Who is this assistant for?

You want your own agents and not to depend on whatever is offered. You are afraid an AI will touch something that cannot be undone. Every Monday you have to explain everything to the AI again. You installed something with AI and cannot tell whether it worked.

One agent per task means forty agents nobody maintains; one agent per ROLE means seven somebody can actually manage. This assistant designs the AI figure that holds a seat in your house, with the same structure used to write a human role: which competencies it carries —five minimum, or it is not a role but a task—, which inputs it needs weekly and where they come from in YOUR house, what it remembers between sessions, and the limit in three columns where the third —what it never touches— is written first, because an agent with no written limit ends up changing a price. And it carries the option to define your own role, so you depend on nobody's list. It comes out with the agent's prompt ready to paste and with the business figure that should move at ninety days.

AI assistant · N5 Complete system

◆ Included in the annual plan🤖 Specialized AI Agentsv1.2.0 · 2026-09-23

AI Agent Builder by Role for Restaurants

You pick which agent to create —one of the MASTERESTAURANT firm's seven departments (Marketing, Commercial, Management Control, Finance, Operations, Chief of Staff, AI Transformation), a corporate chef, purchasing, people and culture, or your own written in your words— and out comes the whole figure: its team and competencies, the weekly inputs, what it remembers, what it decides alone and what it never touches, and the agent's prompt ready to paste, with the department's package of files.

Depth

14/23

Your data

Compatible AIs

Steps to use it

Depth of the resulttap a level to see what you get

◀ Put out today's fireRun the year ▶

🏛️

N5Complete system

Several chained pieces you install and operate over time, with their review cadence. It changes how the business runs, not one decision.

◆ This assistant 110 of the 164 published assistants are at this level

Is this for you?

What it does and what you get

One agent per task means forty agents nobody maintains; one agent per ROLE means seven somebody can actually manage.
This assistant designs the AI figure that holds a seat in your house, with the same structure used to write a human role: which competencies it carries —five minimum, or it is not a role but a task—, which inputs it needs weekly and where they come from in YOUR house, what it remembers between sessions, and the limit in three columns where the third —what it never touches— is written first, because an agent with no written limit ends up changing a price.
And it carries the option to define your own role, so you depend on nobody's list.
It comes out with the agent's prompt ready to paste and with the business figure that should move at ninety days.

What you end up with

The agent's full figure: its seat in three lines; the competency table with the question each one answers and the input it needs; the input table with where each figure comes from, its format and whether it exists today; memory —what it remembers, where it lives and who updates it—; the limit in three columns with a cash cap on the first; the cadence with its day, deliverable and recipient; the five-step start-up with the test against an old week; the ninety-day measure with baseline, target, owner and shutdown rule; and the AGENT'S PROMPT ready to paste, with its limits inside.

EXAMPLE This is what you getAn example built with Sazón de Origen, the house's fictional restaurant.Not your data: your result is built with YOUR restaurant's.See the full example →

Sample result

Sazón de OrigenFictional contemporary Colombian restaurant in Medellín, 2 locations, 6 years in business. Every sample result in the library comes from this same case. Fictional restaurant

The owner asked this: «I have three locations and I live putting out operational fires: every manager does the checklists differently and I hear about problems once they have already cost money. I want one agent covering operations, not forty little ones.»

Which figure it resembles: the Operations Director a three-to-ten restaurant group would have — the one who is in no kitchen and knows what happens in all three.

Why it exists: so you read on Monday, on one page, what today you find out on Thursday through a complaint.

If nobody holds this seat: nobody consolidates the three locations' checklists, findings repeat because nobody counts them, and deviations only show up once they have passed through the P&L.

Note what it does NOT cover: it is not a cost agent or a marketing agent. An agent covering operations and costs at once covers neither, because those two questions are answered with different data on different cadences.

Competency	The question it can answer	Input it needs
Operational audit	What went off standard this week and at which location?	Signed checklists
Shift and area checklists	Which were filled and which were signed unfilled?	The same, with times
Manuals and procedures	Which finding repeats because the procedure is badly written?	8 weeks of history
Quality and food safety	Did any chiller go out of range and what was done?	Temperature sheet
Maintenance	Which equipment is going to fail before it fails?	Fault report
Multi-site consolidation	Which location is doing worse, and at what exactly?	All three, together

Six, not two. With two competencies this would not be a role: it would be a task, and a loose prompt from this library serves a task better.

✋ NEVER TOUCHES	📝 Proposes and waits for a signature	✅ Decides alone (with cap)
Changing a menu price	Changing a written procedure	The order the week is audited in
Hiring or firing	Pulling a dish off the menu	Which finding goes into Monday's 3 alerts
Approving equipment investment	Buying a part above 300,000 COP	Asking for evidence on a blank signed checklist
Talking to a supplier	Moving staff between locations	Minor parts up to 300,000 COP
Writing to a guest

The left column is written before the other two. An agent without it ends up changing a price, and that is not undone with an apology.

