The first time I watched Return of the Jedi, I wondered how we were supposed to believe that a gelatinous slimy mound of evil who could barely move around and wholly depended on sycophants to cater to his every whim could have possibly amassed such a fortune that he would inspire fear and be running things.
I get it now.
there's still a massive gap between people who use chatgpt and people who actually know how to make agents do work
that's an opportunity
if you know how to build with agents, go meet people. learn how they work, what sucks, what they need
building is getting ridiculously cheap, access to real problems isn't
just go
Claude Code has several permissions modes you should be aware of:
Planning Mode: mostly prevents you from even planning.
Manual: you approve everything not explicitly allowed until carpel tunnel sets in.
Accept Edits: the same as manual mode but called something else.
Auto: you spend 25% of your time convincing a really really stupid “auto classifier” model to let your agent do what you JUST asked it to do, 25% of your time trying to figure out which part of the job the classifier wouldn’t let the agent do, and the last 50% of the time watching your agent burn mountains of tokens trying to circumvent the auto classifier.
Dangerously Skip Permissions: the only way to get work done with Anthropic models. Also dangerous. Have you tried auto classifier mode?
Which one is your favorite?
To Lane Lambert, Thursday's 5-1 loss to Colorado was "mind boggling," calling out an effort that was "unacceptable." #SeaKraken
» https://t.co/2VLyCfNYz3
@EmeraldCityHky Given how we see other teams stretching out their cup-competitive windows in the last ~10 years, is there anything that the Kraken could learn from those teams before our window opens? Anything we should change about player or contract management?
@EmeraldCityHky Hey do you guys have a plan to move to bluesky? I think you’re the main Kraken creators that haven’t moved yet (and I would *love* to dump this account after 20 years!).