Software builder @verze_ai, where I break things in production so nobody else has to.
Zurich 🇨🇭
Joined May 2008
- Tweets3.3K
- Following841
- Followers2.2K
- Likes14.4K
Silvan Hagen ⚡️ retweeted
None of our repos install npm packages younger than three days anymore.
One line in .npmrc: min-release-age=3 (npm 11.10+)
Matters more when an agent updates every repo on Friday nights.
Catch: it only applies when resolving, not in npm ci. Older npm ignores it silently.
Silvan Hagen ⚡️ retweeted
We use a small app to manage maintenance chores and minor fixes. It works, updates come through, but it doesn't always commit the lockfiles.
So on Friday, I had to patch the PRs by hand. After some cursing and tinkering, it commits lockfiles automatically while I sleep.
What’s the last small but infuriating fix you gave your clanker?
Silvan Hagen ⚡️ retweeted
Design is hard 😅
We asked ourselves @verze_ai: who’s currently using our app, and who should be able to use it?
Most of our users understand performance marketing deeply. Not someone who boosted a Facebook post twice.
Last week, we started working on making our app easier to use, and the first improvements are already out.
We’ve still got a long journey ahead, and I doubt we ever will or need to pass the mom test. But there are already fewer places to get lost.
Silvan Hagen ⚡️ retweeted
Big up @contra this is dope work (and a pretty site ;)
AI is making creativity more valuable than ever before and..
we have 3 years of data to prove it.
Introducing the Creative Intelligence 2026 report from @contralabs.
Full report: c87.to/creative-intelligence…
None of our repos install npm packages younger than three days anymore.
One line in .npmrc: min-release-age=3 (npm 11.10+)
Matters more when an agent updates every repo on Friday nights.
Catch: it only applies when resolving, not in npm ci. Older npm ignores it silently.
Wasn't happy with the reframe video apis out there.
I need subject tracking, but as little outpaint/filling as possible.
Agents will be cooking over night, will see if it works.
Silvan Hagen ⚡️ retweeted
"Cannot find" is my favorite kind of bug report.
We moved from a user facing Filament panel to a custom built panel using React and Inertia. We reached feature parity quickly, but I knew there will be some features missing.
And a few days after the release user reports started coming in: "Group by link missing." "Cannot find Create with AI." "Where did the UTM preview go?"
Some of the features just moved into a different place. Out of 12 reports in 6 days, only 3 where gone from the new UI.
We quickly brought them back or moved them to more obvious places. All in all I was just happy to see our users using the app, even the features that are not used daily.
What is everyone using for product analytics these days? Hand-rolled? PostHog? Other?
We use a small app to manage maintenance chores and minor fixes. It works, updates come through, but it doesn't always commit the lockfiles.
So on Friday, I had to patch the PRs by hand. After some cursing and tinkering, it commits lockfiles automatically while I sleep.
What’s the last small but infuriating fix you gave your clanker?
Silvan Hagen ⚡️ retweeted
PHPStan 2.3.0 is out with massive leap in performance, generics improvements, detecting unused variables, improved closure type inference, inferring static variable types from assignments, and engine rewrite!
Design is hard 😅
We asked ourselves @verze_ai: who’s currently using our app, and who should be able to use it?
Most of our users understand performance marketing deeply. Not someone who boosted a Facebook post twice.
Last week, we started working on making our app easier to use, and the first improvements are already out.
We’ve still got a long journey ahead, and I doubt we ever will or need to pass the mom test. But there are already fewer places to get lost.
We now do brand checks for the social ads copy before they go out.
This was such a cool feature to implement using the Laravel AI sdk and lots of trial and error to get it right.
Who else is absolutely loving this time to be software builder?
Silvan Hagen ⚡️ retweeted
Inbox zero for reviewing ads 😅
Before: Our users filtered a table of all their ad entries by “Ready for Approval.”
Now: Quickly review each ad in its own UI.
Your review queue used to be a table. Now it’s the ad. A approves. R rejects. S skips.
That’s the whole flow now.
Tonight I’ll be talking about an internal project that saves us countless hours every week.
If you are in Zurich, come and join us ⬇️
Reminder: our next Laravel Switzerland meetup is on 30 September in Zürich, back at the Liip Arena. Two talks, drinks and snacks.
→ luma.com/gufser9e
#Laravel #PHP #WebDev #Switzerland
"Cannot find" is my favorite kind of bug report.
We moved from a user facing Filament panel to a custom built panel using React and Inertia. We reached feature parity quickly, but I knew there will be some features missing.
And a few days after the release user reports started coming in: "Group by link missing." "Cannot find Create with AI." "Where did the UTM preview go?"
Some of the features just moved into a different place. Out of 12 reports in 6 days, only 3 where gone from the new UI.
We quickly brought them back or moved them to more obvious places. All in all I was just happy to see our users using the app, even the features that are not used daily.
What is everyone using for product analytics these days? Hand-rolled? PostHog? Other?
Silvan Hagen ⚡️ retweeted
Coming soon from VERZE: a browser extension that scores any page for ad readiness before you spend on it.
Same score we use to help marketing teams improve their ads and ad spend.
All in your browser, nothing sent to us unless you send a report.
Free, no account needed.
ALT Score any page before you spend on it. The VERZE browser extension side panel scoring verze.ai 100 out of 100, with factor details, crawler file checks and a story ad preview.
Silvan Hagen ⚡️ retweeted
We are hosting our first webinar 🎉
The AI Operating Model: how to scale marketing without scaling cost.
Thursday 29 October, 11:00 CET. Live, online, in English, with Q&A.
For CMOs and marketing leads planning 2027 budgets
Save your seat 🪑 verze.ai/events/ai-operating…
ALT Our first webinar is here. The AI Operating Model: how to scale marketing without scaling cost. Thursday 29 October, 11:00 CET. Live, online and free.
To find out what our own product feels like when I'm the user, I started running ads for a small side project.
Seven days in: CHF 24.99. 13,490 impressions, 1,073 clicks, CHF 0.02 a click.
The placement mix is interesting: Reels and Stories carry half the impressions, Facebook Feed had the best click rate, Facebook Search sent almost nobody.
I am off in the mountains for a couple days now with my wife, but happy the ads keep running and I don't have to touch them.
ALT Using VERZE to optimize the ads for a little side project: the seven-day performance chart with 13,490 impressions, 1,073 clicks and CHF 0.02 per click, and the per-entry Performance dialog with the placement mix.
In @verze_ai, we render videos in the browser using Remotion. The tab has to stay open and best in front of our users.
We tried to explain that in a tooltip and watched people switch away anyway.
My colleague came up with a little cat runner game to pay homage to the Chrome no network dinosaur.
From the department of doing more silly things: play with us while you wait.
Translating a video is more difficult than translating the words.
Same script, native speakers: German takes about 16 seconds, while English takes 13. French lands next to English even with more letters because it's spoken faster and half of them are silent.
Dubbing fixed-length videos is a science. Either you cut meaning or speed the voice up until it sounds hunted.
In @verze_ai studio, we are experimenting with this: the picture runs as long as the audio needs. The German cut is longer while we still try to stay within the boundaries of social media ads (15s and 30s).
Any tips on how you dub videos are greatly appreciated.
ALT Same script, German runs longer. Bars showing German at 16.1 seconds, English at 13.0 seconds, French at 13.4 seconds, read aloud by native speakers.