Most AI projects start exactly where SpaceX ends.

SpaceX runs an internal method they call “the Algorithm.” Five steps, in strict order, applied to every part of every rocket. The fifth and final step is to automate. The first four are question every requirement, delete whatever you can, simplify, and accelerate. Automation goes last, and it goes last on purpose.

Here’s the thesis, before any tool: the expensive mistake with AI isn’t automating the wrong thing. It’s automating before you delete. When you bolt AI onto a process you never questioned, you don’t gain efficiency. You build a faster version of the waste you already had, now harder to change because you put technology on top of it. Deciding what to delete before you automate is exactly what AI Maestro does before touching a single tool.

The Algorithm, and why the order matters

Marc Andreessen and Michael McGuiness laid out the method with the same detail Walter Isaacson published earlier. The five steps are worth reading in order, because the order is the method:

1. Question every requirement. And each one should come with the name of the person who asked for it, not a department. The requirements from the smartest people are the most dangerous, because nobody dares to question them.
2. Delete any part or process you can. The rule is aggressive: if you don’t end up adding back at least 10% of what you deleted, you didn’t delete enough.
3. Simplify and optimize. But only after step two. The common mistake is simplifying something that shouldn’t exist.
4. Accelerate cycle time. Only after the first three.
5. Automate. It comes last.

Musk himself says the mistake in Tesla’s factories was exactly this: trying to automate first, before requirements were questioned and processes deleted. They automated the chaos. And automated chaos is more expensive to dismantle than manual chaos, because now there are robots, code, and contracts defending it.

AI does the method backwards

Look at how the average AI project starts. Someone decides the company needs to “use AI.” A tool gets picked, wired into a process that already existed, and asked to make it faster. That’s jumping straight to step five. Nobody questioned the requirement. Nobody deleted a single step. Nobody asked whether the process should exist.

The result is already measured, and McKinsey published it. In its report on agentic organizations, 78% of companies say they use AI and over 80% see no impact on operating profit. The reason the report gives is the same one Musk gives about Tesla: they plugged AI into a flow designed years ago for people, the speed entered one isolated step, and the bottleneck moved to the next one. I worked through that in the post on why adoption isn’t transformation. SpaceX’s method is the other side of that number: the impact doesn’t show up because you skipped the four steps that came before automation.

Deleting is the part nobody wants to do

Of the five steps, the one AI makes easiest to skip is the second. Delete.

Deleting is political. Questioning a requirement means going to the person who asked for it, often someone more senior than you, and asking why it exists. Deleting a process means somebody will feel their work matters less. It’s uncomfortable. And AI offers a tempting way out of that discomfort: “we don’t have to delete anything, we’ll just automate it.” That sentence is the entire trap in seven words.

I’ve been doing this for 36 years, since 1990, and I’ve watched the same pattern in every cycle: digital, web, mobile, cloud, and now AI. Each time, the temptation was identical. Instead of questioning the inherited process, put the new technology on top and call it modernized. The winners in each of those cycles weren’t the ones who automated fastest. They were the ones who first had the stomach to delete. The technology changed five times. The discipline of deleting before automating didn’t change once.

What IQ Source does about it

Here’s what changes for a company about to start its AI project. AI Maestro is, in practice, the first three steps of the Algorithm applied to your operation before automating anything. Two months questioning requirements, mapping how the work actually gets done, finding what can be deleted and what can be simplified. Only at the end comes the Go/No-Go gate that decides what’s worth automating.

And the part that sets it apart is the same part that sets SpaceX’s method apart: often the right answer about a process isn’t “automate it.” It’s “delete it.” That ties to something I’ve written from another angle. In the post on what not to automate the point was the damage of slop, the junk content that erodes trust. Here the point comes earlier: before you decide how to automate well, you decide what shouldn’t exist in the first place.

Before your next AI project, ask the room one question. What part of this process could we delete entirely? If nobody can name a single one, you’re not ready to automate yet. You’re ready to speed up waste, put a technology cost on top of it, and bolt it to your operation for years. Step five is tempting because it looks like progress. Steps one through four are the actual work.

Question and delete before you automate, with AI Maestro