Recipe Manager: Step-by-Step Procedures Your Agents Follow
Step-by-step procedures for your agents, tested before they reach you.
The Recipe Manager is the designer of AI agent procedures in the Ai1 platform by MyZone AI: it turns a task into tested step-by-step recipes your agents follow, finds why a recipe goes wrong, re-tests every fix and keeps a version record of each change.
Clear, tested procedures mean fewer mistakes and less rework, and your agents handle the same task the same way each time.
Working procedures, not drafts
You get back a recipe that has already been run and checked, rather than a first attempt you have to debug yourself.
A clear record of every change
Each version is kept with what changed, when and why, so you can always see what a recipe used to do and roll it back.
Know what to improve next
Audits of your existing recipes feed a ranked list of the ones most worth building or fixing next.
Meet your agent
Get to know Somchai
Somchai builds, tests, audits and improves the step-by-step procedures your agents follow. He is patient, methodical and believes a good procedure is like a good recipe.
AI agent. Fictional persona; not a real person.
Somchai, the persona of the Recipe Manager
Name
Somchai W.
Role
Procedures Specialist
Experience
18 years
Work history
Operations documentation lead at a hotel group Process analyst at a manufacturing company
Personality
Patient, methodical
Based in
Chiang Mai, Thailand
Why Somchai has a personality
AI does better work as a specific expert, so the Recipe Manager works as Somchai (Procedures Specialist, Chiang Mai, Thailand). You can rename Somchai or change the personality, experience and profile settings at any time.
What a recipe should achieve and the rules it must follow
Each recipe
Ordered steps, the tools each step needs and plain instructions
Testing
Run end to end in test mode until it passes consistently
When it goes wrong
Traces each wrong result to its step, fixes it and re-tests the failed cases
Change record
Every version kept with what changed and why, so you can roll back
What it never does
Hand over a recipe that has not passed its tests
Capabilities
What the Recipe Manager does for you
Builds new recipes from your requirements
Turns what you need an agent to do into an ordered set of steps, works out which tools each step needs and writes plain instructions for every one.
Finds why a recipe goes wrong
Reads the recipe's run history, traces each wrong result to the step that caused it and fixes what belongs in the recipe itself.
Tests before it ships
Runs the whole recipe against test cases, including the ones that used to fail, and keeps iterating until it passes consistently.
Versions every change
Records each update with the reason behind it, so you can compare versions and step back to an earlier one if you need to.
Hands tool problems to the right specialist
When the fault is in a tool rather than the recipe, it passes the issue on with full context, then checks the fix before closing the job.
Audits recipes and ranks what comes next
Reviews the recipes you already have and keeps a ranked roadmap of the ones worth building or improving.
How the Recipe Manager differs from a written checklist
A checklist tells an agent what to do but does not prove the steps work. The Recipe Manager agrees the rules with you, breaks the task into steps with the tools each one needs and runs the whole recipe in test mode before hand-over. When a result goes wrong, it traces the cause, re-tests against the cases that failed and saves a new version, and it passes tool faults to the Skill Manager.
How this differs from the SOP Agent: the SOP Agent writes procedures for the people on your team, while the Recipe Manager writes and tests the step-by-step procedures your AI agents follow.
How it works
From your request to a finished result
It runs inside your Ai1 system and works in the tools you already use. You ask, it works, it reports, and it stops for your OK where it matters.
Tap the diagram to enlarge it
You approve
Step 1: Agree what the recipe should do
You describe the task and confirm the rules it must follow. If anything is unclear, it asks before it goes further.
Step 2: Break it into steps
It maps each action in order, picks the tools each step needs and writes clear instructions.
Step 3: Test end to end
It runs the full recipe in test mode and checks every step gives the expected result.
Step 4: Fix and re-test
Failing steps are fixed and tested again. Problems in the underlying tools go to the Skill Manager, and the fix is checked before work resumes.
Step 5: Ship with a version record
The passing recipe is saved with its version history, so every change can be traced.
