GUIDES
Book Every Flight And Hotel With Claude 📚
The guy who built Claude Code just tweeted that Cowork on Opus 4.7 booked 8 flights and 5 hotels for him while he worked on something else. Here is the exact.
Why this works now (and didn't 3 months ago)
Cowork has been decent at travel booking for a while, but with Opus 4.7 it crossed the line. Boris Cherny, the creator of Claude Code, tweeted the receipts on May 14: he put his travel preferences inside his Cowork project, asked it to book a stack of trips, and Cowork "opened my browser, navigated a bunch of websites, and booked everything for me." Eight flights and five hotels in one shot, while he hacked on something else.
That is the part that actually changed. The model is now good enough at multi-step browser navigation that you can hand it the whole job. Three steps to do it yourself.
Add your travel preferences once
Open Claude Cowork. Create a new project called "Travel HQ." Open the project Instructions. Paste in everything you would normally check when you book yourself. Once it lives here, you never type it again.
Why this matters: the more specific these instructions, the less back-and-forth you have to do later. Spend 10 minutes on this and you save it on every trip from here forward.
Tell Cowork the trip
You do not need a long brief. Cowork already has your preferences. Just give it the basics, and add any specific date or constraint that overrides the defaults.
What it does next: Cowork compares Google Flights, the airline's direct site, and the hotel's loyalty rate against a third-party rate, and reports back with the best option plus the reasoning. You approve, it books.
Sit back
Cowork opens your browser, navigates Google Flights and the airline site, compares options, fills in the booking form, and pauses for your approval before charging the card. The whole thing runs in a Cowork browser window you can watch live or ignore. Boris described it as "I let Opus get to work" while he kept hacking in Claude Code.
Once Cowork confirms the booking, save this as a follow-up Skill. It produces the one-page brief I read on the plane.
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.