History repeats itself: Aerospace & AI-software-development

The Analogy:-

• When aircrafts were first being designed, they had to deal with the unpredictability that arose from working with fluids (that too, when fluid dynamics was not even a proper subject)
• 1920s-1930s: "Works most of the time"
• When it did fail, invariably it was unexpected and it happened at the worst possible situation; we then end up in a mini disaster, if not worse!
• Eventually, aircraft ENGINEERING, as a discipline, figured out how to systematically handle the chaotic nature of Air, ultimately delivering "deterministic-flights" even in the bad weather!!
• All Aerospace-engineers turned into Systems-Integrators and/or into Simulation/Modeling people. What does it mean for those of us developers who're struggling to survive this AI-armageddon? The Challenges:
• STANDARD CONTROLS & METRICS: Engineering may likely reassert itself into AI-only software-development, once we have globally well-understood and well-standardized metrics + controls, to track progress and to track quality of the work done by AI. if everyone's got their own standards, it's not going to be enough.
• A BIT CLUELESS: If you listen to the famous names and what they're publishing on the various media, as of April 2026, we still have insufficient control on what the models do. I believe, we must seek 100% control. (Dare I say this? Just frequently enough, we sometimes have no clue re: "how/what" the AI did! Forget about "control").
• GOING NOWHERE FAST: As someone who started a career with 3-yrs-long projects, I now ask all end-users: "I CAN COOK SO FAST NOW. BUT CAN YOU EAST FAST ENOUGH?". Let me answer on behalf of the end-users: "No!" Survival Tips (part 1 of 2):-
• STOP: Stop wasting time hoarding/curating prompts/templates. My own recent experience was that GitHub's CoPilot's CodeReview was FAR-superior to my own curated set of "PR Review prompts" (run across multiple agents); FYI: I had copied these from famous people's publications.
• NEW FOCUS: Your career now is NOT to decide which model/framework works best. Focus instead on how you can capture metrics & trends, the change-impact measurements, the cost-estimates, etc .. especially for already-deployed apps.. .. to help others make those decisions.
• Your career will ALSO be about GUARANTEEING / proving that the ROLLBACK in production, was 100% precise. What metrics are needed to provide such a guarantee, and how will you have them in place prior to starting the rollback?
• With AI, prepare for rollbacks to be routine. My gut says: This is the 1st true engineering within the AI-world. SRE may need to adapt.
• BEST TOOLS: Like a traditional engineer, your tools should be the best available: (1) best PR-review tool built & maintained by some corporation (2) best design-review built & maintained by some corporation (3) etc.. ..
• Open Source simply can NEVER keep up with this mind-blowing pace of progress. So, focus on commercial-tools (hopefully those that'll "Not vendor-lock you forever").
• MINDSET CHANGE: Start chanting "Can Not control what you do Not measure". Drink that "kool-aid". Believe in it.
• You must truly believe that it is NOT engineering to just break-apart a prompt into "Role, Context, Objectives, Constraints & Communication-style". Possibilities in the Future:-
• SIMULATION/MODELING: Following the trajectory of the Aeroplane-industry, its quite likely that Engineering will be more "Simulation & Modeling" (of the problems and of the future-solutions), and less about actual software-development (see more below). True Visionaries are so rare! That's wonderful for our job-security! As, we get to have jobs to do modeling of endless scenarios for the clueless C-Suite.
• PRODUCT-MANAGEMENT:- With AI "listening in" into all calls/meetings and with access to all information, it's certainly quite possible this role is no longer human! ☹️. Unless, .. .. like Steve Jobs and Henry Ford (as well-known examples), you know what the customers WILL want.
• REQUIREMENTS-ELICITATION:- This is the MAKE-or-BREAK MOST-IMPORTANT aspect of software-development, that .. .. if AI somehow nails it, .. .. it will destroy all prospects for humans. As one possibility, the day that the employees have a CONSTANTLY ATTACHED AI, that knows everything about that person's role and a bit about their personal lives, .. that will be the day there will be no more IT-Dept. Google MAY soon achieve this in the context of the common-consumer. 99% of the time, end-users do NOT know what EXACTLY do they want. Hence my note above re: "never be able to eat fast enough". It takes a lot of time & effort by us humans, to elicit updated-requirements from a end-user. Survival Tips (Part 2 of 2):- Enroll in a local college re: "Simulation and Modeling". Become familiar with such terminology & tools Sorry. I am Not able to predict whether "feature git-branches" with their own deployed tiers/environments (incl. freshly hydrated new databases/data-repos) counts as "Simulation and Modeling". Feel free to debate about the future. Common Misconceptions about the Wright-brothers:-
• For those who live out their liberal-arts degrees, and for those who don't give a sh*t, and for the uneducated, the Wright Brothers invented the plane! That's it.
• There were many many people who actually "lifted off" quite a bit before the famous success of the Wright Brothers. Their problem? No one could replicate their success ever again.
• From an engineering standpoint, the Wright Brothers spent years (perhaps close to a decade?), looking at it from a true engineering perspective. CONTROL. CONTROL. CONTROL. CONSISTENCY.
• What was the reason the Wright brothers were soooooo obsessed with control & consistency? The one brother who was designing it, was sooooo concerned that his brother would get hurt --- in a TRADITIONAL sense of being brothers. It's a relationship that's hard to explain/understand in this world today. Perhaps it was due to the social / parental pressure back then. BTW, I'm old. My view of the world today:-
• We just FINALLY got past the "It's So Exciting! Another new hot-hot research paper was published. Can't wait to read it and discuss/debate it endlessly."
• We're now in the phase:- "Throw it against the wall, see if it sticks" (Chaotic Experimentation)
• Software-Authoring -versus- Software-development: Claude is PUBLICLY leading in the latter. I'm quite sure Google is doing the same internally too. But, But .. you say ..
• "AI is working so great !!"
• You must be in sales/marketing/general-media/porn.
• Words like "Quality" and "Guarantee/Warranty" never applied to your work.
• And never will.
• Appearances & Impressions do.
• Good for you.
• Enjoy!
• "We're in the hockey stick part of the human evolution. It's different this time."
• Decades may shrink down to years or lesser. But, human-history does rhyme, even if it does not repeat.
• "Nowadays. MVP and TTM have made the "brotherly relationships/objectives" irrelevant. Production is where testing is done / bugs are identified. So, we're unlikely to see consistency, and even less likely determinism."
• Unfortunately, I need to admit that this is a realistic possibility.

If you are a PRACTICING software-ENGINEER (emphasizing the words "practicing" & "engineer"), then you barely skimmed through this article. Even so, you understood everything here and .. perhaps have better insights!!
For all such folks, "L.F.G." !!
If you're more of a thinker/intellectual-elite, then you're more likely to debate this endlessly, perhaps even on some Dwarkesh podcast. Leave me alone.