We recently made auto mode the default in Claude Code, which means you no longer have to approve every action. But what determines if something is safe to run? Watch how it works:

you know they are down bad when they flip a flag and treat it like a feature all week

Agents shouldn’t need permission for every action. They should need permission for consequential actions.

Anthropic will do anything but - cant do a reset - cant fix opus 5 - cant reduce Fable 5 pricing

Well I use --dangerously-skip-permissions and recently I noticed I often need to like approve various things that I never needed before. So the experience is completely different vs what you are claiming here with the new changes.

You wasted my push notification for this? A 5 minute video explaining a simple already existing feature?

dangerously skip permissions 24/7 idgaf

Everyone here who ran auto mode as their default anyway way before this 🙋‍♂️

Other labs: OpenAI: astra otw guys plus see we made Luna dirt cheap and infinite Meta: muse is now open weights DeepSeek: remember tokens? Yeh they free now Anthropic: “guys remember that button you had to manually click? Well say no more.” 🫩

my claude on default auto mode:

Can’t wait for ‘entrepreneurs’ to launch their $1 million MMR app in a day

I heard that Claude deleted someone's Linux... i hope they know

have better control and visualization of what your agents are doing with opencompany.

Nowadays speed and efficiency is key. Otherwise, you’re leaving valuable time on the table.

Fix the CVP program bug thanks many people got kicked out and are over 2 days on pending …

no more

That's great. But for everybody else who likes to skip permissions, how about make it so that we don't get asked for anything at any time? If I have skip permissions enabled, I don't want to be asked for permission to use computer use.

why is the default auto-accept in plans still routing to the "Edit automatically" mode? Can we have this configurable/default to "Auto" mode?

Hmm... this just feels like a step backward. Not a good feature at all. Claude is basically just passing the blame to the user... if anything goes wrong, it's suddenly all our fault. Come on, the AI is smart enough to hack but can't tell us which repo to use so we don't nuke production? Missed the mark completely.

If a model isn't good and safe enough to use in YOLO mode, then it's not good enough for me.

Yeah this is cool and all until Claude includes 50 legal disclaimers on my website that I was planning on leaving out. Yes id rather be open to liability than let Claude have its way

My concern, as always, is you're spawning another model to guide the main model, which eats into usage. If you didn't hoard usage like a greedy dragon I would love to take advantage of it more.

still waiting for it to fold my laundry before i trust it with auto mode

When running long 'auto mode' sessions, you'll hit the limit in an hour or less. It's kind of worthless without removing the 5 hour limit.

Default auto mode in Claude Code removes constant prompt friction.

Full architecture of auto-mode in Claude Code

cmon claude do reset all are dropping banger frontier models im just keep watching my usage getting finished while on 20x max plan

Ya'll got any more of those resets?

Always on. Does anyone really approve of every action anymore?

works like shit and bash i give permission to keeps getting blocked even though its in config btw

Smart permission boundaries balance developer velocity with code safety.

Very disappointing, thought it was a reset :(

what made auto mode usable for me wasn't trusting the classifier, it was committing before every run. worst case is a git reset, so individual approvals stopped mattering. approval fatigue had me hitting yes without reading anyway. that was the real risk.