You type
The brief, your rules, your tone, and your corrections. Ordinary sentences. Bad grammar is fine, it still works.
How-to — — by Mahmoud Zalt
Customize AI with plain English instead of code or prompt engineering: write a brief, upload files, switch on skills, grant access.
The thing that stops most people is not the software. It is the suspicion that somewhere behind the friendly screen there is a proper way to phrase it, and that other people know the trick and you do not.
There is no trick. The old advice about clever prompts came from a time when you got one text box and one shot, so people learned to cram everything into it. That is not how a customized AI Employee works, and copying those habits actually makes your setup worse.
In Sistava, everything you configure is either a sentence you type, a file you upload, or a switch you flip. The brief is a text box. Your knowledge is a drag-and-drop. Skills and duties are checkboxes with descriptions in English. Tool access is a connect button and a permission. The whole surface is designed for the person who runs the business, not for a developer.
It means every input is something you already know how to produce: a sentence, a document, or a yes or no. Nothing has syntax. Nothing breaks because of a bracket in the wrong place. If you can write an email explaining your business to a supplier, you can do every part of this.
It also means there is no hidden expert mode. Some tools call themselves no-code and then hand you a canvas of boxes and arrows that behaves exactly like programming with a nicer font. That is still coding, just slower. Describing a job in words is a genuinely different activity.
The brief, your rules, your tone, and your corrections. Ordinary sentences. Bad grammar is fine, it still works.
Price lists, past proposals, policies, help articles. Drag the file in. Nothing to convert or format first.
Which skills the role uses, which duties always hold, which apps it may reach. Every option is described in English.
When something is off, you say why in a message. That is the entire feedback mechanism. No settings to hunt for.
Risky actions wait for a click from you. That click is the whole safety system, and it needs no configuration.
Write code, chain prompts, tune parameters, or learn a syntax. If a task starts to feel like that, you are on the wrong path.
Because prompt engineering solves a problem you no longer have. It exists to squeeze context into a single message that vanishes the moment the chat ends. When the context lives permanently in a brief, your documents, and memory, cramming it into every request just adds noise.
The tells are easy to spot in your own writing. Shouting in capitals, stacking on threats about how important this is, repeating the same instruction three ways, or opening with "you are a world-class expert in". None of that helps once the setup is done, and the capital letters help least of all.
Write the way you brief a person instead. Say what the job is, who it is for, what good looks like, and what to do when it is unclear. That is a better instruction than any prompt template, because it is how you already think about work.
Five things, about a page in total. Who your customers are, what you sell and how you charge, how you like to sound, what this particular role owns, and the three or four things it must never do without asking you first. That last part carries more weight than anything else on the page.
Write it in one sitting and accept that it will be imperfect. The gaps announce themselves within the first week of real work, and fixing a gap takes one sentence. People who wait until the brief is perfect never finish it, and a rough brief working today beats a perfect one that stays in your drafts.
Grace is fifty-eight and runs a garden centre in Kent with her son. She uses email, a card reader, and a spreadsheet for stock, and describes herself as "absolutely useless with computers", which is not true but is what she believes. Her problem was the enquiry inbox, roughly forty messages a week about plant care, deliveries, and whether something was in stock.
Her son sat with her for one evening. She dictated the brief out loud while he typed, and it came out as ordinary speech: "We are a family garden centre, we have been here since 1974, most customers are local and over fifty, they want practical advice not sales talk, we do not deliver further than twenty miles, and we never tell anyone a plant will definitely survive because that depends on their soil."
Then they uploaded four things. The delivery policy from her website, a two-page care sheet she hands out with roses, last year's price list, and a folder of about fifty replies she had already sent. That took ten minutes and involved dragging files into a box.
The whole setup was under ninety minutes including tea. Six weeks later roughly thirty of the forty weekly enquiries were answered without her, anything about a refund or a bulk order waited for her approval, and she had corrected it eleven times in total. Her exact words about the corrections were that it was like telling a Saturday assistant something once.
| What Grace did | How she did it | How long |
|---|---|---|
| Explained the business | Said it out loud while her son typed | 25 minutes |
| Gave it the facts | Dragged four files into a box | 10 minutes |
| Chose what it handles | Ticked customer enquiries, left stock alone | 5 minutes |
| Connected the inbox | One connect button, one password | 10 minutes |
| Set what waits for her | Ticked refunds and bulk orders | 5 minutes |
It removes the technical barrier, not the thinking. You still have to decide what good work looks like, and nobody can do that part for you. The setup is easy, the clarity is the work, and that is true whether a developer is involved or not.
It also does not mean anything is possible. A plain-English setup will not build you a custom integration with a system that has no connection available, and it will not invent a rule you never decided on. When the honest answer is that something is not supported, no amount of phrasing changes it.
| Dimension | Traditional | With Sista |
|---|---|---|
| What you write | A dense instruction block, rewritten each time. | A page about your business, written once. |
| Where context lives | In the message, and gone when the chat ends. | In the brief, your files, and memory, permanently. |
| Fixing a mistake | Rephrase and hope the new wording lands. | Say what was wrong once, and it holds. |
| Who can do it | Whoever learned the tricks. | Anyone who can brief a new hire. |
| Handing it over | Your prompts are personal and hard to share. | The setup is readable by anyone on your team. |
No, and the habits it teaches actively hurt you here. Prompt tricks exist to cram context into one disposable message. When your context lives in a permanent brief, your uploaded files, and memory, plain sentences work better than any template. Write the way you would brief a person.
Nothing breaks. A weak brief produces vague work, you notice within a day, and you fix it with one sentence. There is no syntax to get wrong and no way to corrupt the setup by phrasing something clumsily. Rough and finished beats perfect and unwritten.
For the work most businesses need, no. The limits you hit are about connections and judgement, not about how the instructions were written. A developer building the same thing would still have to write down the same rules, they would just write them somewhere less readable.
No. Model choice is handled for you, and knowing the name would not change how you write your brief. What actually changes your results is the quality of what you describe and the documents you upload, not which model is running underneath.
Yes, and that is one of the real advantages. Because everything is written in normal English, anyone in the business can read the brief, understand the rules, and update them. A setup built from personal prompt tricks tends to leave with the person who wrote it.
Stop and start using it anyway. A brief plus one connected app already produces usable work, and the missing pieces become obvious once real tasks start arriving. Trying to finish every layer before the first task is the most common way people stall.
The barrier was never technical. It was the belief that there is a correct way to phrase things and that you do not know it. There is not, and the sentences you would use to explain your business to a new colleague are already the right ones.
If you want to go deeper on one layer, the companion pieces cover getting your written voice right, encoding your rules and approvals, and training on your own documents. Pick whichever is costing you the most rework right now.