TL;DR
V1 Strip is step one of the V1 Framework. It is the work of removing the assumptions a team has inherited about how a job is done, so the AI can be aimed at the outcome instead of at a frozen version of the existing process. Most AI failures live here, not in the prompt. Strip is cheap to do and expensive to skip.
- Assumptions are the silent killer of AI work.
- Strip the workflow, the tool stack, and the language.
- The output of Strip is a one-page rewrite of the problem.
- If the rewrite still sounds like the org chart, you have not stripped enough.
- An hour of Strip saves a week of prompt fiddling.
In this article
What stripping means in AI work
Strip is the first step of the V1 Framework because it is the step that decides whether the AI is being aimed at the actual job or at a fossilized version of how the job used to get done. Most teams skip it. The AI then dutifully accelerates the wrong thing, and the team blames the model.
Stripping is the work of pulling layers off a problem until only the outcome remains. The layers are inherited: the existing workflow, the existing tool stack, the existing job titles, the existing internal language. None of them are facts about the job. They are facts about how the company has chosen to do the job, in a world where AI did not exist.
In a world where AI does exist, almost none of those layers are load-bearing. Strip is the work of finding out which ones are.
The output of Strip is not a prompt. It is a rewrite of the problem in language that names the outcome and nothing else. If the rewrite still contains tool names, team names, or process names, the strip is not finished.
Why assumptions are the silent killer of AI
Assumptions kill AI work quietly. The prompt looks fine. The model responds. The output looks plausible. The team ships it. Three weeks later the metrics have not moved, and nobody can point to the moment the work went wrong, because the work went wrong before the prompt was ever written.
Here is the failure mode in slow motion. A team is asked to "automate the email response workflow." They sit down with the AI and describe the workflow: the ticket arrives in Zendesk, a tier-one rep reads it, the rep checks the order in Shopify, the rep copies a macro from a Google Doc, the rep edits the macro, the rep sends the reply. The team then asks the AI to do all of that, in that order, with the same tools, in the same sequence.
What the team did not ask: does the customer need a reply, or does the customer need a refund? Does the answer need to come from a macro, or from the order data directly? Does a human need to be in the loop at all, or only for the 8% of tickets that touch a refund threshold?
Those questions live upstream of the prompt. They are the questions Strip exists to ask. If they are not asked, the AI gets aimed at the existing workflow, and the existing workflow becomes a permanent ceiling on what the AI can achieve.
The AI cannot transcend a workflow you handed it as a requirement. Strip is the only step that pulls the requirement back off.
The three layers you have to strip
In practice, Strip removes three layers in this order. The order matters, because each layer hides the next.
Layer 1: The existing workflow
Most briefs describe the existing process and call it the requirements. They are not the requirements. They are the current implementation of the requirements. Strip the implementation and ask what the actual desired state is.
If the brief says "the AI should read the ticket, check the order, and draft a reply," the desired state is buried. The desired state is "the customer gets the right answer fast." The reading, the checking, and the drafting are implementation steps. Some will survive Strip. Some will not.
Layer 2: The existing tool stack
The tool stack is the most stubborn assumption. Teams describe their AI work in tool names: "we want to use Claude to pull from Notion and post into Slack." The tools are a constraint, not a fact about the job. The job is not "use Claude." The job is "make a decision and put it where a human will see it."
Strip the tool names out of the brief. If the rewrite still works, the tools were never load-bearing. If it stops working, the tools are real constraints, which go in Step 3: Constrain, not in Step 1.
Layer 3: The existing language
The hardest layer to see is the language itself. Every company has internal vocabulary that names the workflow, the roles, and the artifacts. The AI does not share that vocabulary. When the brief says "the CSM kicks off a QBR with the AE on the named accounts," the AI is now solving for vocabulary, not for outcomes.
Translate every piece of internal language into plain English. If the brief makes sense to someone who has never worked at the company, it makes sense to the AI. If it does not, the AI is reading a foreign language and guessing.
A worked example: stripping a CX workflow
Here is a real example, sanitized. A consumer brand asked me to "build an AI for customer support." The original brief, in their language, was this:
"We want an AI agent that lives in Gorgias, reads incoming tickets, looks up the order in Shopify, checks our return policy doc, and drafts a reply that our agents can approve and send. We want it to match our brand voice and follow our SOP."
That brief looks reasonable. It is also fully un-stripped. Every noun is inherited. Every verb is an existing step. The AI gets aimed at a frozen process, not at the job.
Here is the same brief after a one-hour Strip:
"A customer has asked a question or made a request. We need them to get the right answer fast. Some answers are factual lookups against the order. Some are policy decisions against rules we have already written. A small percentage require human judgment because they cross a refund threshold or involve a complaint. The outcome we care about is: the customer is helped, the answer is correct, and the team's time is spent only on the cases that require judgment."
The second brief contains zero tool names. Zero job titles. Zero internal acronyms. What remains is the actual job: route a customer request to the cheapest correct answer, escalate only what needs a human.
