Real engineering work is not linear.
On a real project, you don’t “follow the workflow.” You bounce between modes.
You compare two options. You zoom out to sanity-check the strategy. You question the framing. You keep placeholders because the data isn’t there yet. You chase the one parameter that actually moves cost. And only occasionally do you get a clean, straight line to “done.”
Most tools are great at one of these modes and awkward at the others.
I wrote a post on the workflow taxonomy we use internally when designing utility-scale solar PV layouts — less as UI flows, more as modes of reasoning engineers switch between as the project matures.
If you’ve ever felt like your process is “messy” because it’s not linear, this is for you.
Link in comments.
Depth shows up in strange places.
Today it was an almost two-hour run – steady effort, no shortcuts – with Our Oriental Heritage playing the whole time.
Neither is efficient. Neither is optimized. Both demand the same thing: attention, endurance, and staying with discomfort without reaching for an exit.
You can’t skim a run.
You can’t summarize a 40-hour book.
You just keep going – step by step, chapter by chapter – and something slowly settles into place.
That’s the long arc.
AI amplifies noise as much as it amplifies intelligence. So filtering becomes strategy. Understanding leads to ideas; execution is what turns them into value.
I'm not saying that everyone who holds equity in a startup should work on weekends.
But I do notice that many people who do work on weekends are the ones who own equity.
Having a prototype idea over the weekend.
Writing it down on Monday morning.
Asking Zed/GPT-5 to build it while you're making coffee and staring into the void.
What a ridiculous, amazing world we live in.
"Don't build shitty products for problems that don't exist."
This brutal advice from Maksim Markevich hit different. After spinning off from his first AEC tech company Kreo, he's now building PVFARM - solar design software that's pulling in $40K average revenue per customer.