An agent that never stops looking — and never acts for them.

A 17-year-old applying to university or apprenticeships in the UK is asked to do something genuinely difficult: hold a strategy in their head while moving between fifteen websites that don't talk to each other. UCAS, university course pages, GOV.UK's apprenticeship search, employer career sites, college advice, half-remembered Open Day dates, a referee who needs an email, a deadline that may have moved.

The task isn't really searching. The task is not missing things, the apprenticeship that posted on a Tuesday morning while you were in a Maths lesson, the Open Day that would have changed your mind, the deadline missed by three days because the form needed a reference you hadn't asked for yet.

Search engines hand you ten blue links. They don't notice when something new appears. They don't know your predicted grades, or that you're cost-sensitive, or that last week you said you were done with pure maths. They don't remember.

You begin with a conversation. Not a form, a conversation. You tell Pathfinder what you're studying, where you are, what you're predicted, what you're drawn to, what you're worried about. You don't have to know what you want. Most 17-year-olds don't, and Pathfinder is built around that fact rather than against it.

From that conversation it builds a picture of you, predicted grades, subject strengths, location preferences, financial constraints, the softer texture of what you're hoping for. That picture becomes the lens through which it sees every opportunity it finds.

Then it goes to work. It indexes UCAS-style courses and apprenticeship vacancies, ranks them against you, and re-ranks them as you change. It notices the deadline you haven't acted on, the Open Day you'd want to attend, the apprenticeship that opened at 9am, the degree apprenticeship that quietly solves a worry you mentioned three weeks ago. When it finds something, it brings it to you with reasons, why it fits, where it's risky, what it would take, what to do next.

Most search tools give you results. Pathfinder gives you a strategy. Every shortlist is shaped around how applications actually work: one aspirational first choice, three solid backups that protect you if grades or interviews don't go to plan, and one wildcard, an opportunity adjacent to what you've been looking at, surfaced because it thinks you might not have known it existed.

The wildcard is often the most valuable suggestion. The degree apprenticeship for the cost-sensitive student. The foundation year for the late bloomer. The Manchester course for someone who said London but would actually love Manchester if they let themselves consider it.

When you save an opportunity, Pathfinder turns it into a real plan. The deadline goes into your calendar. So does the Open Day, a session for writing your application, a reminder to ask your teacher for a reference in time, and a block to prepare if there's an interview.

When an email arrives from a university or employer, an interview confirmation, a missing-document request, a scheduling note, Pathfinder reads it and updates your tracker. The status moves. The next action changes. You see what to do next without having to remember to look.

When you need to write something, a personal statement section, a "Why this course?" answer, a cover email to an employer, it drafts from what it already knows about you. You edit. You decide what's true. You send.

Pathfinder doesn't submit applications. It doesn't pay fees, accept offers, withdraw applications, or send emails without you. It doesn't invent grades or pretend a recommendation is a guarantee. It tells you when something is risky, and it tells you why.

The agent does a lot of work behind the scenes, searching, ranking, planning, drafting, watching deadlines. None of that work crosses the line into making decisions that are yours to make. Every shortlist, every draft, every email, every status update is something you can review and override.

Pathfinder is useful with one student. It is significantly more useful with a thousand. When a single student is searching, the agent searches on their behalf, already better than searching alone. When a thousand are searching, the agent searches once and routes the results intelligently. A new apprenticeship posted at 9am isn't discovered a thousand times by a thousand browser tabs; it's discovered once and matched against a thousand profiles. The students it fits hear about it. The students it doesn't aren't interrupted.

The web doesn't get bigger when more students search it, only the duplicated effort grows. An agent that searches centrally and matches individually scales without scaling the work. And there's a quieter second effect: every saved opportunity, every rejection, every late-night change of heart teaches the system how 17-year-olds actually choose, making future recommendations better for every student who comes after.

Pathfinder is designed first for the 17-year-old at the centre of the process. Everything else flows from there. But applications aren't a solo activity, even when they feel like one. Parents want to know there's a backup plan. Advisors want to spot the students who are under-aspiring or at risk of missing deadlines. Teachers want to write references for students whose strengths they actually understand.

In time, Pathfinder will offer the people around the student a lightweight view, never a surveillance tool, never a way of taking over the process, but a way of supporting it. A parent who can see three backups and one wildcard worries less. An advisor who can see thirty pipelines at once helps better. For now, the student comes first. The rest follows.

The most reasonable question an advisor asks about any AI tool is whether it is making things up. Pathfinder answers that in the interface rather than in a policy document: results carry their source, and the student can open it.

When it presents routes to compare, it names where each came from and links out to the original listing. A student who wants to check is one tap from the actual page. A teacher who wants to verify a claim does not have to take the agent's word for it, and neither does the student.

This also shapes what the agent is willing to say. Where it cannot evidence something, it says so rather than filling the gap, and where a recommendation is a stretch, it is labelled as one. The honest downside is presented alongside the reason it fits, not buried under it.

Pathfinder is not a prompt wrapped around a chatbot. It runs on Pea, a governed agent runtime built so that every tool call, retrieval and external request happens inside a bounded, auditable structure. The agent can propose. It cannot quietly grant itself permission to act.

That distinction is the reason Pathfinder can be trusted around a 17-year-old's application year. The limits described in section 05 aren't promises made in marketing copy, they are enforced below the level the model can reach. An agent that is asked nicely not to submit an application is a different kind of product from one that cannot.

Each student's work runs in its own isolated worker, with authentication handled separately from the application itself, and stored state kept apart from the reasoning that reads it.