The problem with how teams adopt AI
Everyone is automating the wrong things.
A sales team uses Claude to draft emails. A marketing team uses it to summarize meeting notes. An onboarding team uses it to format checklists. Each of these is a single task inside a larger process that nobody has looked at as a whole.
The result is what I call tool-assisted busy work. You are doing the same process you always did, just slightly faster in a few spots. Three months later, leadership asks what changed, and the honest answer is: not much.
The issue is not the tools. The issue is that nobody mapped the process first. Nobody asked which steps are actually automatable, which need a human in the loop, and what the dependencies between them are.
That is what this guide helps you do — for one process, in about ten minutes.
What you will get
By the end of this, you will have:
- A step-by-step map of one business process as it actually works today
- An automation assessment — which steps an AI agent can handle, which need human review, and which should stay manual
- A working Skills file — a markdown artifact that encodes the easiest-win automatable step as an instruction set an AI agent can follow
- A gap analysis — what guardrails, approvals, and data boundaries you would need before turning any of this on
This follows the first three pillars of the FORGE methodology: Baseline (mapping what exists), Skills (encoding the work), and Agents (deciding what runs it). We stop short of Guardrails and Schedule — those require looking across multiple processes and understanding how your team's data flows between them.
Before you start
Disclaimer: This content is provided for educational and informational purposes by VibeSec Advisory. The prompt below interacts with Claude Code on your local machine. It does not access external systems, modify files, or execute code unless you explicitly approve. Always review generated output before using it in your work. VibeSec Advisory provides recommendations and frameworks — not warranties on specific outcomes. Use at your own risk. By using these materials, you acknowledge that VibeSec Advisory is not responsible for any damages, data loss, or system issues resulting from their use.
You will need:
- Claude Code (terminal, desktop app, or IDE extension)
- About 10 minutes
- One business process in mind — pick something you or your team does regularly that involves multiple steps and at least one handoff
Good examples: client onboarding, content review and publishing, sales lead qualification, new hire setup, incident response, invoice processing, customer support escalation.
Ready to apply the FORGE framework?
VibeSec helps knowledge worker teams redesign their processes using the FORGE framework: Skills, Agents, Guardrails, and Schedule. Security is built in, not bolted on. Book a FORGE Discovery Workshop to get started.
The prompt
Copy this entire block and paste it into Claude Code. It will walk you through the process interactively.
I want you to act as a FORGE process mapping consultant. You are going to help me decompose one business process into a structured map that identifies which steps can be handled by AI agents and which need human involvement.
Walk me through this step by step. Ask me questions one at a time — do not try to do everything at once. Be conversational but structured.
PHASE 1 — BASELINE (Understanding what exists today)
Start by asking me:
1. What is the process? (e.g., "How we onboard a new client")
2. Who is involved? (roles, not names)
3. Walk me through it step by step — ask me to describe each step, and for each one, ask:
- What triggers this step?
- Who does it?
- What tools do they use?
- How long does it typically take?
- What can go wrong here?
- What happens next?
Keep asking "what happens next?" until we reach the end of the process. Number each step.
After we finish mapping, summarize the full process as a numbered list with this format for each step:
- Step [N]: [Description]
- Owner: [Role]
- Tools: [What they use]
- Time: [Estimate]
- Risk: [What can go wrong]
- Trigger: [What kicks this off]
PHASE 2 — AUTOMATION ASSESSMENT
For each step, assess and categorize it as one of:
- AUTOMATE: AI agent can handle this end-to-end with no human review needed
- ASSIST: AI agent does the work, human reviews before it moves forward
- MANUAL: Must stay human — judgment calls, relationship moments, legal/compliance sign-offs
For each assessment, explain your reasoning in one sentence. Be conservative — when in doubt, mark it ASSIST rather than AUTOMATE. The goal is not to automate everything. The goal is to automate the right things safely.
