Groundwork

Ideas

The frameworks behind the writing.

Location, quality, classification

Location, quality and classification as three panels Location One home for eachkind of knowledge Quality Someone has said it'sgood, final or approved SPA English law Corporate Classification Basic labels: type,jurisdiction, practice

Before a firm points AI at its knowledge, three things need sorting. Location means each kind of knowledge has one home, so people and agents know where to look. Quality means someone has decided what is good, final or approved. Classification means each document carries basic labels, such as type, jurisdiction and practice area. AI raises the cost of getting any of them wrong, because a fluent answer hides a weak source.

You can't tell a document is final by reading it

Five versions that look the same; only the sender knows which was the execution version v1 v2 v3 v4 v5 ???? execution version Only the person who sent it knows

Whether a document is the signed version, or approved as know-how, can't be read from its four corners. Only the person who sent or approved it knows. So curation needs a signal from a human, and the best place to capture it is a step lawyers already take, such as sending the execution version.

Chat is another way into search

A chat question becomes a search query, which returns a list of results with filters Which precedent didwe use for the lastbank-side facility? facility · lender Chat writes the query Search runs it A list you can check

Chat drafts and runs search queries for you. For a quick lookup, that saves time. For research a lawyer has to stand behind, they need what search gives them: filters, a list of results, previews and some confidence they've seen everything.

Seven questions before using AI on a task

Seven questions before using AI on a task 1Will I learnfrom it? 2Howmechanical? 3Creative orstrategic? 4How much doesaccuracy matter? 5Cost ortime-sensitive? 6Must it beconsistent? 7Will the readercare a human did it?

AI native should mean deciding task by task, with seven questions: (1) will you learn from doing it, (2) how mechanical is it, (3) is it creative or strategic, (4) how much does accuracy matter, (5) how cost or time-sensitive is it, (6) does it need to be consistent, and (7) will the reader care that a human did it? Most debates about AI in legal work fix on one of these and ignore the rest.

Automate the keyboard, not the brain

Vendors aim at drafting and research; the groan-worthy keyboard tasks go unsolved Where vendors aim drafting · research What makes lawyers groan formatting · cross-references · time narratives · filing

Vendors aim AI at drafting and research, the interesting thinking, and call it drudgery. Meanwhile formatting, cross-references, time narratives and filing go unsolved. Those are the tasks that make lawyers groan. They're easier to automate, and they get used.

The zone of apathy, and push and pull

Productivity dips below today's level while people learn a new tool, before gains arrive. Push and pull get them across. time today zone of apathy gains push: who's telling me? pull: what's in it for me?

Every new tool makes people slower before it makes them faster. A lawyer focused on today's matter won't cross that dip without a reason. So for each audience, ask who is telling them to change (push) and what's in it for them (pull). The answers differ for partners, associates and trainees.

Skills are knowledge assets

Precedents, prompts, playbooks and agent skills are all knowledge assets needing the same care Precedent Prompt Playbook Agent skill classification curation maintenance someone who owns it

Prompts, playbooks, tabular review templates and now agent skills are all knowledge. They reach further into how lawyers work than any document did, so they'll multiply and vary. They need the discipline any other know-how needs: classification, curation, maintenance and someone who owns them.