There are tons of ways to get leads/clients, curious to hear what you guys think is the scariest. There are no right or wrong answers here.

I just love seeing what the smaller free models are capable of and how one can use them to save on tokens! My results from using Opencode and their free AI models. What I scraped: Pulled structured job listings from DailyRemote (10 jobs as CSV), grabbed full-page screenshots + branding breakdown of OpenClaw.ai, and mapped 370 URLs across Blue Bottle Coffee's entire US site for competitive analysis. Firecrawl did all the heavy lifting — I just pointed it at URLs and said "go." One thing I learned: I was surprised how well the free AI also picked the right Firecrawl tool without me spelling it out. When I said "I want structured data for a spreadsheet," it reached for `firecrawl_scrape` with JSON format + schema instead of markdown. When I said "map out this site," it went straight to `firecrawl_map`. When I asked for branding, it knew to use the `branding` format. It's like it actually reads the cheat sheet. Wild. One use case idea: Competitive auditing for e-commerce clients — map a competitor's entire product catalog, extract pricing and categories as structured data, and track changes over time with monitors.

Hi All, I let Claude Code loose on Firecrawl 🦞☕ What I scraped: Went on a bit of a tour, tried to scrape an AI automation site that turned out to be one (1) chat bubble on a blank page 🤣, pulled 50 structured job listings from a remote job board into a clean CSV, screenshotted + reverse-engineered the entire brand system of an AI agent startup, and mapped out Blue Bottle Coffee's site (turns out it's 75% cafe location pages, 😭RIP my "coffee brand" theory). One thing I learned: Honestly didn't expect Claude Code to reason about which Firecrawl tool to reach for like a debugging flowchart — scrape fails on a JS-heavy site → try waiting longer → check the raw HTML (`<div id="root"></div>`, thanks for nothing) → confirm with a blank screenshot → THEN escalate to the autonomous browser agent. It didn't just retry blindly, it diagnosed why the page was empty before picking the next tool. My own debugging process is less structured than that. 😁 One use case idea: Competitor/market monitoring for a small business — point it at competitor pricing pages or job boards (hiring trends = who's investing where) on a schedule, get a clean CSV instead of me manually copy-pasting like a caveman. Firecrawl even has a built-in `monitor` tool for exactly this — diff-based alerts when a page actually changes.

As promised I completed the same task using Opencode and their free AI model "Big Pickle" What I built: An automated newsletter that writes itself, wraps it all in a slick dark-theme email, and sends it — all in one command. One thing I learned: OAuth scopes are a nightmare. Google silently drops permissions during token refresh if your consent screen isn't configured exactly right. I spent half the day debugging a token that kept reverting to gmail-only access — turns out the Gmail client was loading the token with a subset of scopes, which nuked the others on refresh. The fix: every service that touches the token needs to request ALL the scopes, not just the ones it needs. Classic "one line fix, four hours of debugging" energy. One thing I'd improve: I'd love to add RSS feed monitoring so the newsletter automatically picks up trending topics throughout the week instead of me manually picking one. Basically let it surf the news on its own and come back with a draft ready to go — true "set it and forget it" mode. Big Pickle is actually legit. While I was yelling and swearing at it, it created the whole pipeline, fixed the bugs, and got the system running. Not bad for a free model going up against the big guns. One just needs to give some extra info.

What I built: A newsletter machine that does the thing I'm way too lazy to do myself — research a topic, write the copy, generate matching infographics, and drop a fully designed HTML email as a Gmail draft, personalized per subscriber. Today it cranked out an issue on "AI Agents as Friends," complete with a dark-mode design because apparently I have opinions about that now. One thing that clicked: I learned that Gmail's draft editor has a mind of its own — it quietly reshuffled my inline images so the logo showed up where an infographic should've been and vice versa, like it was playing a prank on me. Turns out Gmail doesn't like short, lazy image IDs and just remixes them if they don't look "official" enough. Gave them fancier names, problem solved. Moral of the story: even your email client has main character energy. One thing I'd improve: At one point every subscriber could see everyone else's email address in the same draft — which is a great way to accidentally start a group chat nobody asked for. So I would like to start a what to look for process that will create a database to store little things like this that can cause big issues.