I Built a Scheduled AI Task That Mines Tech News for Content Ideas – Here's What I Learned About Rate LimitsBuild In PublicAI

The Problem Wasn't the Idea, It Was the Setup Tax

Every morning, before I could actually write anything, I was doing the same twenty minutes of unpaid work: open a few news sources, skim what's relevant to frontend and AI tooling, decide what's worth turning into a post, and manually drop it into my Airtable Ideas base. None of that is hard. It's just repetitive enough that I kept putting it off, which meant the Ideas table stayed thin.

So I built a scheduled task: once a day, Claude scans tech news on its own and writes structured idea entries directly into Airtable — title, short description, source link, no human in the loop until I'm ready to pick one and write.

What It Actually Does

The task runs daily and does three things in sequence: pulls recent tech news relevant to my niche (frontend engineering, AI tooling, dev workflows), filters for anything that's either genuinely new or has a personal angle I could write from, then writes each finding as a record in the Ideas table with enough context that I'm not starting from a blank title when I open it later.

That third part matters more than it sounds. "Here's a headline" isn't an idea. "Here's a headline, here's why it's relevant to a frontend audience, here's the source" is something I can actually act on without re-researching it from scratch.

Where It Got Interesting: Time of Day Matters

Here's the part I didn't expect going in. After watching this run for a few weeks, not from one dramatic failure, just from comparing runs against each other, a pattern showed up: the task's output quality and reliability noticeably dips during peak hours.

This tracks with how the infrastructure actually works. Claude's scheduled tasks, whether running through Desktop, Cowork, or the cloud-based Routines system, share the same usage quota as regular interactive sessions. Automation isn't a separate lane, it's competing for the same capacity everyone else is using at that exact moment. Schedule a task for 9am Monday, and it's running at the same time as a huge number of developers starting their week, asking Claude to review PRs, debug overnight failures, and triage their own morning briefings.

I don't have a controlled experiment here, no A/B test, no logged latency numbers, just a few weeks of paying attention. But the pattern was consistent enough to change how I schedule things: I moved the task to run well outside conventional working hours, and the difference in consistency was noticeable.

What I'd Actually Recommend

If you're setting up something similar, a few things I'd do differently if I were starting today:

• Schedule outside peak hours by default, not as an afterthought. Early morning in your timezone, or genuinely off-hours, not "right when I wake up so I can see it immediately."
• Don't trust a missed run to mean failure. Desktop and Cowork scheduled tasks only run while your machine is awake and the app is open; if it's asleep when the task fires, the run gets skipped and replayed the next time you open the app, not silently lost, but not exactly punctual either.
• If reliability matters more than convenience, look at cloud-based execution. Routines, the cloud version of scheduled automation, runs on Anthropic's infrastructure independent of whether your laptop is open, which removes one entire category of "why didn't this run" debugging.

What This Actually Changed

I didn't go into this expecting an article about infrastructure timing. I went in wanting to stop manually curating a content backlog every morning. The automation itself works exactly as advertised, it writes usable, structured ideas into Airtable without me touching anything.

But the real lesson wasn't "automate the boring task." It was that scheduled AI work isn't a fire-and-forget system just because it's labeled automatic. It's still sharing real, finite capacity with everyone else using the same infrastructure at the same time, and the time you pick matters as much as the prompt you write.

If a scheduled task is producing inconsistent results and the prompt hasn't changed, check when it's running before you start rewriting the prompt. The variable might not be the instructions at all.

I still have the task running daily. I just moved it to a quieter hour, and stopped assuming "scheduled" meant "isolated from everything else happening on the platform at the same time."