Questions,
made tangible.

The work around a newsletter.

When several teams feed one publication, the work around the writing matters as much as the writing itself.

The problem

Two separate publishing teams each needed material, contributors, and editorial decisions brought together before every issue could reach its readers.

The solution

A simple publishing tool built around the organization’s own newsletter design. Staff fill it with stories, milestones, and photos, then send the finished issue to the website and to email.

What the work involved

Editorial and contributor coordination, website and newsletter content, graphics, promotion, interviews with institutional leaders, and video production. Then a newsletter editor: a web app for assembling each issue in the organization’s own templates.

Afterwards

Staff can now publish each issue themselves.

This work informs how we approach communications tools and the editorial work around them.

Learning that fits around people’s work.

Learning sticks when it fits around people’s work instead of pulling them away from it.

The problem

People with full working lives needed learning that connected new ideas with the places they already work, without stepping away from them.

The solution

A new online learning space designed audio-first, so people can learn by listening around their workday. It sits alongside in-person sessions, shared discussion, and one-on-one mentoring.

What the work involved

Curriculum editing, session preparation and facilitation, individual mentoring, and online community work.

Afterwards

A learner can now keep up by listening, then bring their own questions to discussion and one-on-one mentoring.

This work informs how we think about learning resources, alongside our software experiments.

A public gathering, held together.

A public event depends on many people and many details staying in step.

The problem

A lecture series depended on speakers, information, web pages, and announcements staying in step, with different people responsible for different parts.

The solution

One shared plan that tracked every task, kept speaker details in one place, and kept the website and announcements in step.

What the work involved

Project coordination: speaker logistics, task tracking, website work, and event communications, with responsibilities kept visible.

Afterwards

Contributors had one organized source of information and a clear view of who was responsible for what.

Give uncertainty somewhere to go.

What should an assistant do when it cannot tell whether a message matters?

The problem

An assistant screening email cannot always tell junk from a message that matters, and a wrong guess can hide something important.

The solution

Sorting rules with three answers instead of two: junk, keep, or not sure. Anything uncertain stays in the inbox for a person to check.

What the work involved

Framing the screening problem, directing AI collaborators, and testing rules for junk, legitimate, and uncertain messages.

Afterwards

In the tests, uncertain or malformed results stayed visible in the inbox instead of disappearing. A faster model candidate failed several difficult examples, so speed alone was not enough.

One place for the work in your head.

Planning is easier when the work and the help sit side by side.

The problem

Tasks, notes, and AI conversations were scattered, so it was hard to see what needed attention now and what could wait.

The solution

A single workspace where tasks, notes, and an AI assistant sit together, like a desk with everything within reach.

What the work involved

Building software that keeps tasks, notes, and conversations with an AI assistant together in one workspace.

Afterwards

The assistant works from the same information the person sees and helps organize the next step, so the work stays visible.

Talk a thought through.

Some thinking starts out loud.

The problem

Some thoughts are easier to say out loud than to type.

The solution

A way to talk with the assistant out loud and hear it answer, without typing.

What the work involved

Software that turns speech into text, passes it to an existing assistant so it keeps the context of earlier work, and plays the reply back as speech.

Afterwards

A conversation can begin without typing. Timing, interruptions, and hearing the full response are still being worked out.

Assistants that build on one another’s work.

AI assistants are more useful when they can build on one another’s work.

The problem

Different AI assistants each work on their own, so one cannot easily pick up where another left off.

The solution

A shared toolbox several AI assistants can use. The first tool listens to a recording and writes it down, keeping the two linked. Our internal tests explore agent orchestration: coordinating several assistants around one shared task, at a small scale.

What the work involved

Shared tools several assistants can use, starting with one that automatically turns a recording into text.

Afterwards

Each transcript stays linked to the original recording, so a person can check the result.