Hello!
Recently I've been thinking about what we heard at our Crazy Agent Scale event. Some of the numbers were hard to ignore: one agent with one system prompt now does nearly every job at Sentry, and another company passed $100M in ARR with 21 people and no support, finance, or marketing teams.
So what does that mean for product design? When agents are doing that much of the work, the interface problem moves to managing them. AI products went quickly from models behind the scenes, to chat, to agents picking their own tools, to agents orchestrating other agents. Each of those shifts opened up room for new products and new UI.
One line from the event stuck with me: handing agents a shared spec, a task graph, and a polite retirement plan. That lines up closely with what we've found people need to do with agents: start them, schedule them, scrutinize their work, steer them when they go off course, stop them, and see, share, and save their results. The "polite retirement plan" part (knowing when to stop an agent) tends to get the least design attention.
Steering also has to be fast. That's why we built Intent's notch, so you can grab screenshots and videos, mark them up, and send them to the relevant agent to fix UI, bugs, and more without opening an app.
If a team of 21 people can run a $100M business with agents, the tools those 21 people use to direct and check their agents need to be very good. That's where I'd put design effort right now.