Joined February 2010
- Tweets 15,388
- Following 55
- Followers 13,446
- Likes 4,127
1,339 Photos and videos
May 1
Very happy with our move from Heroku PG to Planetscale but oof that egress hits hard haha
7
29
8,422
Apr 15
Cloudflare AI gateway is almost perfect, but I wish there was a way to put a provider in "time-out" when it errors...if the provider errors, skip it for the next 5 minutes and use the fallback, instead of trying the (failing) provider every time
1
5
1,352
Apr 13
Would love to see any experience reports with running Rails workloads (either web server or jobs) on Cloudflare containers
1
1
13
3,027
Apr 11
I honestly don't know how anyone makes a performant and cost effective looping agent product
Hoping it's a skill issue, plz send help
16
19
6,176
Apr 2
btw as part of moving from Heroku PG to Planetscale we now have a small cluster with read-replicas...it was so awesome that Rails supports this
connects_to database: {writing: :primary, reading: :primary_replica}
autoswitch on GET vs POST
2 lines of config, just works
7
1
71
4,828
Apr 2
You are building a Rails app that runs agent flows / pipelines so you end up making lots of blocking inference API calls
Is the future:
- Running a bunch of async jobs
- Offload I/O heavy work and get callbacks
- Something else
Consider that tool calls need to hit Rails app
14
43
6,208
Apr 2
I can see specific use-cases for both:
- Running a deep research task or page crawl -> send to 3rd party, get webhook back when its done
- Agent loops with business rules -> run locally as jobs
Or do you orchestrate fly sprites or CF workers? Or jobs but fanned out with SQS?
4
1,019
matt swanson π retweeted
Apr 1
You've heard of @Cloudflare Durable Objects... but what are they?
What do they enable you to build?
What are the platform guarantees?
What new patterns do they open up?
Good news! I just released a completely free series on Durable Objects!
databaseschool.com/series/duβ¦
38
99
912
141,111
Apr 1
Hard agree...I hate having to tell my agent "bro there 100k records in production this is going to explode"
Apr 1
People and agents would be better at writing code if you could easily check what value a variable usually has in production:
2
18
4,222
Apr 1
cant believe they suckered @jessethanley into paying them to talk about them...rookie move
1
8
984
Mar 31
I honestly think I'm going to end up just keeping compute on Heroku. It's overpriced but literally does not matter to save a few hundred dollars a month for our business.
Heroku wrapped in Cloudflare, db on Planetscale, all errors/perf/logs to Sentry. Easy life.
Mar 31
You really don't appreciate Heroku until you look at other platforms. Just kicking the tires on Fargate ECS and realizing Heroku is also handling:
- Routing
- Scheduler/cron
- Console / SSH / one-off runs
- "Building a container image"
- Env var management
- Custom ACM certs
12
1
63
12,329
Mar 31
We are getting applications for interns now. Last year, we hired one that showed "proof of work" -- a rambling video of them going through our site, talking out loud about their questions, and showing how they researched things with AI / google when they didn't understand
Mar 31
Being human in your job applications more important than ever x.com/NanoClaw_AI/status/203β¦
1
11
2,039
Mar 31
You really don't appreciate Heroku until you look at other platforms. Just kicking the tires on Fargate ECS and realizing Heroku is also handling:
- Routing
- Scheduler/cron
- Console / SSH / one-off runs
- "Building a container image"
- Env var management
- Custom ACM certs
18
68
20,369
Mar 25
"To recap what you accomplished:
- Diagnosed the original cached 404 CSS issue
- Removed CloudFront from the request chain
- Simplified from User β Cloudflare β CloudFront β Heroku to User β Cloudflare β Heroku
- Deleted the CloudFront distribution and WAF ACL
- Confirmed old and new asset hashes both work through the new setup
Youβre done!"
Claude is so helpful on this stuff, even just telling you which screens to go to in the various consoles
4
9
8,168
Mar 20
Just want to keep shouting out @Judoscale -- we migrated all our sidekiq queues to SLA and auto scale off latency. Been running for 3 months so smoothly
1
1
14
2,437
Mar 19
"You can just vibe code all your GTM stuff with openclaw, just cancel all your SaaS and hire an AI consultant"
4
2,803