From the stripped brief, Step 2 (Decompose) is obvious. From the un-stripped brief, Step 2 is impossible, because the workflow has already been frozen into the requirements. For a deeper look at how stripped briefs translate into shipped systems, see Production AI vs AI Demos.
How to know you have stripped enough
Strip is not a feeling. It has a checklist. You are done when:
- The rewrite contains no proper nouns. No Zendesk, no Klaviyo, no Notion, no Salesforce. If a tool name survives the rewrite, you have not stripped it; you have hidden it.
- The rewrite contains no internal job titles. No CSM, no AE, no SDR, no Tier 1. Roles are how the company organizes the work today. They are not the work.
- The rewrite names an outcome, not a process. The verbs in the rewrite are about what the customer or the business gets, not about what the team does internally.
- The rewrite would make sense to a smart outsider. Hand the rewrite to a friend who does not work at the company. If they understand the job, the strip is clean. If they need translation, it is not.
- You can list the assumptions you removed. Strip leaves a residue. Keep the list of removed assumptions; some of them will become constraints in Step 3.
For most workflows, Strip takes about an hour. For department-level programs, it takes a day. For a company-wide AI transformation, it takes a week of conversations with the people who actually do the work. That is the cheapest week you will spend on the program. It is also the week most companies skip.
Strip is cheap to do and expensive to skip. The team that does it ships in two weeks. The team that skips it is still tuning prompts in month four.
Common stripping mistakes
I have run Strip with a lot of teams. The same four mistakes show up repeatedly, regardless of company size.
1. Confusing constraints with the job. "We have to use our existing CRM" is a constraint, not a fact about the job. It belongs in Step 3, not Step 1. Mixing them produces briefs that look stripped but are not.
2. Stripping in private. The person closest to the workflow is rarely the person who can see the assumptions in it. Strip with at least one outsider in the room. They will hear the inherited language you stopped noticing years ago.
3. Stripping to the point of uselessness. "We need the customer to be happy" is so stripped it has no edges. The right altitude is a paragraph that names the outcome, the inputs that produce it, and the success condition. Lower than that is decomposition, not stripping.
4. Treating the existing tool as the AI's job. "Use AI to make our CRM better" is not a job. It is a feature request against a tool. The job is whatever the CRM is supposed to help the company do. Strip until you find that.
The teams that ship production AI fastest are the ones who treat Strip as a discipline, not a workshop. They write the stripped brief, share it with one outsider, revise it, and move on. They do not spend three weeks on it. They spend one hour, then move to Step 2: Decompose.
Where Strip fits in the larger transformation
Strip is the first step of the V1 Framework, but it is also a posture that runs through the entire AI transformation playbook. Programs that strip assumptions at every phase produce systems that change the operating model. Programs that skip stripping produce automation that locks in the operating model as it existed before AI.
The principle is the same at every altitude. If you describe the existing system to the AI, you get the existing system back, faster. If you describe the outcome, you get a chance at a new system. Strip is what makes the second option possible.
The bottom line
V1 Strip is the cheapest, highest-leverage step in the framework. It is the work of pulling inherited assumptions off a problem so the AI can be aimed at the outcome instead of at a frozen process. An hour of Strip saves a week of prompt tuning. A day of Strip saves a quarter of stalled pilots.
The next step is V1 Step 2: Decompose, where the stripped brief gets broken into its irreducible parts.
FAQ
What is V1 Step 1?
V1 Step 1 is Strip, the first step of the V1 Framework. It is the work of removing the assumptions inherited from how the job was done before AI, so the AI can be aimed at the actual outcome instead of a frozen version of the old process.
What does stripping mean in AI?
Stripping in AI means pulling off the inherited layers around a problem: the existing workflow, the existing tool stack, the existing language, the existing org chart. What remains is the actual job to be done. The AI gets aimed at that, not at the encrusted version of it.
How do I identify hidden AI assumptions?
Read your own brief out loud and underline every noun and verb that names an existing tool, role, or step. Each underline is an inherited assumption. Ask whether it has to be there for the outcome to exist. If not, it is convention, and convention is the most expensive thing you can hand an AI.
Can I skip Strip in the V1 Framework?
No. Skipping Strip is the most common reason AI workflows ship broken. The AI dutifully reproduces the old workflow at higher speed, including the parts that were already broken. Strip is the only step in the framework that prevents AI from accelerating bad processes.
How long should stripping take?
An hour for a single workflow. A day for a department-level program. A week for a company-wide AI transformation. Strip is cheap to do and expensive to skip. The output is a one-page rewrite of the problem in language that names the outcome, not the existing process.
What is a common stripping mistake?
Describing the existing tool stack as if it is a requirement. The AI does not need to know that the team currently uses Zendesk, Klaviyo, and Shopify. It needs to know that a customer asked a question and an answer needs to exist. The tool stack is a constraint that goes in Step 3, not a fact about the job.