Predicting the next few years of any fast-moving technology is a good way to be wrong specifically, but the general direction is visible if you look at where the current tools are strong and where they're still obviously weak, because that gap is where the next round of investment is clearly going.
Less magic-button, more embedded tooling
The current framing, describe a whole game in one prompt, get a finished product, is compelling for a demo but isn't actually how the strongest tools are heading. The more durable direction is AI assistance embedded at every stage of a normal development pipeline: a level designer gets a first-pass layout to edit instead of a blank grid, an artist gets a variation generator instead of a blank canvas, a programmer gets a first-draft implementation to review instead of an empty file. Assistance at each step, not a replacement for the whole process.
The testing loop gets more central, not less
As generation gets cheaper and faster, the bottleneck shifts even further toward verification, checking whether what got generated is actually correct and actually good, which is exactly the "plays its own build" loop covered elsewhere on this blog. Expect more investment in automated testing and playtesting specifically, because that's what turns fast-but-unreliable generation into something a studio can actually ship on.
Where the real friction still is
Not generation quality, which keeps improving on a predictable curve. The friction is judgment: is this fun, is this balanced, is this the right creative choice for this specific game. No visible trend line suggests that gets automated soon, and the tools that win long-term are likely the ones that make the judgment calls faster to make and test, not the ones claiming to make the calls for you.
What this means if you're building or buying into these tools now
Pick tools based on how well they handle the boring reliability problem, does it test its own output, does it fail gracefully, not on how impressive the demo prompt looks. The demo is marketing. The testing loop is the product. That's true of the tools covered across this whole blog, this one included, and it's the single question worth asking about any AI development tool before you build a workflow around it.