Stop Asking AI Questions. Start Building Workflows.

Picture it. A junior barrister in London, grounds for a judicial review claim neatly drafted, authorities lined up like soldiers on parade. Five of those soldiers, it turned out, had never been born. In June 2025, in Ayinde v London Borough of Haringey [2025] EWHC 1383 (Admin), the High Court of England and Wales had some pointed things to say about it.

What the court did not say was “put the robots away”. Its message was quieter and far more useful: if you use AI for research, you still answer for whatever it hands you. It’s an English decision, so it binds nobody here, but I struggle to picture a Barbadian judge taking a more relaxed view of cases that don’t exist.

The lesson I take from it is this. The trouble wasn’t the tool. It was the absence of any process around the tool.

And that, I suspect, describes how most of us use AI. We open a chat window, type a question, copy the answer and get on with our day. It’s a bit like paying for a Foreday Morning package, collecting your paint and your drinks, then spending the whole morning at the side of the road watching everybody else jump. You were technically there. You just missed the party. Junior attorneys, I’d argue, have the best seat in the house to do something more interesting: to stop asking AI questions and start building it into the way we work.

From tourist to resident

There’s a difference between visiting a tool and living with it. The tourist pops into a chat window when a task comes up and leaves the moment it’s done. The resident sets up a routine: a handful of saved prompts, a couple of templates, a checking step at the end. One query gets you a paragraph. A workflow gets you your Thursday afternoon back, every Thursday, for as long as you keep it running.

Oddly, being junior is an advantage here. Our habits haven’t set like concrete yet. Our diaries are packed with exactly the repetitive work machines would enjoy, if they enjoyed anything. And hours are never scarcer than at the start of a career.

What that looks like on a Tuesday

Plenty of legal work is necessary, time-consuming and, let’s be honest, not what anyone daydreamed about in law school. That’s where AI earns its keep, as long as a lawyer is looking over its shoulder.

• The blank page. A prompt built from your own template and the facts of the matter can turn out a first-cut clause, a draft pleading or a client update before your coffee cools. It won’t be finished. It will be started, which on a Friday at four o’clock is most of the battle.
• Research, as a scout. AI makes a decent scout and a terrible witness. Let it suggest where to look, then go and look. In my experience, most general-purpose tools know a great deal more about English and American law than about ours, so every Barbados statute and case goes back to the source before it goes anywhere near your signature.
• The mountain of paper. Disclosure bundles, witness statements and hearing transcripts can be indexed and summarised in a fraction of the time it takes to read them. Treat the summary as a treasure map. It shows you where to dig. You still have to pick up the shovel.
• The first conversation. A structured intake form, built once, gathers the basic facts, key dates and documents before a client ever sits down. You walk into the meeting already knowing the story, instead of spending twenty minutes piecing it together.

Build each one once, then reuse it. That’s the whole trick.

Before you paste anything

One question belongs at the top of every workflow: what is this tool allowed to see? Dropping a client’s file into a free chatbot can mean handing it to a stranger on terms nobody read. So read them. Check what the provider keeps and whether it trains on your data, choose tools with proper business-grade protections and strip out names where you can. Confidentiality doesn’t take the afternoon off just because the work got quicker.

From doing to directing

Something shifts once these workflows are humming. You spend less time producing first drafts and more time improving them, less time reading every page and more time deciding which flagged passages matter. Gradually you move from doing the work to directing it, and the directing part, the judgement, is the one piece no software has managed to bottle.

People notice, too. The associate who can walk a senior colleague through a tidy drafting or research process is offering something rather different from the one who simply stayed later.

None of this makes the law any easier to know. AI can be fluent, confident and entirely wrong, citing cases with the serene assurance of a stranger giving directions to a village that isn’t there. The faster the drafts arrive, the more of your job becomes checking them. That part stays yours.

The cost of sitting it out

The tools are cheap, everywhere and getting better by the month. Access stopped being the obstacle a while ago. Treating AI as an occasional party trick can feel like the safe, sensible choice. It isn’t really neutral, though. It simply means doing the same work more slowly than the colleague down the corridor who built a system.

Which brings me back to those phantom cases in London. They were never really a warning about technology. They were a warning about using it without a plan. Build the plan first, with the checking step baked in, and the tool becomes what it should have been all along: a very fast junior colleague who still needs supervising. Much like the rest of us, once upon a time.

Javed E. Daniel (LL.B) (LEC)