The Veterans Still Matter
August 23, 2026
I’m about to begin my part-time Master’s program at UPenn. Doing a program like this is a bucket list item for me, and I’m excited and nervous to be learning again.
I got to meet some of the others in the program last night for an unofficial happy hour. The range of people there was quite interesting - from Engineering Managers to full-time students, to those who have never had a software job. I felt like I was coming in with a unique perspective after having worked and lived in the Bay Area for 8 years.
One of the questions I found interesting was if I felt, in my time in industry so far, if the number of senior software engineering roles was capped. The thinking was that because it feels very difficult to find a job, there’s inevitably going to be people who can’t find opportunities to enter, and even those that stay in industry would eventually be laid off because of AI productivity gains.
My response was that I thought this was adopting a scarcity mindset on the industry in general. I think the right companies will continue growing their slice of the pie, in a way that contributes to more headcount overall, even with AI productivity gains. If you’re truly at a technology company, SWEs are the primary bottleneck to business impact, not just “more AI”. So it’s in the best interest of the company to continue hiring the right engineers, which should create a flywheel of new roles as the economic pie continues to grow.
I do think AI tools have made more companies treat software engineering as cost centers [1], but I also think many of these companies are finding out in a very hard way that they still need their veterans. There seems to be a level of intelligence seniors and staff+ have that LLMs cannot consistently replicate. And so, the shape of the world I’m betting on is that there’s going to be either the same number or more of senior SWEs + AI tooling, driving impact unless there’s another paradigm shift in AI that causes everyone to freak out again.
Maybe it’s not the level of intelligence that helps give seniors their job safety, but the ability to have one foot in the digital world, and the other in the real world. As a person, I don’t need to have the vast knowledge and know-how to quickly do everything the way a LLM does it, but I need to have the systems intuition to know when I see something that could catastrophically break many things, and also prioritize what would have the most impact from a product point-of-view.
Also, as a person, scoping and planning is deeper and richer because it requires talking and building relationships with real people, something that is always for some reason consistently undervalued, and is needed for any sufficiently meaningful project that is not just a vibecoded slop project.
My views on this have evolved a lot over the last year. Generally speaking, I was more of a doomer while I had a job and was grinding out Claude Code every day (maybe it’s my own version of AI psychosis). I partly attribute that to agentic coding being quite new (barely over a year old!).
But having spent some quality time away from the day-to-day SWE stuff over the last nine months and having seen AI intelligence plateau, I’m doubling down on my perspective that AI tooling is just that: a tool. And that any FUD or anxiety over how the current toolset will affect the senior engineering pool will stabilize in the long term, as organizations realize that human intelligence still cannot be beat for the product work that matters.
- Thinking back to just a few months ago when everyone (cough cough Dario) was saying SWEs and white collar were getting automated out very soon. I think anyone that has this perspective is vastly underestimating how much work the average white collar job has, and how slow it is to roll out innovative technology in older organizations. There’s a ton of complexity in reality that you cannot effectively feed into a LLM.