>code that nobody questions usually crashes in production
I don't understand what that means.
>code that nobody questions usually crashes in production
I don't understand what that means.
Probably "code that nobody critiques will fail in production". That's not always true I guess.
But usually the severity was more correlated with a lack of comments on the tests. Giant holes in test cases meant giant bugs. So don’t call a PR ready to land until you’ve gotten a few substantive critiques. Because the absence of evidence is not evidence of absence.
For the bugs I introduced (I am highly bug-averse) it was either zero comments on tests, or bugs introduced by the CR process - changes I was coerced into making that I felt were wrong. And I can’t say which sort of subconscious resistance was at work there. Self sabotage for making changes I don’t want to, or my sixth sense for robust code telling me the suggestion is an antipattern. Probably both.
What I learned from that last, which I confirmed in subsequent years, was that as a team you should only tolerate major pushback on a CR/PR at the beginning of the review process. Anyone who jumps in late with Needs Work demands, especially after a round of feedback changes has already landed, has lost their right to participate in the review. Because as a PR drags on, everyone gets tired of looking at it and has a less critical eye for spotting bugs that have been introduced by committee. It quickly becomes better odds that the original bugs the early reviewers did not catch are less dangerous than the ones that will be introduced by work-hardening the PR.
It’s the same mechanic that makes pushing code or a deployment after 4pm a bad idea. Confirmation bias is greatly amplified by a desire to be somewhere else.
This dynamic flourishes when the stakes and/or uncertainty are low enough.
High stakes and high uncertainty means everyone’s pushing their intuition and their reasoning as far as they can. They’re at their limit of what can be communicated efficiently. This results in an uneven distribution of communication bandwidth across the edges in the team network. Accountability induces leadership and competing views are ascendant and in decline.
I think it’s reasonable to wonder that, if the temperature never rises about room temperature, the team might not be fully challenging itself.
I suspect this only seems reasonable if you've never experienced a healthy work environment. I probably would have agreed with you when I was in my twenties, working at a startup with another bunch of twenty-year-old guys and a CEO who was in over his head. It wasn't unusual for the whole company to yell at each other in a meeting room. The stakes seemed high then, but they seem ridiculous in hindsight, as does my own behavior.
Thirty years later, the stakes are much higher, and the temperate is much lower. This is precisely because we can't afford this behavior, and we can't afford to deal with people who can't control themselves and behave professionally in high-pressure situations.
Doesn't the article refute exactly this point of view? In "The hidden cost of “nice” teams" section:
"Those teams weren’t actually harmonious—they were conflict-avoidant. The disagreements still existed; they just went underground."
A toxic consensus culture usually does not even allow for a serious, critical discussion, but dissenters are just ignored or verbally patted on the head.
In my experience, this is not true because, in high-trust teams, there is "harmonious conflict." People offer criticism without getting heated.
Getting heated often results from a strong opinion combined with a lack of faith that other people are genuinely considering your opinion. People who get heated feel they have to be forceful to convince others to listen. Knowing your opinion will be hard and carefully considered, you don't need to get heated.
We have a project to develop a new engine, its teklaover budget, very late, and requires a ton more work and excuses are running thin.
My team is running barebones with 60-70 hour weeks trying to get the ratings done, the test team is waiting on us for the ratings to get their work done and THEIR manager is pissed because they're billing for time on their books waiting for us to get our shit together. There is no more room to delay the tests because another team also needs their project tested and has waited for the test team to free up.
Meanwhile, my bosses boss is pressuring my boss to submit what ratings data we have to get approved to start testing. The test teams boss is begging us to submit the data too so he doesn't have to bill time for being idle on standby.
Tired of fighting, we submit the ratings data and tests starts. And boom 2000 hours into testing the gearbox fails and blows out the back of the engine at 4am. The test facility is now inoperative until cleanup finishes and several hundred hours are now needing to be spent figuring out why a $3 million engine just blew itself up.
The cleanup team is now being paid a fuckton of overtime to clean up the mess, and an army of people are being summoned to provision a new test engine.
The test team now needs to spend extra time writing reports, the ratings team now needs to go overtime to detail why they fucked up and why we suck so much at our jobs, on top of their normal work, and the customer is super fucking pissed.
So yeah, there is no anger if stuff doesn't matter. When stuff does matter, anger happens. It's very easy to sit back and claim "no heated discussions" when it doesn't involve millions of dollars and hundreds of hours of other peoples time when its your team that caused the fuck up
This is toxic BS and mismanagement. It's people not accounting for and communicating risk properly. None of that anger and finger pointing makes this work faster or better. Everything here could happen exactly as it happened, but without anger, if people communicated risks properly and took the failure (which will happen) as part of the process, people would be less stressed about it, and maybe they'd make fewer mistakes too.
Maybe it was correct to progress without better data, or maybe the cost of doing that was too high. Either way, someone should've redone the schedule to match reality, so blocked teams don't get anxious. Then someone should've done the risk/benefit analysis and signed off on the "proceed with testing anyway, we are aware of and accept the risk" path, so nobody talks about fucking up, but concentrates on doing the best they can instead.
I've worked in large cross-company teams on extremely complex products with millions of LOC and thirty-year-old obsolete code handling millions of US$s of bank transactions. I've been in situations where people made mistakes, systems went down, and transactions were lost. I've been in incredibly stressful situations.
Do you know what would have made none of these situations better? Anger.
It sounds to me like lots of discussions should have been had before getting to this point? So sure at the point where you ended up, things get heated because everyone is overworked and frustrated and being pressured from all sides, but that's not really a function of a complex project, but more of a poorly managed/executed one?
The hubris in this sentence is chef's kiss. You have no idea.
> When stuff does matter, anger happens
No, anger is completely unnecessary. It's a poor way to act and communicate, often counter-productive and terrible for morale.
But with less jest - I think we have a very good discussion culture in the team at work. No one on the team is scared to disagree, or to point out that this doesn't seem to go towards a positive direction. It's just not a heated discussion.
At times during these planning sessions, we just sit in silent thought and maybe doodling something in Excalidraw or on paper for a few minutes until someone is like "I need more input on X", or "So I have a rough plan I think?" or "I think there is an issue with that idea", even if it's not clear why - if 2 or 3 people are vaguely not fine with something, it's probably a bad idea. And we've also started to learn each others tells if we're not happy.
Shit code or architecture that other devs didn't call out.