Software's Centaur Era
19 points by carlana
19 points by carlana
An honest reflection on the the future of code writing with LLMs. Some changes are inevitable but the future is different from how it is depicted by today's CEOs and social madness.
the future is different from how it is depicted by today's CEOs and social madness.
Sadly, that is the case for many reasons beyond AI...Plenty of CEOs and their mad teams are all too happy to listen to the likes of management consultants who will whisper sweet nothings indicating that offering X or platform Y or solution Z or "supposed innovation" W will solve all the problems of the CEO, so might as well get rid of thosw pesky meatbags who have independent thought and actually bring value to an org.
You can hire a grandmaster-level software engineer, leave them pretty much alone, and trust that they will start making valuable contributions to long-term projects and will reliably be making your software system better.
This statement struck me as profoundly wrongheaded and naive. In my experience, there is little correlation between being good at programming and being motivated to produce valuable contributions to an existing codebase -- if anything it's the opposite. There is a huge amount of friction to overcome when learning to work in a new software system, and the better programmer you are, the more likely you are to find someone else's code objectionable. I've known many first-class programmers who were terrible procrastinators, and have largely gotten away with it because they could quickly produce the necessary code by deadline.
Grandmaster-level software engineers have a broader skill set then you're counting here and meet the description in the article.
The people you are describing are something else. Good coders, maybe.
I encourage these people to not vibe-code, but rather vibe-write-documentation, vibe-understand-requirements, and vibe-soften-their-MRs, which can also help those individuals perform at a higher level.
Yes there are definitely people who are "grandmaster-level" at something smaller than the task of shipping valuable software as part of a team. If you need to be steered in order to produce valuable contributions, then you remain competitive as long as you produce more marginal value per unit of steering than the AI does. "Brilliant but needs guidance" is going to be the engineering archetype that suffers most, because that unfortunately seems to be profile that these AI coding agents occupy also.
I am glad that I only have 8 years left in the industry. Having said that I think human ai (with a skilled thinking person) are an unbeatable combination. I think the future belongs to those who can ask the right questions.
I'm trying to eek out another 10 years, and after that then pivot to easier work (mentally easier work, i mean)...but i'm fearful that too many orgs will believe that they can do more with less humans...and that, i won't reach those last remaining 10 years.
Chess and software engineering are different things. I would have taken the example of CNC and manufacturing. Or secretaries and word processing and spreadsheets. When the economy is doing well we go to 5% unemployment but the jobs are different. They are new, different kinds of jobs. perhaps we will all be hyper-local poets, making poems about our neighborhoods. Maybe that will be the future jobs.
Or perhaps it won't be so drastic and we will each be senior engineer + product manager + sales for some product that before needed a whole team.
Or perhaps we will each be one person companies, you know, like the village fisherman 3000 years ago.
I recently ran across the idea of a transition to "product pods", basically like ordinary corporate software teams but smaller and with a different role ratio, more like one product manager + one QA manager + two developers. The theory is that with AI assistance in all the roles, two developers can keep up with all the ideas from one product manager, and it'll take a full-time QA person to maintain whatever process is necessary for the codebase to stay sane. Or maybe the numbers aren't quite right, or maybe it won't work at all — but at least it seemed like a coherent thing to try.