Pinned Tweet
Lets grow together, me and the legend(you) 鉁笍
We're near at 600,
Than you for everything 鉂わ笍馃弳馃檹
We're near at 600,
Than you for everything 鉂わ笍馃弳馃檹
1
1
5
1,565
3 signs your code review has become a rubber stamp:
1. 800-line PRs get approved in under two minutes
2. The only comments are about naming
3. Bugs get caught in prod, never in review
Which one did your last merged PR hit?
1. 800-line PRs get approved in under two minutes
2. The only comments are about naming
3. Bugs get caught in prod, never in review
Which one did your last merged PR hit?
2
0
1
67
Researchers used AI to dig up years-old bugs in PHP. Then the AI-assisted attacks started.
The same tools that find the hole now help weaponize it, fast.
Patch windows used to be measured in weeks. Name one team that can still afford that.
The same tools that find the hole now help weaponize it, fast.
Patch windows used to be measured in weeks. Name one team that can still afford that.
0
0
1
56
Hot take: rollback plans for database migrations are mostly fiction.
Once new writes land in the new schema, "down" loses data.
The only real rollback is a forward fix you've already tested.
Name one team that ran a down migration in prod and was glad they did.
Once new writes land in the new schema, "down" loses data.
The only real rollback is a forward fix you've already tested.
Name one team that ran a down migration in prod and was glad they did.
0
0
1
67
Strong opinion: most "flaky" tests aren't flaky.
They're race conditions your production code also has, just triggered less often.
Retrying until green is shipping the bug with extra steps.
Name one flaky test that turned out to be truly random.
They're race conditions your production code also has, just triggered less often.
Retrying until green is shipping the bug with extra steps.
Name one flaky test that turned out to be truly random.
3
0
1
99
seridarivus 馃嚪馃嚧 retweeted
Seven shot dead by gunmen at church funeral in Jordan, state media say bbc.in/4idgdiC
551
700
2,406
1,340,255
3 signs your monitoring alerts stopped meaning anything:
1. Someone set up a filter that auto-archives them
2. The same alert fires every night and nobody fixes it
3. Real outages get reported by users first
Add the one your team is living with right now.
1. Someone set up a filter that auto-archives them
2. The same alert fires every night and nobody fixes it
3. Real outages get reported by users first
Add the one your team is living with right now.
0
0
2
79
The new Coding Agent Index 2.0 has Claude Code + Opus 5.5 on top at 61.7.
It also lists a setup at $0.39 per task. The leader costs $11.42.
That's ~29x the price. Name one real ticket where the top score is worth it.
It also lists a setup at $0.39 per task. The leader costs $11.42.
That's ~29x the price. Name one real ticket where the top score is worth it.
1
0
2
93
Unpopular opinion: the bottleneck in AI-assisted coding isn't writing code anymore. It's reviewing it.
A PR generated in minutes still takes a human the same hour to trust.
Name one team that actually changed its review process to keep up.
A PR generated in minutes still takes a human the same hour to trust.
Name one team that actually changed its review process to keep up.
1
0
5
114
A pharmacy chain's firewall was reportedly still accepting the default password "password".
The scariest vulnerabilities aren't zero-days. They're defaults nobody changed.
Name one security audit that actually checks default creds on every box.
The scariest vulnerabilities aren't zero-days. They're defaults nobody changed.
Name one security audit that actually checks default creds on every box.
0
0
1
80
3 signs your feature flags have turned into tech debt:
1. Nobody can say which flags are still read in prod
2. A flag has sat at 100% for a year and the old branch is still in the code
3. Turning one off needs a meeting
Got a 4th? Worst one wins.
1. Nobody can say which flags are still read in prod
2. A flag has sat at 100% for a year and the old branch is still in the code
3. Turning one off needs a meeting
Got a 4th? Worst one wins.
1
0
1
73
OpenAI's frontier models are coming to Google Cloud's Model Garden, right next to Gemini and Claude.
The moat isn't the model anymore. It's whichever cloud already holds your spend commitment.
Name one enterprise that picks the model before it picks the cloud.
The moat isn't the model anymore. It's whichever cloud already holds your spend commitment.
Name one enterprise that picks the model before it picks the cloud.
0
0
1
93
Hot take: an AI agent's quality isn't its success rate. It's what it does on the step where a tool call fails.
Retry, give up, or confidently make something up.
Most demos never show that step. Name one that does.
Retry, give up, or confidently make something up.
Most demos never show that step. Name one that does.
0
0
1
92
3 things that quietly inflate an LLM API bill:
1. Re-sending the full chat history every turn
2. No prompt caching on a system prompt that never changes
3. Client timeouts that retry requests the server already finished and billed
Which one bit you last? Name a 4th.
1. Re-sending the full chat history every turn
2. No prompt caching on a system prompt that never changes
3. Client timeouts that retry requests the server already finished and billed
Which one bit you last? Name a 4th.
2
0
3
208
Exponential backoff without jitter is just a synchronized retry storm on a delay.
I've watched it knock the same service over twice in one outage.
Show me a retry config running in prod that actually sets jitter.
I've watched it knock the same service over twice in one outage.
Show me a retry config running in prod that actually sets jitter.
2
0
3
346
Most p99 latency panels I've seen just average the p99 from each pod.
That number is basically fiction.
Has anyone's dashboard actually merged the histograms instead? Show me the query.
That number is basically fiction.
Has anyone's dashboard actually merged the histograms instead? Show me the query.
0
0
2
360
AI code review catches the unused import and approves the race condition.
I've yet to see one flag a real concurrency bug on its own.
Show me the screenshot where it did.
I've yet to see one flag a real concurrency bug on its own.
Show me the screenshot where it did.
0
0
4
514
Every database migration I've reviewed had a "rollback" section.
I've never seen one actually run in prod and work.
Tell me yours did. Table size and how long it took.
I've never seen one actually run in prod and work.
Tell me yours did. Table size and how long it took.
0
0
4
603
seridarivus 馃嚪馃嚧 retweeted
Lets grow together, me and the legend(you) 鉁笍
We're near at 600,
Than you for everything 鉂わ笍馃弳馃檹
We're near at 600,
Than you for everything 鉂わ笍馃弳馃檹
1
1
5
1,565