Introducing: the Less Assistant

If your job is turning data into answers for the rest of the business, the best part is the analysis itself - digging into the data, working out what it's telling you, shaping it into something that answers the question in front of you. That's the part you're good at, and the part worth your time.

The friction is everything between you and it: finding the right source, checking where a figure came from, working out why last night's run didn't finish, connecting a new tool, keeping schedules and access in order. None of it is the analysis. All of it slows you down on the way there.

A lot of AI tools respond by trying to do the analysis for you. We think that's backwards. The assistant works the other way round: it speeds up everything around the analysis, so you reach the part that matters sooner.

What it is, and what it does

The assistant is a chat panel inside Less. You ask it something in plain language, and it works across your workspace to answer or to act.

Most of what it does is read-only. It can find any source, model, table, or orchestration and explain how they connect; trace a number's lineage upstream and down; read the change history on anything to work out when something broke; give you a health overview of recent runs and read the logs down to the step that failed; and answer questions about your verified tables, with a plain-English explanation of how it got there.

When you ask, it can also make changes: run a model, set or clear a schedule, edit configuration, reorganize folders, connect a new source or destination, or manage who has access. Connections are set up end to end, and credentials go into a secure form the assistant never sees.

A faster path, not a different one

Nothing moves behind a prompt. Everything the assistant does, you can still do yourself - same screens, same clicks, exactly as before. It's simply a quicker way through the parts of the job that slow you down. When asking is faster, ask; when you'd rather do it by hand, do it by hand.

Ask instead of digging

Much of the friction is just looking things up: open the source, check what ran, find the table behind a model, trace where a number came from. Each lookup is a minute or two, and they add up across the day.

So ask instead. "Where does this number come from?" and the assistant follows its lineage back through every step it passed through. "Why is this stale?" and it checks what ran, what failed, and which step broke. "What's in this table?" and it reads the columns, row counts, and a sample, so you know what you're working with before you build on it.

It can do the work around your analysis, too

The assistant isn't read-only. Ask it to run a model now, change a schedule, connect a new source or destination, reorganize folders, or manage who has access - and it carries it out, without you hunting for where each setting lives. As always, you can still do any of it yourself; the assistant is just quicker. Anything that changes your workspace happens at your request, and credentials are entered in a secure form the assistant never sees.

Why it understands your setup

Connecting an AI to a tool is straightforward. Getting it to reason correctly about your data is the harder part.

Two things make it work. Less keeps a full record of what actually happened to your data, not just metadata describing it. And because Less is built around a visual canvas rather than code, every transformation is explicit - there's no hidden SQL to misread. The assistant can see exactly what each step does, so its answers hold up.

What's next

The one thing the assistant doesn't do yet is help on the canvas, where the models are built. That's not a line we've drawn for the first release - but it's the next thing to build.

When canvas support lands, it'll work like the rest of the assistant: a faster way through the work, with you still deciding what the data means.

The assistant is available now.