He asked "can you do it faster" and I agreed, thinking I'll make a throwaway version first and fix it later. Needless to say the project was a disaster, rapidly became unmaintainable.
That's how I learned my job isn't to do what the client asks, it's to make sure their project succeeds even if it means making them (temporarily) unhappy.
And I learned that doing it my way will get me fired because the manager has asked to do faster.
The way I have learned to get around this is by making the manager publicly document the request to go faster. If they don't document, I don't see or act on it. Once they document it, I happily go faster and let it all crash and burn. Then when they inevitably try to blame me, I point at the public documentation of the request to go faster and let the blame fall on the person responsible.
if time is the only actual concern for the project's success, a good approach is to explicitly re-scope the feature list and start asking managers things like:
"do we really need feature X to release? can feature Y wait until after beta? did the request for feature Z come from a user or a stakeholder?"
document all that too of course ;) but then at least there is a chance for safety _and_ success
Yes the project will fail. But if my manager only cares about speed, why should I care about anything more? Why should an engineer be responsible for a manager's poor decisions?
Because the corporate hierarchy demands it. Front-line workers are expendable, much like front-line soldiers. Corporations are not democracies, and neither are militaries.
If your job is to write and commit code, you are the corporate equivalent of infantry. In short: a grunt.
And if I am being treated as a grunt, I will act like a grunt aka I don't care more than my bosses. I will even abandon ship at the sign of incompetence and point out the incompetence to others.
Regardless, it’s an important life skill to learn that it’s generally not enough to ask people to do what you want, you need them to actually “buy in” to what you want, and then they’ll actually care enough to at least try make it happen.
This applies to managers and their employees, or also when trying to get on the ground employees to adopt a product initially sold to their manager/execs.
I don't feel the need to get "buy in" for bad faith managers.
An overly broad statement which, my experience, is not true. Managers come in many shapes.
Although, I suppose ironically, if you act in bad faith as a response to this perception, then I think that perception will quickly become true.
If the project fails, and the manager tries to blame it on you, they are de facto acting out of bad faith given the request. Documenting it just identifies this. If the project succeeds or fails, but no one blames you, then the documented request is just there for the record.
If we don’t know if anyone will want the product, the quality of the product is less valuable than validation of product market fit.
Later, I care much more about avoiding accidental complexity and having a great technical foundation.
This sounds like an improvement over the opposite, a code base that is rarely touched and uses eol frameworks. Software is a living thing and if you don’t act as a ruthless gardener you wind up a museum curator with 1990s DEC hardware running in the 2010’s.
The right balance of staying current and not reinventing the wheel is not trivial.
And yes, 10 years is quite possible these days. Besides the infamous Javascript framework churn, things are moving quite a bit slower in recent years.
In part, this is due to the risks inherent to change have been spread out over 10 years instead of backloaded to the end.
Practicing mitigating risks by proactively taking them has value of its own. The parable of ergodic cakes shows this.[0] What % of cakes will fail if 10 people each bake one, versus one person baking 10 over time?
reactjs came out in 2013, so it meets the 10y metric, but it has some internal churn, such as classes, hooks, etc
There are PDP-11s running nuclear plants today with support contracts to keep them running until 2050.
PDP-11s.
Indeed, business people do not typically model depreciation curves for software as they should. That doesn't mean that the plant control system becomes less valuable (the value is tied to the operation of the plant, probably for the lifetime of the plant), but in many other situations it does mean that.