Why the clerk's room is next

The clerk's room is the load-bearing wall of a barrister's practice and a small law firm. Fee notes, listings, briefs, conflict checks, cheques in and cheques out — the clerk holds it all in their head, in a leather diary, or in a desktop product designed for the office of 1998.

For twenty years, legal technology has aimed almost everywhere except here. There's AI for partners drafting opinions, AI for trainees doing discovery, AI for associates pulling case law. The barrister has been handed Dragon and a Dictaphone. The clerk has been handed nothing at all.

Why the clerk has been skipped

The honest answer: there's no obvious enterprise sale. Clerks' rooms are small, distributed, and run without IT budgets. At the Irish bar, barristers are sole practitioners who share infrastructure — not a chain of command with a procurement team. A six-figure annual contract isn't possible. So for two decades the venture money has gone elsewhere.

The second answer: the work is hard to model. Clerking is interruption-driven, multi-tasked, and deeply local. It depends on knowing that Mr. Cassidy SC won't take an interlocutory on a Friday because he's at his daughter's school sports, that the Smith and Jones briefs go to the same junior, that the BBA fee needs reconciling every quarter. That's not a CRM problem. It's a working-knowledge problem.

What AI has changed

Three things shifted in the last two years that make the clerk's room a real target for software:

1. Speech-to-document, not speech-to-text. The output is a structured fee note, brief, or attendance note — not a wall of transcribed words a clerk then has to format.
2. House style as a first-class input. A model can learn your headings, your fee-note format, your reference conventions, and produce work that looks like it came from your own desk — not a vendor's template.
3. EU-only deployments. Models from Microsoft Azure OpenAI now run in EU regions under Standard Contractual Clauses, with no US infrastructure in the path. Privileged material can finally use modern AI without a CLOUD Act conversation.

What this looks like on a Tuesday in the Library

A senior clerk in Dublin we work with does roughly the same thing every day at half past six. The fee notes for the day are still to be drafted; the next day's listings still need to be summarised; counsel are emailing about cancellations. The day's billable work is already three hours behind the day's actual work.

With clerk&, the fee notes start drafting themselves the moment counsel walks back from court and records a thirty-second voice memo. Attendance notes from a consult are finished before the kettle boils. The clerk reviews, edits, and dispatches — and closes the day at six instead of nine.

It's not magic — it's a first draft, every time

Worth being clear about what this is. clerk& doesn't replace the clerk's judgement, and it doesn't replace the barrister's voice. It produces a first draft, in your house style, with the structured fields populated correctly. The clerk reads, corrects, dispatches. The difference is whether you're starting from a blank page at nine at night or from a 90%-correct draft at four in the afternoon.

Most of the work in a clerk's room is editing, not authoring. AI is finally good enough to take the authoring off the table.

What to look for

• EU-only data path — Dublin storage, EU AI processing, no US transfer.
• House-style learning — the output should look like yours, not a vendor's template.
• Mobile capture — dictation, photograph of a handwritten note, paste from email. Court isn't a desk.
• Audit trail — every draft, every edit, every send, captured.
• DPA on request — Article 28 GDPR, signed and returned, not a sales conversation.

The practices that adopt this in 2026 will be twenty years ahead of the ones that don't. Not because the technology is exotic — because the previous generation of software for clerks simply gave up trying to help.