When to use it
Hand these situations to the Recipe Manager
Illustrative photo. Ask for a task to be done the same way every time.
If this is you…
…here's what it does
You want an agent to handle a repeatable task the same way every time
It designs the procedure, such as qualifying and filing a new lead, and hands it over already tested.
A recipe keeps producing wrong results
It finds the exact step that fails, fixes it, and re-runs the cases that were going wrong.
A recipe works most of the time but not always
It tightens vague instructions and re-tests against the cases that slipped, such as a report that sometimes skips its summary.
You are not sure which procedures to improve first
It audits what you have and gives you a ranked list of what to build or fix next.
What you get
New recipes with ordered steps, tools and clear instructions
A failure analysis tracing each wrong result to its cause
Before and after test results on the same cases
A version record showing what changed, when and why
A ranked roadmap of recipes to build or improve next
What it won't do
✕Build or fix the tools and skills that agents useHandled by: The Skill Manager
✕Run automated or browser testing of apps and websitesHandled by: The Tester
✕Decide what a recipe should achieveHandled by: You, the person asking for it
Example tested recipe
Example: returns recipe fix and re-test
Example with a fictional company. Names, people and figures are invented to show the agent's output. Any resemblance to a real company or person is unintended. Industry facts are real and cited with their sources.
What it was asked: Our returns recipe keeps going wrong: some late returns get a 'yes, send it back' draft and damaged parcels never get asked for photos, so find out why, fix it and make it reliable.
Example report
The request, and why the recipe kept getting it wrongThe recipe, version 2.0: seven steps, three changedThe re-test: same cases, old recipe versus newVersion record, who did what, and what to build next
What it found: 3 findings, with the numbers
Nine of the 19 wrong results came from one gap: when the order number was only in the email subject, the recipe skipped the return-window check and assumed the order was in time.
Six damaged-in-transit requests were filed as change-of-mind returns because the reason list was vague, so version 2.0 separates damage from faults and asks for photos before anything else.
The other 4 failures were in the order lookup tool, not the recipe, so Skill Manager fixed the tool and version 2.0 then passed all 24 test cases, three runs each.
The Recipe Manager read 30 days of the recipe's run history, traced each wrong result to the step that caused it, rewrote the steps after the operations manager approved the rule changes, and re-ran 24 test cases three times each in test mode, sending nothing to customers. The tool fault went to Skill Manager and its fix was re-tested before closing; reply tone was not scored, and a person still approves every refund. Run date: .
Covers automated and browser testing of your apps and sites, which sits outside recipe testing.
Common questions
Frequently asked questions about the Recipe Manager
What are standard operating procedures for AI agents?
Written, ordered steps an agent follows to finish a task the same way each time. In Ai1 by MyZone AI they are called recipes: the Recipe Manager designs each one, picks the tools every step needs and tests it end to end before your agents use it.
A recipe works like a script, except it is tested and improved over time. Recipes are written for AI agents; procedures for the people on your team are the SOP Agent's job.
Why does an AI agent workflow keep giving wrong results?
Usually one step has a vague instruction, or a tool it relies on is at fault. The Recipe Manager reads the run history, traces each wrong result to the step that caused it and fixes the recipe. If the fault is in a tool, the Skill Manager fixes it and the recipe is re-tested before the job is closed.
How do you test a step-by-step workflow for an AI agent?
Run the whole workflow in test mode, check each step gives the expected result, then re-run the cases that used to fail after every fix. The Recipe Manager does this before hand-over, so what you receive already works.
Every version is kept with a note of what changed and why, so you can compare versions or roll back.
About Ai1
Do I get just the Recipe Manager, or the whole platform?
The whole platform. The Recipe Manager is included on every Ai1 level, including Developer Core, with no per-agent charge. It works alongside the other Ai1 agents on your account. Compare plans →
Two paths, one platform. Build it yourself on a developer plan, or let our team run your AI operations for you. Every plan runs on its own private server, and every price is shown in full. See Ai1 pricing →
Product & Engineering
Product & Engineering
Product & Engineering
Product & Engineering
Product & Engineering