So, You “10x’d” Your Work…

There are lots of ways to “10x” things.

• Drive 300 miles per hour down a busy city street.
• Eat 15,000 calories in one meal.
• Own ten dogs.
• Have ten children.
• Make hundreds of friends.

All these things have fairly obvious consequences and side effects. Even having lots more friends would force you to embrace a much shallower notion of friendship than you could otherwise enjoy. So, why is it that when AI fanboys speak of “10x-ing” their productivity, they never say a thing about the side effects?

It’s because they are speaking and behaving recklessly.

Increasing productivity by multiples is already an extraordinary claim. Doing so without incurring great risks and repercussions is simply unbelievable. Unless of course, the claimant has low standards. If you don’t care about quality, you can achieve any other goal you like.

When developers ask AI to write code for them, that may happen very fast. But some things are still slow:

• the developer’s process of learning about what they are building
• the developer’s process of deciding what to do next
• the product owner’s processes of learning and deciding
• the end users’ processes of learning and deciding
• the time it takes for an elusive bug to reveal itself during normal use

If someone claims to have 10x’d their product development, my inquiry would look something like this:

1. Tell me your test strategy. (Oh, you don’t know, because you asked the AI to test everything for you.)
2. I would ask your AI for its test strategy, then. (It produces a lot of text that may or may not be what it actually did.)
3. I would ask your AI for its analysis of the weaknesses of that strategy. (It produces text that may or may not represent important weaknesses.)
4. I would review its claims about the risk coverage and product coverage achieved by its own checks.
5. I would learn enough about the product to make my own analysis of product risks and how well they are addressed by the test strategy. (This would either reveal important holes in the testing or it wouldn’t. Either way, I could begin to take responsibility for the strategy.)
6. I would acquire and review the chat transcripts that show how the product came together. (This would give me clues about weaknesses in the development approach and potential undiscovered bugs.)
7. In the course of doing this, I would review test artifacts and results. (My goal would be first to make sense of them, and then to evaluate them against the claims the AI made about them and also against the overall business risk.)
8. I would especially focus on security problems and user experience problems. (These are common in slop-coded products.)

Steps 4-6 cannot be significantly accelerated by AI, because the whole point is to achieve human understanding of how we know that the goal was achieved. That understanding cannot be manufactured outside a human mind and imported. Instead, some actual person must construct it. And that person must be credible.

We have all seen AI confidently assert things that are not true. I’ve seen claims that tests have been run, even when they weren’t; bugs that were fixed, even though they weren’t; products that are working, even though they don’t work at all. This is why we can’t just take AI’s word for it. This is why I say that “AI” stands for “automated irresponsibility.”

Testing is really not about quality, it’s about responsibility. What does your “10x” productivity mean, and how do you know?