This recalibration reflects the impact of Claude Cowork and its autonomous mode on the software lifecycle. While traditional development followed a predictable curve, AI-native engineering front-loads code generation. This mirrors Andrew Ng's analysis that the primary bottleneck has moved from writing code to strategic verification.
You should adjust roadmaps to account for this "illusion of completion" where a functional prototype no longer signals a task is nearing the finish line. As Claude Desktop integrations accelerate the initial build, focus must shift toward the final sliver of work — debugging and polish — that still requires human time.
AI has broken my mental model for time estimates. I used to lean on Pareto's 80/20: first 80% of the work takes 20% of the time. I think the underlying principle is still true, but both ratio and curve have moved. Getting to "basically works" is nearly instant, sweating every single last detail still takes a decent chunk of time. Maybe it's time to switch to 90/10, 95/5, 98/2?
More like this
AnthropicApr 24