

The sequence is straightforward. The first session tests scope and feasibility; the second establishes accounts, APIs, connections and a development environment. By the third, teams are expected to have a thin end-to-end prototype rather than a collection of prompts or slides. Later sessions cover production readiness, governance, iteration, logging, feedback and observability, before a final showcase and handover. (digital.je)

For a well-chosen use case, that should answer some valuable questions quickly:

• Can the organisation access the data it needs lawfully and reliably?
• Does the proposed workflow produce outputs accurate enough to assist a human decision-maker?
• Where are the failure modes, and can staff spot and correct them?
• What supplier, hosting, security and support arrangements would be needed beyond the prototype?
• Is there enough operational value to justify a proper production project?

Those are better questions than “which chatbot should we buy?”. They separate an attractive interface from a usable operating model.

The limitation matters too. Twelve hours of facilitated contact, plus the expected work between sessions, can validate a narrow workflow. It isn’t enough to deliver a resilient, integrated production service for a complex regulated process. Organisations should treat the final prototype as evidence for a deployment decision, not as a way around engineering, procurement, security testing or change management.

How does the lab deal with governance, data and human oversight?

Governance isn’t left to a closing disclaimer. Digital Jersey’s fourth session explicitly covers data policies, human oversight, access controls, security and relevant Jersey Financial Services Commission and Jersey Competition Regulatory Authority considerations. The stated output is a risk-and-governance canvas alongside a practical deployment plan. (digital.je)

That is particularly relevant in Jersey, where much of the Island’s economy handles confidential client material and regulated activity. The JFSC’s AI guidance does not create a new standalone rulebook, but it is clear that firms remain accountable for outcomes. It calls for a proportionate, risk-based approach, with stronger controls where AI affects regulated activity, customer outcomes, confidential data or regulatory submissions. Higher-impact examples include onboarding, lending decisions and investment advice. (jerseyfsc.org)

The practical distinction is clear. An internal drafting assistant using only approved, non-sensitive source material may justify lightweight controls. A system that recommends risk ratings, prepares regulatory material or materially influences customer treatment does not. That second category needs named accountability, testing, documented controls and a real route for human intervention.

Data protection belongs in the build, not after the demo. The Jersey Office of the Information Commissioner says personal data remains personal data when fed into an AI system, and that AI-generated outputs about identifiable people can themselves be personal data. It advises organisations to consider data protection by design, supplier terms, data location, retention, security and whether automated decisions are subject to additional protections. A human merely rubber-stamping an AI decision is not meaningful human involvement. (jerseyoic.org)

So Project Lab’s requirement for teams to understand data availability and access constraints before they begin is more than sensible housekeeping. It guards against a familiar failure mode: an impressive prototype built with material that cannot legally, contractually or safely be used once the system is live.

Why is this a more mature stage of Jersey AI adoption?

Training still matters. Senior leaders who don’t understand model limitations, cost patterns, data exposure or automation risk make poor decisions. But education has a predictable ceiling: people return to work with enthusiasm, a few possible use cases and no route through the next set of organisational blockers.

Project Lab is built around those blockers: scoped work, accountable owners, technical execution, governance and a near-term adoption plan. Digital Jersey says the programme responds to a gap seen after its first Let’s Build AI season and similar programmes, where promising ideas did not become working solutions. (digital.je)

It also fits the Government of Jersey’s wider Digital Economy Framework. The framework assigns Digital Jersey targets for AI-provider relationships, an AI Hub, co-ordination through the AI Council and two to three AI pilots focused on professional and financial services by the end of 2026. It frames safe AI adoption, data use and proportionate regulation as part of Jersey’s economic-development agenda. (gov.je)

There is still a gap between policy ambition and repeatable delivery capability. A small cohort programme won’t close it on its own. It can, however, produce something more useful than a broad awareness campaign: local evidence of which problems are worth automating, which need better data first and which should be stopped before more money is spent.

That is a better measure of progress for Jersey’s digital economy than counting AI licences or well-attended seminars.

Which Jersey organisations are most likely to benefit?

The strongest candidates are not necessarily the largest firms or those with the most elaborate AI strategy. They have a narrow, repeatable problem; a process owner able to make decisions; a technically capable participant; and enough authority to access the necessary systems and data.

Likely examples include document-heavy professional services, internal knowledge workflows, controlled customer-service triage, operational reporting, compliance-support tasks and specialist research processes. In financial services, the first sensible projects will often assist employees rather than make autonomous customer decisions. That gives teams room to test usefulness and error patterns while preserving human responsibility.

Firms should hesitate if the proposed “use case” is really an aspiration such as “use AI across the business”. They should also pause if nobody can explain the source data, where it will be processed, who will review outputs or what happens when the tool is wrong. A lab cannot compensate for absent ownership.

What should teams do if they are not in this cohort?

Organisations outside this cohort can still borrow the discipline. Pick one process where delay, rework or search effort is measurable. Define what a good answer looks like. Identify the data source and its permissions. Decide who remains accountable for the result. Then build the smallest test that can disprove the idea as well as support it.

Make the production decision explicit. There are three valid outcomes: proceed to a controlled deployment, improve the data or process and test again, or stop. The last is not failure; it is why prototypes exist.

Digital Jersey’s Project Lab should not be judged by how polished its showcase demos look in October, but by what happens afterwards. If participants leave with a defensible decision, clear ownership and a 90-day path for the work, it will have done more for Jersey AI adoption than another round of generic AI literacy.