I switched Claude Code over to GLM 5.2 and ran it all day. It's a 756 billion parameter open source model you can route straight into the Claude Code harness for about five times cheaper than Opus, and for most of my knowledge work it held up fine. In this one I show you what it can build, where it beats Opus and where it doesn't, and exactly how to set it up so you can switch between models per project. Here's the config I use. Drop this into your .claude/settings.local.json and swap in your own Z.ai key: "env": { "ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic", "ANTHROPIC_AUTH_TOKEN": "your-z-ai-api-key-here", "ANTHROPIC_API_KEY": "", "API_TIMEOUT_MS": "3000000", "ANTHROPIC_DEFAULT_OPUS_MODEL": "glm-5.2", "ANTHROPIC_DEFAULT_SONNET_MODEL": "glm-5.2", "ANTHROPIC_DEFAULT_HAIKU_MODEL": "glm-5.2", "ANTHROPIC_SMALL_FAST_MODEL": "glm-5.2", "CLAUDE_CODE_SUBAGENT_MODEL": "glm-5.2" }

I thought Day 2 was going to be pretty straightforward. Get Firecrawl connected, learn how MCP servers work, scrape a few pages, check the box, and move on. That lasted about five minutes. Once I had access to real web data inside Claude, I stopped thinking about the tool and started thinking about actual problems I could solve. The setup took less time than expected. What I didn't expect was how fast I stopped thinking about "the MCP server" as the thing I was building, and started just using it. The first real test wasn't a demo page — it was a problem already sitting on my desk. I needed actual product data: door styles, colors, finishes, starting prices, for close to 100 cabinet collections on a supplier's site, for an intake app I'm building for my own home-services business. Before Firecrawl, my plan was to open every page and copy it by hand. Once Firecrawl worked, I pointed it at all 97 pages with a schema telling it exactly which fields to pull, and let it run. One pass, and I had a clean, structured dataset instead of a stack of browser tabs — collection name, brand, door style, color, finish, price range, ready to use instead of retyped by me one page at a time. That's the moment the assignment stopped being "set up a tool" and turned into "the tool just did three hours of my work in one request." The credit system bit me. A search costs 2 credits, and you get 1 back by calling a separate feedback function right after — but only if you remember to. I didn't, more than once, and burned through more of the monthly allowance than I needed to before I noticed the balance dropping faster than it should. A tool only does what you actually tell it, not what you meant. While I was poking at test pages that didn't matter much. Once I pointed it at a real problem with real data going into a real product, every shortcut I skipped showed up as a real cost or a real gap. Has an assignment that was supposed to be "just set this up" ever turned into you solving an actual problem before you even finished reading the instructions?

I needed structure and i'm getting that here! Day 1: I built a newsletter on getting US work from home jobs particularly in the health care field. Attached email 3 & 4 Learned: organization is PRIMERO. My other claude ecosystem is a token consuming mess. This folder structure, sub, mds, instructions, etc is amazing. Improved: ORGANIZATION Day 2: Scraped Health care remote jobs (attached) Learned: you can really have the agents do some deep research and firecrawl is amazing but even Claudes scraper was good Case: helped guy in my network scrape spanish speaking skool niche he was interested in. Very fast Day 3: Skill built: Parallel keywords for big tasks Optimization: splits that huge task into actionable keywords I can just put in multiple Claude windows and get it to execute on various model levels depending on the need. Skill md attached Have to ask: 1) Any blindspots? 2) Feedback? Note: Didn't mean to batch this all BUT I didn't realize I had enough likes to finally post....

I keep seeing click bait videos on youtube about people "turning Claude into a hedge fund," but I haven't seen anyone using it to create and manage algorithmic trading strategies/portfolios that weren't coded in a day. My algo trading has been largely theoretical with a lot of Tradestation Easy Language and quant connect python testing, but no live strategies. Do any of you have experience with this?

I closed them in March, and I remember the discovery call very clearly. We analyzed their ↳ Website's core web vitals ↳ On-page SEO ↳ AI visibility ...and every single thing was terrible. In fact, they had a toxicity warning from Semrush ↳ Because of so many bad backlinks Their, ↳ Toxicity score was super high (super common in new brands) ↳ AI presence was downright bad ↳ Monthly revenue was ~$7,000/month (w/ paid ads) Here's what I did: ↳ Improved their on-page SEO ↳ Made their technical SEO sexy ↳ Delivered 2 brand listicles ↳ Delivered one brand review ↳ Did 10+ targeted reddit and forum posts ↳ Published 10 high value BOFU content pieces and a bunch of other updates... Outcome: ↳ MRR Increase from $7,000 to $16,000 ↳ 12% decrease in impressions (positive - we removed irrelevant audience) ↳ 64% increase in AI visibility and citations Next up: ↳ More targeted SEO (Gonna add new pages) ↳ Focusing on ICP targeting and structuring P.S. If you're also a founder trying to make a splash by getting cited by LLMs - hit me up! Only 3 spots are open for July.