Introducing Underleau, a tool for the rest of us

You have read the report. You have opened the spreadsheet. You have a few notes from a conversation that changed how you see the problem. Somewhere in those things is an answer, or at least a better question.

Getting there is the work.

It rarely looks like the polished demonstrations of artificial intelligence. There is no perfect prompt waiting to be typed. The information is incomplete. Two sources disagree. An important detail lives in a document you almost forgot about. Before you can ask for an answer, you have to work out what deserves to be in the question.

This is the space we want Underleau to serve.

We are still building it. What we can introduce today is the belief behind it: useful AI should make thoughtful work more accessible, and leave people better able to understand the decisions they make.

The work between the tools

Software has given us remarkably good places to put things. Documents hold our writing. Spreadsheets hold our numbers. Project trackers hold our tasks. Messaging apps hold our conversations.

Understanding lives across all of them.

A customer concern becomes a support ticket, then a meeting note, then a line in a proposal. Each tool preserves a piece. The person making the decision has to reconstruct the relationship between those pieces, often while the next message is already arriving.

Microsoft's 2025 research on the expanding workday, based on Microsoft 365 activity and a survey of knowledge workers, describes work spread across messages, meetings, and hours that used to sit outside the working day. It is a view of a particular population, but it gives shape to a familiar experience: being busy communicating can leave little room to think.

Our reading of that problem is that making each individual task faster will only take us so far. We also need better ways to hold the bigger picture together.

AI has changed what is possible

The excitement is justified. Being able to ask a machine to explain something unfamiliar, compare arguments, or help articulate a half-formed idea is a meaningful change in who can do what.

There is real value in automation, too. Routine work should not consume a person's best hours simply because it always has.

But the most interesting question for us is what happens to the work that still needs a person: interpreting a situation, deciding what matters, noticing a contradiction, and taking responsibility for a conclusion.

Anthropic's Economic Index research distinguishes delegated tasks from collaborative uses of Claude, and finds substantial use of both. Those observations describe one provider's users rather than the whole economy. They nevertheless suggest a useful distinction for product design: people use AI to do things for them, and to think through things with them.

We want to pay particular attention to that second relationship.

An answer needs somewhere to belong

A conversation is a wonderful way to begin. You ask a question, get a response, and follow your curiosity.

As a project grows, you also need to see what you already know. Which source supports this claim? What changed since the last discussion? Is that conclusion grounded in evidence, or did it just sound convincing enough to keep?

A flowing conversation can make those relationships difficult to revisit. The response may be clear while the path to it remains obscure. And generating more text can add to the pile you were trying to understand in the first place.

Consider someone trying to understand why customers are leaving. They might have interview notes, a usage report, a pricing proposal, and an internal discussion. A summary of each is useful. The harder question is how they fit together. Does the report support what people said? Are two customers describing the same problem in different language? What evidence is missing?

That work needs room for comparison, uncertainty, and revision. It needs a place where a promising idea can sit beside the material that supports it, and beside the material that challenges it.

This is why we are drawn to a canvas. A visible arrangement can help you hold several things in mind at once. It can let a question remain open while you explore its connections. The aspiration is to make the structure of your thinking easier to inspect and return to.

The rest of us

There is a lot of attention on what technically sophisticated teams can build with AI. That work matters. But the ability to benefit from these systems should reach well beyond people who can write code, assemble an automation pipeline, or spend their evenings keeping up with model releases.

The rest of us have difficult work, too.

A researcher trying to reconcile conflicting accounts. A small business owner deciding where to invest limited time. A student learning enough to challenge an explanation. A project lead piecing together what happened before they joined. Each needs a way to bring their own context into the work and exercise judgment over the result.

They should be able to begin with something ordinary: a file, a question, a few things they want to understand. Useful assistance should meet them there.

Accessibility also means respecting cost and attention. A tool that requires constant configuration or makes every experiment feel expensive puts a new barrier in front of the people it promises to help. We want Underleau's design and business model to take those constraints seriously.

Room for judgment

The quality of an AI tool cannot be measured only by how quickly it produces a finished-looking answer.

Sometimes the valuable result is discovering that you do not have enough evidence. Sometimes it is realizing that you asked the wrong question. Sometimes it is a disagreement made clear enough that you can finally investigate it.

Those are productive outcomes. They are also easy to overlook when every interaction is designed to end with confident prose.

For Underleau, the aim is to keep people close to the substance of their work. Sources should remain visible. Questions should remain revisable. AI assistance should help someone examine a conclusion rather than encourage them to accept it because it arrived fluently.

Trust belongs in that vision from the beginning. People bring sensitive work to the tools they rely on. Our commitment is that customer data will never be used to train models. As we build, we have to carry that commitment through the services and choices behind the product.

What we are building toward

Underleau is our attempt to create a calmer place for this kind of work: bringing material together, exploring how it connects, and developing an understanding you can explain to someone else.

The product is in development. The canvas is a prototype, and there is substantial work ahead before production AI, storage, and connections are ready. This introduction is a statement of direction. The useful test will be what people can actually do with what we build.

We hope that means fewer hours spent reconstructing context. More opportunities to notice what does not add up. An easier way back into a question after a week away. A little more confidence that comes from understanding your evidence.

There is room in this industry for tools that take those ordinary ambitions seriously.

That is the reason for Underleau, and the work ahead of us.