I want to use this monthly note to do something other than simply list the features we shipped in May. The list is at the bottom for anyone who wants it. What interests me more is the argument the list adds up to.

For two years the working assumption in legal AI has been that the models are the constraint. Spend any time with in-house teams and you learn the opposite. The models are fluent. What they lack is context, the things a lawyer reaches for before signing. There are three: what the company has agreed to before (its precedent), what the company has decided it wants (its standards, written down as a playbook), and what the wider market will actually accept. Almost everything we shipped in May was an attempt to put one of those three closer to the point of decision.

‍

Market Standards now live in Microsoft Word

The clearest example is Market Intelligence, which we released into the Simple AI (and Law Insider) Microsoft Word add-in this month. It lets you take the clauses in front of you and compare them against Law Insider's benchmark data: what is common, what is unusual, what is missing, with the reasoning shown rather than asserted.

Our advantage here is the data underneath the AI. Law Insider has spent years building the world's largest structured corpus of public contracts, and that is hard to copy. Other tools have shipped (or claimed to ship) market comparison features before us, and some are further along on the analysis. What we have shipped is an early version anchored to a stronger foundation. I would rather say that plainly than oversell a first release. For many of our customers, who know and trust our data layer, this is fast becoming their favourite feature.

‍

Internal standards at the centre of your general reviews

The largest single change this month was General Review, which replaces the old playbook-driven flow with one organised around issues. Each issue becomes something you can act on directly, without waiting for every rule in the playbook to finish first. For a change that took a lot of work, it moves the playbook from a checklist running in the background to the thing the review is actually built around.

Two related releases serve the same idea. Preferred Language lets a team save its exact approved wording and have the system insert it during review, unchanged, with no AI rewriting. For many legal teams the wording is the standard, and they do not want a model paraphrasing it.

‍

Custom playbooks available to every user

Including legacy Law Insider customers. Supporting these, redline suggestions now read the full document rather than a single paragraph, first-party paper has its own review strategy, and the rules that flag specific conditions trigger more consistently across however a clause happens to be phrased.

‍

The repository is where organisational precedent lives

The third kind of context is the one almost no tool handles well: what your own organisation has already agreed to. Most of our repository work in May was aimed at this.

Document Relationships uses AI to detect and map how agreements connect, including amendments, renewals, supplements, and parent and child chains, with the structure visualised and governed metadata flowing across the chain. An amendment that updates an expiry date should update the parent agreement's expiry date, automatically. We all need that type of quiet bookkeeping within our document repositories, and now we have it.

Around it: you can now scope the repository's AI chat to a saved view, so a question runs against exactly the set of documents you mean; you can set bulk, date-based alerts across every document that matches a view, which is how renewals stop slipping; and the extraction underneath improved in accuracy for the fields people actually filter on, with company-level language settings for teams whose documents are not in English. None of this is glamorous. A repository is only as useful as its data is correct, and most of the month went into that.

‍

What is still early

Market Standards is a first release, and the analytical layer will take time to mature. Document Relationships detects with high confidence but still asks for human confirmation, which is the right default for now.

‍

Context is the part that has to be earned

If there is a thread here, it is that we are spending our time on context rather than on the model. The model is the part everyone can buy. The context is the part that has to be earned, document by document and standard by standard. That is the slower work, and I think it is the work that decides whether any of this is actually useful to a legal team.

‍

Also shipped this month

• Request API: attach support documents and optionally merge them into the final signed PDF.
• Custom NDA: assign the correct internal signer through a new API field.
• SharePoint and OneDrive: large folders import instantly and run in the background.