Coding too fast to collaborate

There's a quiet crisis brewing inside software teams that have gone all-in on AI coding agents, and it has nothing to do with the code itself. It's about what happens to everyone else in the room once one engineer can produce ten times the output alone.

Start with technical design. Teams used to hash out architecture through standups, Slack threads, whiteboarding sessions, ADRs, the whole informal-to-formal spectrum of getting humans to agree on an approach before writing a line of code. Increasingly, engineers are having that conversation with their coding agent instead. It's not malicious, it's just the path of least resistance: the tool is built to be chatty, so asking it architectural questions feels natural, and skipping the colleague who might slow you down feels efficient. The catch is that a chatbot doesn't build the team's collective judgment the way a real design review does. You get speed today and a knowledge gap tomorrow.

Then there's the product side, which hasn't sped up nearly as much. Product managers still spend most of their time on research, stakeholder alignment, and requirement-gathering, work that resists automation because it's fundamentally about human consensus. So when engineering throughput jumps by an order of magnitude and product output doesn't, teams start starving. The backlog empties out, specs get thinner, and some teams have responded by collapsing the exploratory phase into engineering itself, building prototypes first and figuring out requirements as they go. Nobody knows yet if that's a smart adaptation or a recipe for building the wrong thing very quickly.

Code review is where the pressure really shows. Review has always done two jobs at once: catching bugs and spreading knowledge across the team so no single person becomes an irreplaceable bottleneck. Handing that job to an automated AI reviewer solves neither problem well, since AI models tend to share the same blind spots on unfamiliar code and, obviously, nobody learns anything from a robot skimming a pull request. Teams that keep humans in the review loop end up shifting the bottleneck downstream instead of removing it, and review capacity becomes the new constraint on delivery.

None of this means the old practices were sacred or that new ones won't emerge. But the piece makes a fair point: not every meeting, review, or design chat was red tape. A lot of it was how teams built expertise that no single engineer, or agent, actually holds on their own. Losing that while chasing raw velocity is a trade a lot of teams are making without quite realizing they've made it.