Professional-services firms, which make up 13% of US GDP, run on time. Traditionally, people reconstruct their day on a timesheet, and that guesswork becomes the firm's record of what got done.
In Laurel's Act 1, the company had built an agent that automatically captures work and hands people a pre-filled timesheet at the end of the day, sparing lawyers and accountants from reconstructing it by hand.
In Laurel's Act 2, we wanted to unlock for firms the ability to see where their profitability risks were, including what the ROI from their AI spend looked like, and for individual timekeepers, where their hours were spent. If those insights were compelling enough to change someone's behavior for the better (help them hit their targets, make a case for promotion, and so on) then it might be a trade even a skeptic would welcome.
Throughout, the team held to what the company saw as its fundamental mission:
How do we ensure our honesty here about genuinely wanting to “give time back” to people comes through?
This was very much a team effort. As one of its co-leads, I helped steer the product through several iterations, each one narrowing and hardening it around the use-case that mattered most: helping firm leaders see the risks to their profitability in fixed-fee and time-and-materials matters. The information architecture and the data visualizations at its heart were the part I owned most directly, built as part of a new design system that worked across a firm's various ranks.
Narrowing to the use-case. Getting there meant staying close to users—1:1s, on-site deployments, weekly prototypes, and designs we pressure-tested with users before committing engineering to them. Each iteration cut what didn't serve a firm leader's core question and sharpened what did.
Designing for agents. A design system used to exist to guarantee the quality attributes you cared about—usability, readability, consistency—by getting humans to adopt a set of patterns. The bigger shift now is that we design for agents as much as for people.
Whether it's a data visualization component or a UI pattern, it should be packaged so someone joining can point their agent at the project and spin up a new "look-alike" view or set of components. New colleagues become productive almost immediately, because the design system is tuned for the agents we all now use.
Democratizing a specialist skill. A skills-enhanced agent can bring non-experts up to speed quickly. I wrote Claude skills that let non-experts create data visualizations that are polished, on-brand, and consistent with a shared grammar of graphics, so that high-quality dataviz is no longer bottlenecked on a specialist.