Autocomplete was magic in 2023. You'd gasp at a correct three-line suggestion. By mid-2024, you were annoyed when it got things wrong. The magic became the minimum.
Chat blew your mind when it first explained your codebase back to you. Six months later, you were frustrated it couldn't apply the fix itself. So smart edit arrived, and that was incredible. Until you realized you were still babysitting every change. Then agents took over, and for a week you felt like you had superpowers. Now you're irritated when an agent loses context halfway through a task.
Each breakthrough becomes your new baseline before you've finished adapting to it. What amazed you in January is just expected by March. The wonder fades, the expectations ratchet up, and the next thing has to be even better just to register.
This is the hedonic treadmill. And in AI coding, it's running at a pace we've never seen before.
WWDC used to be once a year. You'd watch the keynote. Spend a few months absorbing the changes. Ship something that used the new APIs. Then you had almost a year before the next wave hit.
In AI, we're getting a WWDC's worth of change every month. Sonnet 3.5 to 3.7 to Opus 4.5 in under twelve months. MCP shipped and became standard in a single quarter. Agents went from experimental curiosity to default workflow before most teams finished their planning cycles.
Mobile had WWDC once a year. React came out in 2013; Hooks arrived in 2019 — six years to absorb a paradigm shift. Cloud infrastructure moved at re:Invent pace: once a year, a flood of announcements, then eleven months to figure out which ones mattered.
It's not that you're slow. The clock is running faster than any previous technology shift.
What used to take a decade now takes a year. What took a year takes a quarter.
Why the treadmill keeps speeding up
Two things unlock each new era of AI-assisted development:
Model capability. What the AI can actually do.
Developer trust. What we're willing to let it do.
You've seen the impressive demos that never made it into anyone's real workflow. That's capability without trust. And you've felt the other side too: the gap between what was promised and what actually worked when you sat down to use it.
The eras shift when both line up. A model announcement doesn't immediately change how people work. The capability shows up first. Trust takes longer. You need to see it work on your code, in your repo, on a Tuesday afternoon when you're tired and just need the thing to not hallucinate.
Then trust catches up. And a new era starts.
Here's what makes the treadmill compound: every generation of models makes fewer of the mistakes that burned you before. So trust builds faster. Adoption accelerates. The next baseline arrives sooner. The cycle tightens.
We went from "I'll check every line" to "I'll review the PR" in under three years. Now we're in multi-agent territory: multiple agents working in parallel, git worktrees for isolation, specs as the coordination layer.
And we can already see what's forming after that. Agents in the cloud. Triggered by your SDLC, not your prompts. Proactive. Reaching out when they detect issues before you do. A PR lands, an agent runs analysis. Errors spike, an agent opens an investigation. A dependency update drops, an agent tests compatibility overnight.
The finish line keeps moving
On a Tuesday afternoon when you're tired and just need the thing to work, you don't remember what it was like before. You only notice what it still can't do. That's the treadmill. You're running faster and the finish line keeps moving.
If you feel like you're never caught up, you're not alone. I talk to people building this stuff and they feel it too. The best practitioners I know don't have this figured out. They're experimenting. Sharing what works. Admitting what doesn't. Everyone in this industry is relearning how they work, in public, in real time.
Within eighteen months, the developer who mass-accepts agent PRs today will feel the same way about orchestrating ten agents that they currently feel about tab-completing a single line. It'll be the minimum. The new baseline. And we'll be talking about whatever comes after that.
The question isn't whether that future arrives. It's whether you'll have built the muscle memory to work at that speed when it does.
Written by
Sylvain Giuliani
Sylvain Giuliani is the head of growth at Augment Code, leveraging more than a decade of experience scaling developer-focused SaaS companies from $1 M to $100 M+ ARR. Before Augment, he built and led the go-to-market and operations engines at Census and served as CRO at Pusher, translating deep data insights into outsized revenue gains.