Sean McCann

@onnbase Consider venturing beyond soma

Visiting New York always reminds me how SF lacks savoir faire. New Yorkers just know how to live well and make the most of life.

@BlakeMycoskie When did you have kids?

I started TOMS at 29. Sold it at 43. Fell into the deepest depression of my life at 44. And I'm 49 now. Here are the lessons I wish I'd learned:

@CAcannabisdept Why are you using tax payer dollars to promote this post?

If it’s Real CA Cannabis, it’s on the map. Check here to see if your shop checks out.

@hamburger Can you multiple that by 12 or 365?

So Poke had a billionaire user who agreed to pay $136k/mo for the service during the haggle-boarding 🤣

California Profit: $800 Tax: 100%

@zephyr_z9 90-95% of tokens are cached. That’s where the margin is.

"One data point stands out. A single Claude Max subscriber generated over $5,600 in equivalent API costs in one billing cycle. On a $100 plan." PSA: API has a 90%-95% margin. So actual costs are somewhere around $300-$600

@based_coded Surprised you noticed a difference from the sunnyvale tp

Perhaps I’m naive, but before moving to San Francisco I had never met such a large cohort of people that just straight lie all the time and are comfortable with it. I think this is a big city thing and not SF specifically.

@astuyve What percentage of your token usage is cached

I've been watching the dashboards of my token spend across all providers and whew, let me tell you This website is not ready for a (rapidly arriving) world where they pay market rates for inference

@RozenMD That bill though

you could drop 400 million requests per second on that mf and it'd Just Work

I suddenly feel better about spending 100+ hours on SOC 2 over the past year

I'm noticing in my Claude Code usage stats (from ccusage) that 95-99% of tokens are "cache reads." If true, then we're really not using as many tokens as we think. Instead of $573 it would be more like $5 of usage. Maybe ccusage is broken because I doubt I used 691M tokens yesterday, but still.

@jessemontano_ They’ll blame Darnell Nurse for everything

"Again, I really don't give a crap if the goalie is hurt. That's on their D-man, not our guy" - Jared Bednar on MacKinnon ejection

@Stammy We need self hosted Mint for 2026

don't make me vibecode my own web analytics tool.. i'll do it i might spend more on tokens and hosting, but i'll do it 😄

@jenslystad @dakshgup It sounds like 50 reviews so could be more than 1 review per PR

@sea @dakshgup I assume its 50 PRs, not commits

Today we’re releasing Greptile Agent v4, our best code review agent yet. It’s catches more tricky bugs than v3, while also having far fewer false positives. We're also updating our pricing to help us support power users.

@dakshgup Wait, 90% of users are shipping less than 50 commits a month?

Lastly, we’re modifying our pricing to a base + usage model, similar to many popular AI coding tools. The new pricing is $30/developer/month, which includes 50 reviews, after which reviews cost $1 each. Less than 10% of active users will exceed the included usage. Existing users will remain on the existing pricing until the next billing month. Better coding agents have led to a drastic increase in the number of commits reviewed by Greptile among power users. This new pricing model allows us to focus solely on providing an uncompromising AI code review experience.

@vfnuness @anuraggoel Was the llm call cached when it was 25x faster?

One of the points I validated was a call to OpenAI. Today I do some processing, and in the render process it takes close to 50 seconds to respond, whereas in Railway the same request responds in less than 2 seconds, even during their validation period. Same code and same request in both.

I'm migrating the entire @parsaa_ai infrastructure currently on @render to @Railway. I ran some tests, and Railway delivered better performance and speed, not to mention the cost: with them, I only pay for what I use, whereas with Render, everything was charged separately, whether I used it or not, it was always the same price. Hey @Railway, do you have a discount coupon? 😅

@PenguinsJesus @spittinchiclets @barstoolsports @espn @yanismydj get on this

I’m begging you. Bring back Pro Beach Hockey. Somebody with a vision and money needs to give the people what they want.

@DavidTaggart @QuinnyPig Disclaimer, I worked there in the early days and personally built some of it.

Checkr is not a fraud. Criminal data is scattered across 3k+ counties, some counties are digital, some are not. It’s not feasible to check all 3k+ counties for a name/ssn so you have to narrow it down to the counties they’ve lived/worked/etc. There are strict regulations about what can we reported on a background check as well. Checkr is best in class in a messy data world.

I, apparently a naive idiot, had assumed that this was what “we background check our drivers” always meant.

I’m far too old to be this online