Lots of good answers about here. I think the individual coder's struggle is pretty well covered via many good points. But I think we haven't looked at this question from much of a big picture. So here's some of the challenges that make programming difficult today, on a bigger scale:
Weirdly, I think the biggest difficulty is that we are much better served. A huge amount of what we do is well paved. React, bundlers (which are fairly isomorphic to each other webpack/rollup/esbuild/parcel/snowpack/vite), Systemd, Kubernetes, React, gRPC/protobuf, npm/node, even uring or eBPF... the list of well entrenched technologies is high. There's better set ways to do things, well served, than there used to be.
The difficulties show up in a number of fashions. First, it precludes innovation, is highly stasist, when there is a humming along underbelly that maintains life. In the Matrix, the Elders of Xion admitted it was just highly automated machines that kept life alive, and in some ways our ascent upwards has decoupled us similarly; we're rarely liable to go back & reconsider the value of trying other ways. We're kind of "stuck" with a working set of system layers, that we don't innovate on or make much progress on. Our system layer is pretty old and mature. Everything is ingrown & interlocked, depends on the other things.
When we do try to break out, rarely is it precisely targetted reconsiderations: often the offshoot efforts are from iconoclasts, smashing the scene & building something wildly different or aggressively retro. Iconoclasts seek pre-modern times, rather than a new post-modern alterations or rejiggerings.
Another difficulty with having vastly more assumed is that there's less adventurers in the world, less find-out-the-truth/roll-up-your-sleeves/dig-in/read-through-the-source mentality & experience (and more looking only on the surface for easy "solutions"). Being generally well served means we rarely go off the beaten path. So our wilderness survival skills/resourcefulnesses are attrophied, and newcomers are less likely to have developed these deep hunting skills that used to be both simpler (because our systems back then were simpler, had less code) and more essential. A lot of these modern works aren't even that hard to dig into. But there's shockingly few guides for how to run gdb or a debugger on systemd, few guides to debgging kube's api-server or it's operators, few people who can talk to implementing gRPC.
I don't think we're at crisis levels at all, but i think the industrial stratification & expectations of being well served will start to haunt us more and more across decades, that we'll lose appreciation & comprehension (alike the Matrix problem), we'll fail to make real progress.
GraphQL is an interesting case-study to me. It rejected almost all common web practices & went back to dumb SOAP like all-purpose endpoints. The advantage of just getting the data you ask for was good, of not having to think about assembling entities, of having a schema system. But so many of these things are things the actual web can and should be good at. We spent a long time having schema systems battle each other, but that higher-level usable web just kept failing to get built out, so total disruption made sense. We still haven't a lot of good replacements for GraphQL, still haven't made strong gains, but still, it feels like GraphQL is somewhat fading, that we're less intimidated by making calls & pulling data than we used to be.