GUIDES
AI Is Killing The Doomscroll đź“š
Social media usage peaked in 2022 and has been falling ever since. AI is doing the work it used to do. Here’s the data, the Jobs-to-Be-Done breakdown, and 5.
What the Data Shows
In October 2025, the Financial Times’ John Burn-Murdoch published a GWI analysis of more than 250,000 adults across 50+ countries. The headline: social media use peaked in 2022 and is now down nearly 10% in the developed world. The first sustained decline in a decade.
In March 2025, Appfigures reported ChatGPT had become the #1 most-downloaded app globally, beating Instagram and TikTok for the first time. AI apps haven’t just caught social media — on download charts, they’re leading.
The 5 Jobs Social Media Used to Do
A strategist named Matt Paige published a great breakdown on May 6, 2026, using the Jobs to Be Done framework — the idea that every product gets “hired” to do a specific job in your life. Read his full post: Social Media Scrolling Just Dropped for the First Time in a Decade. And I Know Why.
Paige’s argument: people scrolled on social media for 5 jobs. AI does 4 of them better.
Information
What’s happening, what’s new in my field, what should I know. Twitter and Instagram used to be the answer.
AI WinsA two-line prompt to Claude beats 30 minutes of scrolling for getting a real summary of what’s happening this week.
Entertainment
A laugh, a story, something interesting. The reels that made you snort.
AI Wins (mostly)AI doesn’t replace funny videos — but it absolutely replaces the long tail of “just show me something interesting.” Image generation, story generation, Claude as a sparring partner on whatever idea you’re curious about.
Distraction
Ten minutes of not-doing-the-thing-I-should-be-doing. The endless scroll as procrastination.
AI Wins (annoyingly)It turns out asking an AI a curious question burns the same dopamine cycle — except you usually come out with something useful at the end.
A Quick Hit of Dopamine
A small jolt of novelty or validation. Likes, surprising posts, the algorithm catching you off guard.
AI WinsGenerative AI is a novelty engine on demand. You don’t have to wait for the algorithm to surprise you — you ask, it surprises.
Feeling Connected
The sense of being plugged into people you don’t see in person. Friends, family, parasocial accounts you actually care about.
Social Still WinsThis is the one social media still owns — and the only reason most people haven’t deleted Instagram. AI companions are getting better, but watching your nephew take his first steps is still on the platform where his mom posted it.
5 Things to Build With Claude This Weekend
If you’re already part of the shift — spending more time in AI, less in feeds — the next move is the obvious one. Don’t just consume AI. Build with it. Here are five small projects you can finish in a weekend that turn the time you used to scroll into something compounding.
A Claude Project that pulls from your favorite Substacks, podcasts, and Twitter accounts and gives you one weekly digest. The thing Twitter used to be, but better — you control the sources, AI summarizes the signal, you skip the noise. Build time: ~2 hours.
Build a tiny app on Lovable or Claude Code that tracks the 3 habits you’re trying to build, sends you a Telegram nudge at the right time, and writes a weekly review of how you did. Replaces three different apps you don’t actually use. Build time: ~3 hours.
A Claude Skill that reads your Gmail, surfaces the 5 emails that matter today, drafts replies, and ignores the rest. Saves the average professional 30 minutes a day. Pays for itself in week one. Build time: ~1 hour.
A Claude Project where you paste any business idea and Claude runs market research, identifies competitors, sizes the opportunity, and tells you if it’s a real lane. Stop launching things on hunches. Build time: ~2 hours.
Reads your calendar, your priorities, and your last 7 days of work. Tells you what to focus on this week. Reads your Sunday-night brain. The closest thing to a chief of staff you can build in a single Saturday. Build time: ~2 hours.
The real unlock of AI agents is not full autonomy, but human-in-the-loop systems design. Structure your agent inputs with explicit constraints, negative prompts, and automated test checkpoints. When building tools, keep token consumption lean by caching system prompts and isolating tasks into specialized sub-agents.