After the assessment, show me:
- A summary table of all steps with their category
- Total estimated time saved per cycle if AUTOMATE and ASSIST steps are implemented
- The single highest-impact step to automate first (the "quick win")
PHASE 3 — SKILLS FILE
For the quick win step identified above, generate a working Skills file in this format:
```markdown
---
name: [Skill name — verb + noun, e.g., "Draft Client Welcome Email"]
description: [One line — what this skill does]
trigger: [What kicks it off]
owner: [AI agent or human role]
review: [Who reviews the output before it moves forward, if anyone]
tools: [What tools/systems are needed]
---
## Context
[2-3 sentences: why this step exists, what it accomplishes in the larger process]
## Instructions
[Step-by-step instructions written for an AI agent to follow. Be specific about inputs, outputs, format, and constraints. Include what NOT to do.]
## Output
[What the skill produces — format, where it goes, who receives it]
## Guardrails
[What this skill should NOT do. Data it should not access. Decisions it should not make. When to escalate to a human.]
PHASE 4 — GAP ANALYSIS
Finally, give me a brief gap analysis:
- What data does this process touch that would need access controls?
- What approvals or sign-offs exist today that must be preserved?
- What happens if the AI agent makes a mistake at any step? Is it recoverable?
- What would you need to verify before putting any of this into production?
Present this as a bullet list under the heading "Before You Go Live."
IMPORTANT GUIDELINES:
- Ask questions one at a time. Do not dump all questions at once.
- Use plain language. No jargon unless I use it first.
- Be honest about what AI cannot do well. Not every step should be automated.
- If I describe a step that involves sensitive data (PII, financials, credentials), flag it and note that it needs guardrails.
- The Skills file should be practical and specific enough that someone could actually hand it to an AI agent and get a useful result.
## What to expect
Claude will start by asking you to describe your process. Just answer naturally — you do not need to use any special format. Talk about it the way you would explain it to a new team member.
The conversation usually takes 8-12 turns depending on how many steps your process has. A typical process has 6-12 steps.
At the end, you will have three artifacts:
- The process map (your Baseline)
- The automation assessment with a recommended quick win
- A working Skills file for that quick win
## What this does not give you
This maps one process in isolation. That is useful, but it is not the full picture.
In practice, business processes do not exist alone. Your client onboarding process touches your sales handoff process, your billing setup, your project management tooling, and probably three Slack channels. The Skills file you generate here might overlap with a Skill someone else on your team would write for a different workflow.
The hard problems in AI process redesign are not about individual steps. They are about:
- **Dependencies between processes.** If you automate step 4 of onboarding but it feeds into a manual step in billing, you have created a bottleneck, not a solution.
- **Shared Skills.** The same "send a formatted email" Skill gets reused across onboarding, support, and sales. Building it three times is waste. Building it once and sharing it is a library.
- **Guardrails that cross boundaries.** Your onboarding agent should not have access to your billing system, but it needs to trigger a billing workflow. Those access policies need to be designed across the whole system, not per-process.
- **Schedule and sequencing.** Which processes run autonomously? Which run on a trigger? Which need human approval before the next step fires? That is an orchestration problem that requires seeing all the processes at once.
This prompt gives you one piece. The full picture requires mapping the system.
## Try it on a process you know well
Pick something you have done a hundred times. Client onboarding. Content review. Lead qualification. The more familiar you are with the process, the better the map will be — because you will catch the steps that are not documented anywhere but exist in practice.
If the result surprises you — if you discover steps that are pure busy work, or handoffs that exist for no reason, or entire chunks that an agent could handle while you focus on the parts that actually need your judgment — then you are seeing what a systematic approach to AI adoption looks like.
One process is a proof of concept. Across your whole team, it is a transformation.
---
*VibeSec Advisory helps teams redesign their business processes for the age of AI agents using the FORGE methodology. If you want to map your entire operation — not just one workflow — [book a Discovery Workshop](/pricing) and we will do it together.*