n8n
The most expensive way to improve a process is to automate it first. You build the workflow, add an AI agent on top, wire in a few integrations, and only then notice the step never needed to exist. Elon Musk has a five-step process for avoiding exactly that, and the order is the entire point. I am an AI automation engineer, and I run this same order on a GoHighLevel or n8n build before I add a single thing. Here are the five steps, why automation comes last, and the open-source tool I built so the order never gets skipped.
Key Takeaways
- The algorithm runs in a fixed order: question requirements, delete, simplify, accelerate, automate.
- The order is the whole point. Automating a part that should be deleted multiplies the waste.
- Tesla's flufferbot is the cautionary tale: a robot automated and optimized a part that should not have existed.
- Engineers and language models share one bias. They add. The algorithm is a forcing function against it.
- I built a Claude skill that walks the five steps in order, refuses to skip ahead, and refuses to fabricate quotes.
What is Elon Musk's 5-step algorithm?
Musk's algorithm is five steps, run in a fixed order: question every requirement, delete any part or process you can, simplify what remains, accelerate the cycle time, and automate. He laid it out in Walter Isaacson's 2023 biography, in the chapter called The Algorithm, and walked through it on camera in his 2021 Starbase interview with the Everyday Astronaut.
- Question every requirement, and make it less dumb.
- Delete any part or process you can.
- Simplify what is left.
- Accelerate the cycle time.
- Automate, and only last.
Make the requirements less dumb. Requirements from smart people are the most dangerous.
Step one is the one people skip. Question every requirement, and make it carry the name of a real person, not a department. Musk's own phrasing is to make the requirements less dumb, and his warning is that requirements from smart people are the most dangerous, because everyone else stops questioning them. Step two is to delete the part or the process. His rule of thumb: if you do not end up adding about ten percent of it back later, you did not cut enough. Step three, simplify, runs only on what survived. Step four, accelerate, speeds up the cycle, but only after the first three. Step five is automation, and it comes last.
Why does the algorithm automate last?
Because automating or optimizing a step that should be deleted just multiplies the waste. A clean, fast n8n workflow running a step that should not exist is still running a step that should not exist, now wired into three other scenarios and an AI agent, and far harder to pull out. The order spends the cheap question first, should this exist, before the expensive one, how do we automate it. Most teams run it backward. They automate, then optimize, then maybe simplify, and only question the requirement when the bill arrives.
What was the Tesla flufferbot?
Tesla learned this on the Model 3 line. A machine was built to place a fiber mat on top of each battery pack, meant to dampen noise. The team automated the placement, sped the robot up, and optimized how it applied glue. The robot kept failing, because soft fiber is hard for a gripper to handle. On the Q1 2018 earnings call, on May 2 2018, Musk said the company had automated some pretty silly things. They tested cars with and without the mat, found no measurable difference in noise or safety, and deleted both the mat and the robot.
A week and a half earlier, on April 13 2018, he had put it more bluntly in a post on X: excessive automation was a mistake, and humans are underrated. The team had run the algorithm backward. They automated, accelerated, and optimized a part, and only at the end asked whether it should exist. The honest answer was no.
What is the Backward Check?
There is a reliable tell for when you have gone backward through the algorithm. You are on your third fix for the same problem, and every fix added something: a retry, a filter, a wrapper, another node, another agent. The symptom moves but never resolves. When that happens, stop adding. Ask one question. If this part did not exist, what would break? If the honest answer is nothing measurable, the fix is deletion, not a fourth addition. It is the single most expensive pattern I see, and it is almost always a part that step one or step two would have removed.
How do I run it on GoHighLevel and n8n?
This is not just a factory story. It is how I work. When I take on a GoHighLevel or n8n build, I run the algorithm on the setup before I add anything, and most of the value lands in steps one and two, before automation even enters the picture. That audit-before-build discipline is what an AI automation engineer actually does, and it is the opposite of bolting on more zaps and more agents until the bill gets scary.
Here is one from a GoHighLevel setup. An operator had a new-lead alert that fired into a team channel, and it kept firing twice for every lead. The earlier fixes had all added something: a delay, then a filter, then a deduplicate step in n8n. The alert still double-fired. Running the algorithm, the step-one question was simple. Why are there two events at all? There were two triggers on the same record, one native GoHighLevel workflow and one inbound webhook. The fix was to delete one trigger, then delete the delay, the filter, and the deduplicate step that existed only to mask it. One source event, one alert, three fewer moving parts. That is the difference between adding automation and engineering it.
The Claude skill that enforces the order
I turned this into a Claude skill so the discipline runs every time, not just when I remember it. It is called five-step-algorithm, it is open source under the MIT license, and you can install it from GitHub. It walks the five steps in order, writes the work out so you can see what it questioned and what it deleted, and refuses to skip ahead when you ask it to just tell you what to automate.
It runs the Backward Check on its own when you have been iterating on the same fix, and it refuses to fake a Musk quote. Every quoted line carries a source and a year, and anything it cannot verify it paraphrases in its own words instead of putting words in his mouth. It is the same method I use on client work, packaged so anyone can run it on their own process, automation, or AI agent.
Frequently asked questions
What is Elon Musk's 5-step algorithm?
Five steps in a fixed order: question every requirement, delete any part or process you can, simplify what remains, accelerate the cycle time, and automate last. It is documented in Walter Isaacson's 2023 biography of Musk, in the chapter called The Algorithm.
Why does the algorithm say to automate last?
Because automating a part that should be deleted multiplies the waste. Automation is expensive to build and maintain, so it pays off only after you have questioned the requirement and removed what does not need to exist.
What was the Tesla flufferbot?
An automated machine on the Model 3 line that placed a fiber noise-damping mat on the battery pack. It kept failing, and testing showed the mat made no measurable difference, so Tesla deleted both the mat and the robot. Musk described over-automating silly things on the Tesla Q1 2018 earnings call, on May 2 2018.
How do I apply the algorithm to my own automations or AI agents?
Run steps one and two before anything else: question each requirement and delete what you can. If you have already tried several fixes and the problem persists, run the Backward Check and ask what would break if the part did not exist. The five-step-algorithm Claude skill walks this for you and refuses to skip steps.
The order is the whole trick. Question, delete, simplify, accelerate, automate, and never let a later step jump the queue. I am an AI automation engineer, and I run this on a GoHighLevel or n8n setup before I build, so you do not pay to automate steps that should be deleted. If you want it applied to your stack, book a call or send a brief. If you want the tool, install the skill on GitHub.
Sources
- Walter Isaacson, Elon Musk (Simon & Schuster, 2023), chapter The Algorithm.
- Everyday Astronaut, Starbase Tour and Interview with Elon Musk (2021). everydayastronaut.com. Retrieved 2026-06-26.
- Tesla Q1 2018 earnings call, May 2 2018, reported by Slate. slate.com. Retrieved 2026-06-26.
- Elon Musk on X, April 13 2018. x.com/elonmusk. Retrieved 2026-06-26.
GoHighLevel