@bcherny Can I add a console wishlist. I'd like to limit API key access to specific models. Eg I only want a key for haiku and can't elevate to sonnet/opus etc
hit 28% of my weekly claude code (+design now) limit in a single 3 hour session with opus 4.8. haven't hit a day/week limit for months. I ended up re-installing @opencode for another try. Whatever model "big pickle" is using is much better than last time I tried OC!
@michaelsnape@SB_Duggan As they say death and taxes are inevitable, they didn't say what form each takes. It's up to you to exercise good advice to extend both within the limitations set.
@michaelsnape@SB_Duggan Family trust, lower wage in sell year, with $500k super contribution each pre CGT. Do they have kids? Tax loss? Multi year earn out?
@levelsio I have been running this. Seperate log pull from claude. Download logs, check for error etc then pass to claude to process into a ticket.
Then another process for claude to code in the same way. Pick tickets, git checkout/branch, write code, commit etc.
Exactly. Do the dumb stuff on compute time.
Download logs and parse for errors. Download docs parse to MD. Clone git code first.
Do the dumb task in python/she'll etc, parse if there's a task needed. and then call Claude.
Experimented with an AI onboarding flow to personalise default user settings.
User enters email/phone. Backend extracts company domain and scrapes website for context. Claude API analyzes company type, industry, size. Returns personalized default settings.
Instant "ah ha" moment
The @AcquiredFM portfolio index:
Buy $1K of that company when the episode drops. From 44 public companies I reviewed from the episode air date to today
$44K invested → $144K today (+227%)
vs S&P 500: $76K (+74%)
Episodes 2015 Instagram (Meta) and 2022 NVIDIA did well here!
unpopular opinion: this PMF scale is why most startups fail
they think they're at 'product-market fit' when they're actually still at 'customer problem fit
Absolutely true for me about using the latest AI tooling.
"And in fact, many of my friends who have been programming for decades will say this is the most fun they’ve had in years. The productivity unlock is completely wild."
So far all the signs appear to be that people who are good at their field or craft are able to take advantage of AI the most.
Those who can write the precise specs for a coding agent to go off and do work, or craft the clearest prompts for various knowledge worker tasks, are getting the most leverage from AI today.
And in fact, many of my friends who have been programming for decades will say this is the most fun they’ve had in years. The productivity unlock is completely wild.
There's no evidence that this will ever change course. Which means that, like any other digital advancement that we've experienced, those that have the baseline (or better) skills to utilize the technology get the biggest gains.
This is why it’s way too early to make claims about what kind of jobs won’t be as in demand in the future.
My 1st day at Cloudflare, the now CTO gave me 1 big piece of advice:
Ship small wins early. Like ASAP.
The easiest way to build trust in a new org is to ship small wins early.
Ship small and fast (otherwise you're just the new PM working on the same darn feature for 6 months).
Hearing that AI models are collapsing because AI bros poisoned their own well by flooding the internet with loads of slop that’s being fed back into the training data with no way to sift it out. It fills me with such schafenfreude