Now, if you're willing to accept a lower probability of success, the numbers get even more interesting.

Claude will tackle a task that takes about two hours for a human being, and it will get that right about 50% of the time. Now, 50% might sound low, but here's the key thing to remember: it won't take Claude two hours to do that task. It'll rip through it in a matter of minutes.

So there are fairly dramatic productivity improvements available to software engineers who can make use of technologies like Claude Code and who have the confidence to say, "No, that failed. Let's start it again and try again."

This is where that abundance mindset comes in. With a two-hour task, you might get it wrong half the time. But Claude completes it so quickly that you can afford to try multiple times. The total time investment is still dramatically less than doing it manually.

And of course, this success rate is also improving on that same exponential curve. By the end of next year, we'd expect 50% success on much longer, more complex tasks.