HE BUILT A SECOND BRAIN WITH CLAUDE AND OBSIDIAN, NOW EVERY NEW REQUEST TO CLAUDE STARTS NOT FROM ZERO BUT FROM EVERYTHING IT KNOWS ABOUT HIM.
at 0:03 the camera holds on the graph view in obsidian, hundreds of notes connected like a knowledge map, projects, ideas, decisions, context, everything linked together.
and claude has access to all of it.
most people open a new chat and start explaining everything from scratch, who they are, what they are working on, which decisions they already made, what did not work last time.
every time a new intern who needs to be told everything again.
he built it differently.
obsidian stores everything, meetings, business documents, previous decisions, important context, and claude has access to this entire structure.
now he opens claude and it already knows what he is working on, which decisions he made, what matters and what has already been tried and dropped.
this is the difference between AI that answers questions and AI that understands where you are in your project.
HE SWITCHES CLAUDE MODELS MID-TASK AND NEVER PAYS OPUS PRICE FOR SONNET WORK
00:14 he nails the core idea
efficient Claude usage needs two things - the right model and the right moment to switch.
the default model is Sonnet. fast, cheap, handles every research question, every small bug, every draft version - not Opus, not by default.
the moment is one slash command. no new chat, no lost context, no re-explaining the project from scratch.
describe the task → Sonnet handles the routine → type the command → Fable takes over only when the decision is actually expensive.
all of this without burning premium tokens on a five-minute fix, without restarting context, without paying frontier price for work a cheaper model already handles well.
most people leave the strongest model on for everything and pay full price for every reply, regardless of whether the task needed it.
he chose a different rule. harder to remember at first, but he controls exactly where the budget goes.
the hard part here isn't typing the command. the hard part is knowing which five minutes actually deserve the expensive model.
HE SWITCHES CLAUDE MODELS MID-TASK AND NEVER PAYS OPUS PRICE FOR SONNET WORK
00:14 he nails the core idea
efficient Claude usage needs two things - the right model and the right moment to switch.
the default model is Sonnet. fast, cheap, handles every research question, every small bug, every draft version - not Opus, not by default.
the moment is one slash command. no new chat, no lost context, no re-explaining the project from scratch.
describe the task → Sonnet handles the routine → type the command → Fable takes over only when the decision is actually expensive.
all of this without burning premium tokens on a five-minute fix, without restarting context, without paying frontier price for work a cheaper model already handles well.
most people leave the strongest model on for everything and pay full price for every reply, regardless of whether the task needed it.
he chose a different rule. harder to remember at first, but he controls exactly where the budget goes.
the hard part here isn't typing the command. the hard part is knowing which five minutes actually deserve the expensive model.
YOUR CLAUDE COWORK GIVES GENERIC AI SLOP BECAUSE YOU SKIPPED ONE FILE
00:23 the moment he shows the fix that saves everything
it's not the prompt that's broken - it's the empty context
without this one file, Claude is guessing who you are, how your team works, and what actually matters to you - and guessing badly, that's exactly where the "generic AI slop" comes from
add your name, your role, your working rules to it - and suddenly the reports sound like you, not like a template pulled off the internet
the detail from the video almost nobody does - save the process as a skill so you stop re-explaining the same thing to Claude every single time
99% of people close Cowork after the first generic report instead of fixing one file
don't be the 99% - follow for more fixes like this
ANTHROPIC LEAKED THEIR INTERNAL AI AGENT PLAYBOOK AND IT CHANGES EVERYTHING
00:41 the moment he drops the rule that matters most - one agent does the work, a second one checks it
because a smarter model - a better system
real systems don't need the smartest worker, they need a controller - the one deciding what to do and whether it actually worked
three rules from:
1. if it's not written into context, it doesn't exist for the model
2. give the agent a role and a goal, not a random task
3. never trust it blindly - a second agent always checks the first
this is exactly how we run Claude as the manager, not the worker - full architecture breakdown coming in the next post, follow so you don't miss it