 | Something I didn't expect when I started building with AI agents: the interface problem. My agent handles 15+ automations, runs night shifts, processes tasks across CLI, Discord, email. It's capable. But I had no way to see what it was doing without asking. And asking "what's your status?" every time is not a real workflow. It's a workaround. Humans process information visually. We scan, we group, we notice patterns at a glance. That's not how agents communicate. They give you text. Logs. Summaries. And when your agent is doing 20 things in parallel across 5 channels, text stops scaling. So I built a custom visual dashboard. Kanban board, real-time updates, native apps for macOS and iOS. Three platforms. 54 commits. It worked for about 6 weeks. Then I hit what I'd call the productivity paradox of AI agents: the more capable your agent becomes, the more things happen, and the more you need from your interface. I was adding features to keep up with the agent. Every feature added maintenance. Every simplification broke something. I was spending more time on the dashboard than on the actual work the agent was helping with. The fix wasn't building better custom software. It was finding a solid open-source foundation (in my case, Fizzy by 37signals) and building only the integration layer on top. A 94-line adapter between my agent and the board. That's the custom part. The board itself shouldn't be my problem. Two things I learned: I wrote up the full journey for anyone thinking about this problem: https://thoughts.jock.pl/p/wizboard-fizzy-ai-agent-interface-pivot-2026 Curious: for those of you running agents beyond chatbots, how do you keep track of what they're doing? [link] [comments] |
