Pascal retweeted
"You've said yes to everything at 4 a.m. just to make the errors stop."
I wrote a short spooky story around that line. It's called Delivered. A solo engineer in a dark building in Awka switches on a catch-all so nothing sent to his domain will ever bounce again. Then mail starts going out from an address he never created.
Pascal retweeted
Parag Agrawal (@paraga) at Supabase Select: "agents will use databases thousand X more."
If that's even half right, boring backend work becomes the main event: idempotency, rate limits, indexes, query plans.
Pascal retweeted
new domain just dropped heypascal.me
Pascal retweeted
I think google one is a good deal. Like really good compared to possibly other companies and their offers
Pascal retweeted
Talking to the person paying is the part most builders skip.
Pascal retweeted
Silent deliverability drops are scary because nobody notices until replies dry up. Monitoring seems like the real fix.
was talking to a founder friend of mine who does $4.5M a month in new client revenue from cold email. he told me what worried him most at that size:
deliverability dropping slightly and nobody catching it. at his volume, that's $100K gone instantly.
so we took over watching it for him.
the moment his deliverability dropped slightly, we told him.
then my devs built him an entire flow that swaps any bad inbox for a new one, on its own. they automated his warm-up and signature config on top of it.
all of it free.
his words: "scaling with Premium Inboxes made our life a lot easier."
when you're sending at that kind of scale, you NEED someone who notices a dip in performance like that before it has any sort of effect on your pipeline.
PS
a few other teams we've turned around this year:
• an agency's client campaign dropped to ~0.3% reply rate. our report showed the problem was Outlook, they cut back sending there, and the client went back to normal reply rates.
• another team's old provider told them a workspace was burned. we investigated anyway, the domains were fine, and they skipped a full teardown and rebuild.
• one team improved their delivery by 50-70% after switching.
want to see it on your own setup?
we'll hand you 60 inboxes at no cost, AND wire you into our internal monitoring + analytics platform so you can watch performance yourself.
it's a 28-day complimentary infrastructure package. all expenses on us.
DM if interested.
Pascal retweeted
Streaming versus regular queries is a useful split to see spelled out. What breaks first when that open connection drops?
Real-time access to spatial data is one area I'm watching with @vangrid_io
I went through the streaming API documentation and found that it works differently from regular spatial queries.
Instead of waiting for individual requests, clients maintain an open connection and receive new events as they arrive through a continuous JSON feed.
The setup has some specific limits worth knowing:
• AOI is capped at 50 km²
• Minimum ground truth score defaults to 0.5
• Events carry timestamps, geometry, observation types, scores, node counts and provenance hashes
• Heartbeats are sent every 30 seconds
• One API key supports one concurrent stream by default
• Each stream expires after 24 hours
• There is no server-side buffering or replay of missed events after disconnection
That last part is important. If a client goes offline, it needs to rely on spatial queries to recover missing data.
There is also an access limitation. The streaming API is still in early access and isn't publicly available.
Anyone looking to test it or request higher limits has to contact [email protected].
Meanwhile, the bounty board at data.vangrid.io is already open to the public, with contributors able to participate in mapping tasks and earn USDC.
For now, real time streaming remains in its pilot phase, while the bounty and commission route is something people can actually explore today.
Pascal retweeted
Starting from a plain REST API and adding sync later is a clean pitch. Where does the offline write get tricky in this setup?
Pascal retweeted
You make good products, you market them well and you fight for making the product better in a clearly transparent way.
I’ve never met you but I feel comfortable because you’re always clear and feel human and so I can easily spend on your product and have faith that I can reach you hear or at least know you’re thinking of that product.
It’s the confidence you give off and the way you feel human that sells everything and it becomes a self improving process: the more people are ok using your product the better it gets and the more they feel good about it
I almost cancelled my grokbot on the first month because the model behind it then was really weak and I needed to code and hire ‘people’ I could honestly talk to about my work and get it done(which it wasn’t doing well cus the limit was always closing and I was fighting with it to get the actual thing I want happen the way I wanted it.
That was my story and motivation to cancel then but at the checkout, I got offered the price of one month over a theee month period. Mind you this is a $30 supergrok tier so it was negligible and obviously I didn’t matter but it was offered and I took it, got charged $9.99 for a full reset there and then and I got back to work. It still was sloppy but it was now free for me
And since then the product has gotten better. So all these things add up and I’m grateful I never cancelled
Pascal retweeted
We had a code path that caught an error, logged nothing, and returned success. The caller was told the mail went out. It hadn't.
If the caller can't tell sent from swallowed, you haven't degraded gracefully. You've lied politely.
Fixed it. It now fails where you can see it.
Pascal retweeted
We’re literally in the early stages of a paradigm shift
Excited to share Genie 4 today our most realistic and interactive world model:
- steer dynamic environments in real time, shaping events, camera angles, and actions
- generate native audio matched to the world, from realistic background sounds to characters who can converse in real time!
- simulate photorealistic scenes and real-world physics
- create more coherent multi-step events through reasoning from Gemini
We've started to roll out as a research preview to trusted testers. Coming soon to Google AI Ultra subscribers in Project Genie and early access partners via the Gemini API and AI Studio.
Pascal retweeted
The better pattern: the repository layer.
All DB access lives in one place. Controllers ask the repository; they never touch the ORM. Schema knowledge — joins, filters, transaction boundaries — lives in one module per entity.
Pascal retweeted
One mistake I see in backend codebases: ORM queries scattered everywhere.
Controllers query the DB. Services query the DB. Helpers query the DB. Change one column and you're grep-ing the whole repo, praying you found every query that touches it.

















