China has just released GLM 5.2, and Iβve tested it against Claude Opus 4.8 and Kimi K2.7 on identical tasks
same prompts, same tasks, same rules
GLM 5.2 won 4 out of 5 tests
> best endless runner
> best product landing page
> best simulation
> best interactive experience overall
And thatβs not even the most interesting bit
> 1M context tokens
> agent support
> lower prices than the competition
The models coming out of China right now are improving at an absurd rate
Most people donβt realise just how competitive this race has become
ONE-STAR GOOGLE REVIEWS ARE QUIETLY CREATING A $5K/MONTH AI SERVICE
A dental clinic may see a 4.1 rating and assume everything is fine, while hundreds of reviews are describing the same missed calls, long waiting times and unexpected charges
Claude can read the entire history, group recurring complaints, estimate where revenue may be leaking, then prepare replies, updated FAQs, staff instructions and a weekly report for the owner
> Outscraper collects the reviews
> Claude turns them into decisions
> n8n keeps the process running every week
The business has already paid for this data through disappointed customers.
The full article shows how to package it into a recurring service
A $20 CLAUDE SUBSCRIPTION CAN TURN INTO A FULL AI TEAM IF YOU STOP USING IT LIKE CHATGPT
Most people still use Claude like a smarter search bar
Ask, copy, close, repeat tomorrow.
Skills change that
A skill is just a folder with a SKILL.md file, but inside it you can package an entire workflow once: PRDs, refactor plans, GitHub issues, code review, TDD, docs, marketing research, SEO, sales strategy and multi-agent orchestration
That is not "better prompting"
That is installing labor
The article lists 50 Claude Skills with repos and install commands, from Anthropicβs official collection to Matt Pocockβs skill library and SkillsMP with 66k+ community skills
The useful part is not the list
It is the shift from asking Claude to remember your process to giving Claude the process already packaged
You do not explain the same workflow 50 times
You encode it once
The model provides intelligence
The skill turns it into labor
Check full article below
AI has made LARPing 10 times cheaper and 10 times more profitable at the same time
Previously, you had to hire a Lamborghini for $3,000 a day, hire a cameraman for $500-1,000, and spend $3,000β10,000 on a single dayβs filming
Now, a single person with a laptop can do the same for $400 a month
Claude writes scripts for 30β50 videos in a matter of minutes
ChatGPT generates photos of you standing next to an Urus in Dubai
GPT writes a complete 30-50-page mini-course in half an hour
And Telegram bots reply to comments, qualify leads and make sales automatically
Real figures: a beginner with 1 account and 1 million views earns $4,000-12,000 a month
The average level with 3 accounts is $18,000-45,000 the top level with 5+ accounts and a team is up to $500,000
The funnel is simple: a video of a Urus and a Daytona on the wrist > "TG below" > a free channel with trading signals > a paid private club for $299β1,000 a month > consultations at $750β3,000 an hour > a $500 course that ChatGPT wrote in 30 minutes
Most people realise this is just LARPing and thatβs precisely why it works
32 hacks in Claude Code have saved me over 300 hours in half a year
> /init generates CLAUDE.md with the architecture and conventions in seconds
> Agent Teams coordinated agents that communicate with one another, with roles such as Lead, Researcher, Strategist and Reviewer
> Sub-agents on Haiku instead of Opus 10 times cheaper for read-only tasks
> Git Worktree 3-5 parallel sessions without a single conflict
> Plan Mode before every line of code Claude thinks, asks questions, and only then writes
The real trick isnβt in any single component itβs in combining them all together
/init + CLAUDE.md + skills + sub-agents + worktrees = a development team that delivers 100%
One subscription for $20, the output of a five-person team