I Stopped Reading the News. My Agent Does It for Me.

I haven’t opened a news tab in six weeks. I’m not on a digital detox. I’m not doing some mindfulness thing. I just built a system that reads the news for me, filters it, and surfaces what actually matters. Everything else gets logged and ignored.

This is a solved problem if you’re willing to build the infrastructure. Most people aren’t. Most people are drowning in push notifications, algorithmic feeds, and the low-grade anxiety of feeling like they might be missing something. The solution isn’t discipline. The solution is delegation.

Let explain what I mean.

Logging vs. Alerting in Information Systems

There’s a pattern I keep coming back to in how I think about agent tooling: the difference between MCP inline tools — tools that run in the main conversation context — and subagent-scoped tools that run independently in their own context. It maps cleanly onto a distinction that matters a lot in information systems: logging versus alerting.

Logging means record everything. Alerting means notify on signal.

Most people’s information diet is all alerting, no logging. Every app on their phone has notification permissions. Every newsletter lands in their inbox. The signal-to-noise ratio collapses and they start making decisions about what to pay attention to based on what’s loudest, not what’s important.

My setup inverts this. Scheduled NanoClaw tasks run on a cadence — monitoring X/Twitter for relevant AI research, tracking product releases, watching for papers in specific areas. Everything gets captured. Almost nothing triggers a notification.

Designing Alert Logic and Passive Cadences

The alert logic is the part that took me longest to design. What actually warrants interrupting me? I landed on something like: novel capability claims with code attached, API-breaking changes to tools I use, and anything that’s getting enough signal from people I trust that it’s clearly not noise. Everything else goes into a log file that I can query if I want to, but I don’t have to.

The honest part: this is hard to build well. The failure mode isn’t missing things — it’s getting over-alerted until you start ignoring the alerts, which puts you back where you started. I’ve tuned this system through multiple iterations and I still sometimes get batches of notifications that make me want to turn the whole thing off.

What actually works is passive monitoring. The agent doesn’t push to me; I pull from it when I want a briefing. There’s a difference between “I will tell you when something matters” and “here is everything I found, go through it.” The first is a system that respects your attention. The second is just outsourcing your inbox management.

Attention as a Finite Resource

For an ADHD brain specifically, the passive monitoring model is almost medically necessary. My attention is a resource I have to protect. Every interruption has a cost. A system that checks things on my behalf and only escalates when the signal is actually there isn’t a luxury — it’s infrastructure for functioning.

The real insight I keep coming back to: you don’t need to read less. You need to delegate filtering. The information isn’t the problem. The lack of structure around how it reaches you is the problem.

Building this took time. It required being honest about what I actually needed to know versus what I was reading out of anxiety or habit. It required designing alert logic that didn’t collapse under its own false-positive rate.

But I haven’t opened a news tab in six weeks, and I’m better informed than I was when I did it every morning.