I test AI tools, build real workflows, and write down what's actually worth your time - costs and dead-ends left in. Everything's based on what I've shipped, not what I've read about. Past the hype and the doom.
SoN 2.25: Stop guessing and use AI to pull the data
Apify is like a giant app store of (at time of writing, just north of 44,000) robots that can visit websites on your behalf, collect information, and plug that data into your tools and AI agents so you don’t have to do it manually. One of them reads a public Instagram account: the profile, the recent posts, the likes and comments on each. I pointed it at the rescue’s account and pulled the last fifty posts, then ran it again on three other cat rescues of a similar size to compare against. The whole thing cost under a dollar and took an afternoon. Then I handed the raw numbers to Claude and asked it to look for patterns. And the content wasn’t the main problem. The link in the account’s bio — the one every caption points to, the one that says “fill in the form, link in bio” — went to a donations page instead of an adoption form. So: someone reads a post, falls for a kitten, taps the link to apply, and lands on a “give us money” page. The entire adoption form was unreachable from Instagram. Nobody would have caught that by scrolling the feed, because the feed looks fine. It only showed up when you traced the actual path a real person takes. I called on Claude again to do some research of competitors in the region, to see how they were performing and what they’re doing differently. One of the other rescues with the same kind of shoestring setup, no building, fostering cats in volunteers’ homes had three times the followers on fewer posts. They weren’t spending more. The data showed what they did differently: they follow one cat’s story across several posts, they end each post with a single clear action instead of a menu of donation options, and they celebrate every adoption. You can argue with an opinion about good content; you can’t argue with three times the followers on thirty percent fewer posts. And the total cost to run the Apify bots was about $0.04. Diagnosis and reportingA diagnosis stuck in my head is no use to the rescue. So I had Claude turn the data into a short report they could act on: the problems in plain language, the fixes ranked by impact, and a simple content plan to follow. Because the team works in another language, I had it produce the report bilingually in one pass — a polished version in their own language rather than a rough translation they’d have to redo. The afternoon’s scrape became a plan they could pick up and run with. Try it on a question you’d otherwise guess atMost of the questions a small business guesses at are sitting in public data: how are my competitors actually pricing, which of their posts land, what do their reviews complain about, where does my own funnel leak. You used to either guess or pay someone to find out. Now a cheap scraper pulls the real data and an AI reads it back to you, for about a dollar and an afternoon. Apify’s store has a ready-made scraper for most public sites. And if you’re handing the job to an AI agent rather than running it yourself, Apify exposes the whole catalogue through an MCP server, so the agent pulls the data and reads it back without you leaving the chat. Don’t scrape to confirm a hunchThe skill is knowing when to bother — because scraping a feed to confirm something you already believe is just a slower guess. The test I’d use: is the question measurable?
those have answers sitting in the data, so go and get them. A vague “should we rebrand” doesn’t, and no amount of scraping will settle it. And watch out for the trap the cat rescue fell into. When results disappoint, the cause usually turns out to be something structural nobody thought to check: a link pointing at the wrong page and a form no one reaches. - Jim Just launchedTalking Utilities went live this month — a daily news-and-intelligence publication I designed and built end to end for a niche UK energy-sector publisher.
Read the full case study at Solopreneur Superpowers → Paid Member’s AreaSecond Brain ChroniclesRebuilding a year of lost receipts without doing it by handThe bookkeeping version of the same problem. My invoicing system had a year of expenses logged with the receipts missing — 36 of them — which is the kind of thing that stays invisible until it becomes a tax-time emergency. So I pointed a few agents at it. Codex read through my mail archive and pulled out 19 receipt PDFs, which is fiddlier than it sounds: most software doesn’t bill you in its own name, it bills through Stripe or Paddle, so you match on the vendor and the date and never the amount. A headless browser rendered three more that only existed as web pages, and the whole lot got filed into the right folders automatically. The shape is the same as always: the agents do the boring archaeology and I do the judging. It also ran into a wall worth respecting — a browser robot can’t safely log into accounts unattended, and knowing when to stop one is part of the job. → This and two other full write-ups are waiting on your Member Hub:
A tool I actually use: this newsletter’s member hub and most of my task system run on Notion. If you’re a startup or a solo founder, you can get 3 months of Notion Business free, with unlimited AI and no credit card, through my link. Try Notion free for 3 months → Notion affiliate link — I get a small credit if you start a trial, and I only run ads for tools I actually use. You’re reading the free edition of Signal Over Noise. Paid members got a build log this week in the Member Hub:
Paid membership is €7 a month or €69 a year, and you get:
Signal Over Noise helps you sort the AI worth your time from the hype. If it was useful, pass it to someone who’d want it too.
| |
Jim Christian - Signal Over Noise
I test AI tools, build real workflows, and write down what's actually worth your time - costs and dead-ends left in. Everything's based on what I've shipped, not what I've read about. Past the hype and the doom.
Signal Over Noise Vol 2, Issue 29 | Jul 24 2026 Don’t have time to read this week’s issue? Why not copy/paste it into your AI agent and ask it for insights? My AI already gets into my password manager. It has done for months — but on purpose. Every morning something on my MacBook Air reaches into 1Password, pulls out one scoped token, and gets on with the job — filing a note, checking a balance, handing me a draft to look over. I never paste a key into a script. The AI agent asks for the one...
Field Note: Tell AI What You Want and Who You Are Two things did the rounds this week and caught my eye as they’re closely related: First, a post laying into how everyone prompts their coding agents — “no em-dashes,” “stop saying delve,” “don’t sound like AI” — with one flat retort: you never gave it a writing system. So give it one. Their pick: ASD-STE100 Simplified Technical English. If you’ve not met it, STE is a 'controlled English' that the aerospace industry built in the late 1970s and...
Signal Over Noise Vol 2, Issue 28 | Jul 17 2026 Don’t have time to read this week’s issue? Why not copy/paste it into your AI agent and ask it for insights? When I think about AI and automation, I see an opportunity opening up that goes well beyond the chatbots everyone’s being handed — well beyond prompting, too. One of the most important things I ask my AI setup on every project is this: "That’s good, how can we make it better?" Case in point — my friend Marcus is a working actor, and he...