i can't believe people still let AI make changes with no undo button
the usual solution is to ask for permission before every action.
that gets annoying fast.
the better solution is to make more actions reversible.
code changes happen on an isolated branch. files are versioned. emails stay in drafts. deletes go to trash. database writes use transactions.
now the agent can move without asking me every thirty seconds.
the question isn't only "should the agent be allowed to do this?"
it's also "can i undo it if the agent is wrong?"
reversible actions can move quickly. permanent actions still stop and wait for me.
i give agents more freedom when undo gets easier.
NVIDIAs NEXT AI SERVER DOESNT PUT THE CPU IN CHARGE OF AI IT PUTS IT IN CHARGE OF EVERYTHING ELSE
STORAGEREVIEW JUST REVEALED ONE OF THE FIRST SERVERS BUILT AROUND NVIDIAs VERA CPU
the architecture offers a clear look at where ai infrastructure is heading
inside you get
vera cpu
connectx8 networking
bluefield dpus
high speed e1s ssds
rtx pro gpu support
this isn't just another server with more cores and higher clock speeds
its built with agentic ai in mind
the gpus handle generation
the network moves data
the dpus manage infrastructure
the cpu coordinates the entire operation
one of the most interesting details is the veras monolithic design which helps reduce latency between cores
that matters when thousands of ai tasks need to be coordinated across multiple accelerators
this represents a different approach from traditional AI servers designed mainly for model training
the new generation of AI hardware is increasingly focused on
ai agents
long running inference
multi-agent workflows
enterprise automation
massive ai orchestration
the desktop ai revolution brought powerful models to individual users
servers like this reveal what the next phase could look like inside the datacenter
the ai race is no longer only about building bigger models
its about creating hardware capable of coordinating thousands of AI agents simultaneously
ONE OF CLAUDE'S DEVELOPERS SHOWED HIS OBSIDIAN GRAPH AFTER 4 YEARS.
THIS IS WHY HE NEVER STARTS A PROJECT FROM ZERO.
the funniest part? he did not build a "second brain."
he built a system where every idea, meeting, project, mistake, and decision has somewhere to go before it disappears.
four years of it.
then he opened the graph.
at first, it looked like a few random notes.
then the links started spreading.
one cluster became ten.
ten became a map of everything he has worked on.
want to know what is inside?
one file, called CONTEXT.md.
it holds the current version of his work: what he is building, what changed this week, what is blocked, what he decided, and what claude needs to remember before touching anything.
claude reads it first every single session.
one file changed everything.
because now the ai does not open with "what are we working on?"
it already knows.
it sees the project.
it sees the last decision.
it sees the thing he tried three weeks ago that failed.
people spent years searching for the perfect prompt, the right temperature, the magic formula.
and the answer turned out to be not how you ask ai.
but what ai knows before you even open your mouth.
then the guy went deeper.
every project has its own folder.
not just notes.
context.
inputs.
decisions.
outputs.
when he works on something, claude sees that project only.
not the chaos of his entire life.
anything he does more than once becomes a skill.
research a company.
prepare a call.
turn a meeting into tasks.
write a project update.
one line and claude already knows the process.
and every morning, the system walks through the vault on its own.
finds what changed.
links what belongs together.
flags what went stale.
writes a short summary before he starts work.
he wakes up and the brain already worked.
and here is what got me the most:
it is all just text files inside obsidian.
no cloud.
no lock-in.
if a better model drops tomorrow, he points it at the same graph and the system keeps working.
he basically turned four years of work into an external memory claude can walk through.
most people use claude like google with manners.
this guy built a world for it to live inside.
the full graph is in the video.