Delete before you automate. The 5-step method that stops teams from building expensive infrastructure around a process that should never have existed.
You've added an approval step nobody remembers approving. A dashboard tracking a metric nobody acts on. A sync meeting that exists because a sync meeting used to be needed. And now someone wants to "fix" it by automating the whole thing, faster.
That's backwards. Automating a broken process doesn't fix it. It just makes the broken version run at scale, and much harder to ever remove.
This is the method Elon Musk drills into every engineering team at SpaceX, Tesla, and Boring Company (documented in Walter Isaacson's biography Elon Musk, 2023). Five steps. Always in this order. Skip one and you optimize, speed up, or automate something that should have been deleted first.
1. Question every requirement Every requirement needs a named human owner: a person, never "legal," "the department," or "best practice." Then interrogate it, especially if it came from someone smart. Unquestioned requirements from smart people are exactly how bad ones survive. No named owner? Delete it.
2. Delete the part or process Before optimizing anything, try removing it entirely. The tell: if you're not adding back at least 10% of what you deleted, you weren't aggressive enough. Most people are too conservative about cutting. Bias hard the other way.
3. Simplify or optimize Only after 1 and 2. "The most common error of a smart engineer is to optimize a thing that should not exist." Don't polish a step you should have deleted.
4. Accelerate cycle time Speed up whatever survived steps 1-3. Doing this earlier just gets a bad process to a wrong answer faster.
5. Automate Dead last, always. Automating a step before it's been questioned, cut, and simplified means building expensive, rigid infrastructure around something that shouldn't exist.
cp the-algorithm.md ~/.claude/commands/the-algorithm.mdThen in any Claude Code session:
/the-algorithm
[describe the workflow / paste the process / paste the agent or automation spec]
Output:
1. Question: [which steps have no named owner? which survive interrogation?]
2. Delete: [what can be removed entirely, aim to cut at least 10% more than feels comfortable]
3. Simplify: [the leanest version of what survives]
4. Accelerate: [where what's left can move faster]
5. Automate: [only what's left, and only that]
Then the rebuilt workflow, end to end, with what was cut and why.
- Before building any new agent, tool, or automation. Question and delete BEFORE you propose the build
- Any internal process that's quietly accumulated approval steps, hand-offs, or review gates over time
- A sales process, tech stack, or team structure that's gotten heavier without getting better
- An offer, sequence, or funnel with declining conversion. Check for dead steps before bolting on more automation
- Before scaling anything. Automating a broken process just breaks it faster, at volume
Good innovation is simplification. Adding complexity is not progress. The usual failure isn't laziness, it's skipping ahead: optimizing what should have been deleted, automating what should have been questioned. Fewer moving parts break less. Do less, but make sure it's the right things, done in the right order.
Before (typical "improvement" request):
"Can we add an approval step before this goes out, and build a dashboard to track how long approvals take?"
After (The Algorithm):
1. Question: Who owns the requirement that this needs approval at all? If no one can name a person and a reason, it's a leftover from when the team was smaller.
2. Delete: Remove the approval step. Replace it with a clear standard the sender checks against themselves.
3. Simplify: One checklist, three items, applied before sending. Not a review queue.
4. Accelerate: Sends now go out same-day instead of waiting on a reviewer's calendar.
5. Automate: Nothing yet. Automate the checklist check only once the manual version has run clean for a month.
Built by Strategy Sprints, Simon Severino.
Before Strategy Sprints, Simon personally coached 2,000+ teams on sales and negotiation, including Google, BMW, and Airbus. Strategy Sprints now teaches B2B founders the 8 Steps of the Repeatable Sale: how to run sales, offers, and internal operations with fewer moving parts and more control. The Algorithm is one of the tools we use to decide what to cut before we build anything new.
Want the full method and a spotter to run it with you? → strategysprints.com