It’s not necessarily faster to do this for a single task. But it’s faster when you can do 2-3 tasks at the same time. Agentic coding increases throughout.
Until you reach the human bottle neck of having to context switch, verify all the work, presumably tell them to fix it, and then switch back to what you were doing or review something else.
I believe people are being honest when they say these things speed them up, because I'm sure it does seem that way to them. But reality doesn't line up with the perception.
True, if you are in a big company with lots of people, you won't benefit much from the improved throughput of agentic coding.
A greenfield startup however with agentic coding in it's DNA will be able to run loops around a big company with lots of human bottlenecks.
The question becomes, will greenfield startups, doing agentic coding from the ground up, replace big companies with these human bottlenecks like you describe?
What does a startup, built using agentic coding with proper engineering practices, look like when it becomes a big corporation & succeeds?
That's not my point at all. Doesn't matter where you work, if a developer is working in a code base with a bunch of agents, they are always going to be the bottleneck. All the agent threads have to merge back to the developer thread at some point. The more agent threads the more context switching that has to occur, the smaller and smaller the productivity improvement gets, until you eventually end up in the negative.
I can believe a single developer with one agent doing some small stuff and using some other LLM tools can get a modest productivity boost. But having 5 or 10 of these things doing shit all at once? No way. Any gains are offset by having to merge and quality check all that work.