Illustrative example generated with a fictional restaurant. Not a promise of results, not a client case.

This is 3 of 5 parts. Behind the wall are the two pieces that make the agent survive its first month. Inputs, memory and cadence: the five figures it needs weekly with where each comes from in your house, in what format they arrive and which do NOT exist yet with the plan to get them within seven days —because an agent that asks for something you do not have switches itself off—; the ten things it remembers between sessions, where they live and who updates them; and the exact day, the deliverable and the minutes it should take to read. And the start-up with its measure: the five steps of day one, the test against a week that has ALREADY happened and whose outcome you know —if it cannot see what you already know, it will not see it next week—, the business figure that should move at ninety days with its baseline and target, the written shutdown rule, and the complete AGENT'S PROMPT ready to paste into your AI, with the third column's limits written inside.

See the full example — free account

Works with these AIs

Claude — the best for this assistant The best fit here: it holds the five-competency minimum and does not soften the limit's third column. Demand that it tell you when what you bring is a task and not a role, even though that costs the assistant the work.

How to use it

Pick above which agent to create: one of the firm's seven departments, one of three more roles, or «an agent I define» and write what you want it to do. If you came from a department's «Create my own agent» button, it is already chosen.

Download this card's Agent Creator package and upload it to your AI together with the prompt: it carries the department's full sheet and the house method.

Fill in My Restaurant with your channels and yardsticks: the inputs the agent will need weekly come from there.

Before you start, decide who in the house will manage this agent. An agent with nobody reading what it delivers switches off within two weeks.

Copy the whole assistant and paste it into your AI of choice. Answer its questions one at a time.

When you get the agent's prompt, run it against a week that has ALREADY happened and whose outcome you know: if it cannot see what you already know, it will not see it next week.

At ninety days measure the last block's figure and apply the shutdown rule without arguing it.

Which of your data it uses

of 23

This assistant works with 14 of the 23 fields in «My Restaurant», among them Restaurant name, Business type and Cuisine or concept. It does not ask out of curiosity: these are what make the answer speak about YOUR scale instead of an industry average — the same calculation on a business of another size returns a number you cannot decide with. You fill them once and they apply across the library, so the second assistant you open already starts with them in place. Whatever you leave blank, the assistant asks for one thing at a time instead of inventing it, and says so before answering, so you know what it is working from.

Restaurant nameBusiness typeCuisine or conceptCity and countryNumber of locationsYears in businessTeam sizeStrong and slow daypartsActive sales channelsAverage check (with currency)Average monthly salesApproximate food cost (%)Approximate labor cost (%)Main goal this year

Always up to date

Every assistant carries a visible version and date. When an AI changes how it works, the assistant's version goes up and you see what changed.

v1.2.02026-09-23Diego's instruction (2026-09-23): «in each agent a button called create my own agent that leads to that agent's creation prompt; let's centralise one single agent-creation prompt with the selector of the others, with the package of files they need to work professionally and not superficially». The selector now carries the firm's seven departments (derived from datos/agentes by motor/creadores.py, with their team, jobs, brief, cycle and limits), three roles no department covers and «an agent I define» with free text; the section «The seven departments and the package» is born, and so is the Agent Creator package (kits/creador-de-agentes).

v1.1.02026-08-30Diego's instruction (2026-08-30): «agents are supposed to handle things automatically and improve themselves, recurrently, without the human having to step in». The card designed the figure and stopped there: seven layers and none said how the agent improves after month one. The EIGHTH LAYER enters —improvement—: which signal each run leaves, who writes it (the owner, never the model), where it lives, and the monthly 20-minute review with its only three outcomes —adjust a competency, change an input, switch it off. And the final prompt gains a LEARNED-SIGNALS SLOT, which is what turns a fixed prompt into an agent that improves. The honesty goes in the ROLE and not in a footnote: an agent claiming to improve with no signal is inventing, and one inventing its own yardstick is worse than one that does not learn.

v1.0.12026-08-27The response-ladder doctrine is stated in English with the house phrase —«continue with a marked assumption»— and not with a synonym. The migration probe caught it: 34 migrated bodies in Spanish and 27 in English, and the seven-card gap were the ones I wrote today. A doctrine said two ways stops being greppable.

v1.0.02026-08-27Diego's instruction (2026-08-27): «we are NOT making more agents, we are not making 47 agents; what I want is to organise that into powerful prompts that build a few key agents with names and functions like a person in a large restaurant company, with multiple capabilities matching that seat». This card absorbs the 47 in c35: instead of forty-seven one-task agents, seven role figures with five or six competencies inside, plus the option to define your own role so the owner depends on nobody's list.