Replacement instead of a refund — one repair, not two. The reply is ready to send.
Every agent is good at something else
An agent takes a whole area — sales, service, people — and carries dozens of skills, your context and a full set of tools. You hand over the task and it reaches for whatever it needs.
Keeps an eye on the whole funnel: prepares quotes, notes after meetings and deal summaries.
You hand over one sentence.You come back to finished work.
Not every task fits into one answer. The agent goes into the background, works through source after source and comes back only with a finished result — while you get on with your own work.
Collect the terms from every supplier quote this quarter and put them into one comparison.
Three quotes stand out from the rest — one 18% cheaper, but with a lead time three weeks longer.
Your standard, not generic best practice.
A skill is three to five sentences written down once: where we start, in what order, what we never promise. One agent carries dozens of them and reaches for the one that fits the task.
One skill applies to every agent that gets it — including the one answering from Slack.
Hello,
we are replacing your unit with a new one — the courier will collect the damaged one and leave the new one on the same run.
A refund comes into play after two failed repairs; here there was one repair, which is why we propose a replacement.
We have the collection planned for Tuesday, 9 September — the earliest date confirmed by the courier in the system. Compensation for the downtime will be decided by your account owner; I have passed the case on to them.
Does Tuesday work for the collection?
Best regards, Marta
Customer service · Polmar Systemy
One skill applies to every agent that gets it — including the one answering from Slack.
AI rollout quote for Kamex Ltd
1 · Situation
Today each of your four teams works with AI its own way, and the results end up in private notes. This quote is about bringing that into one place.
2 · Scope
A 22% discount subject to board approval — provisionally PLN 12,600–13,100.
3 · Terms
Quote valid 14 days, until 19 September 2026. Payment 14 days from invoice, the rollout starts in the week after acceptance.
Next step: a technical call on Thursday 11 September at 11:00.
One skill applies to every agent that gets it — including the one answering from Slack.
We close the quarter if Zenit signs by 20 September — without them we are 180k short.
Risk: Kamex has been waiting for budget approval on their side for 12 days — owner: Marek.
It has a browser of its own.
The agent clicks in its own browser, on appto’s side — not on your computer and not in your tab. It searches, opens an item, copies it into a sheet, moves on. So it also works where there is no integration at all.
It starts as yours.It ends up the company’s.
An agent starts out private. Once it proves itself, one switch moves it to the team or the whole company and it becomes standard equipment — with nothing rewritten from scratch.
The scope is set by someone with the permission to set it — not by everyone who uses the agent. That is why a team agent does not change by accident, and does not reach the whole company before somebody decides it should.
@appto what is happening with the Zenit complaint?
Replacement, collection on Tuesday. The draft reply is waiting in the thread.
✓1Send me a summary of the cases for tomorrow.
Seven cases, two waiting for the account owner’s decision.
Set up your first agent
You decidewhat the agent has.
Which model it runs on, which tools it reaches for and who has it at all — three settings, not three promises. The agent does not go beyond what you leave it.
Claude Sonnet 5Anthropic · long instructions
Assigned permanently. Nothing switches by itself in the background.
Four out of close to a thousandGmail, HubSpot, Jira, Drive
The agent reaches only for the ones you leave it.
Team: Customer service11 of 48 people in the company
The scope is changed by someone with the permission, not by every user.
Say who you need.appto sets up the rest.
You do not program an agent or click through a form with thirty fields. You write one sentence and appto picks the role, the skills, the company knowledge and the tools — you only correct what should be different.
You fix it once.It works for everyone.
An agent and its skills have versions. A correction saved in one place applies from the next task for every person on the team — no “from today we write differently” email.
We give an approximate date, usually up to 30 days.
Quote only the dates confirmed in the system.
Your first agent by the end of the day
Take a task somebody on the team does the same way every week. One sentence is enough to create an agent — the rest you see after the first result.
Before you build the first one
The usual questions about agents: what they see, how long they work and what they do without asking. Want to see it on your own process? We will build the first one with you.
Does an agent only handle one task?
No. An agent takes a whole area — sales, customer service, people — and carries dozens of skills it reaches for depending on what you hand it. A sales agent will prepare a quote, write a follow-up and summarize a meeting, because all of that belongs to its area. Even a narrowly named agent, say “Quote builder”, does not do one thing: it prices the deal, assembles the document, keeps the bands and reminds you about the validity date.
How is an agent different from an ordinary AI conversation?
A conversation is one-off — every time you explain the background from scratch. An agent is a setting that stays: it knows its role, which part of the company’s knowledge it uses, which tools it reaches for and which writing standard it follows. That is why you hand it a task in one sentence rather than in half a page of instructions.
How long can an agent work on one task?
A task does not have to fit into one answer. An agent works in the background for tens of minutes as well — going through source after source, checking and correcting. The thread is free the moment you hand the task over, so you can hand over more things or close the laptop. The result waits in appto, and if you set it up that way — in Slack or Microsoft Teams.
Will an agent send something to a customer without us knowing?
Not unless you set it up that way. For every agent you decide what it does on its own and what needs confirmation. By default anything going outside — an email to a customer, a change in the CRM — waits as a draft for approval.
What does an agent see in our data?
Exactly as much as the person using it. An agent has no permissions of its own, so the same agent answers two people from different sources — each based on what that person can access. Changing someone’s permissions immediately changes the reach of their agents.
What if our system has no integration?
The agent opens it in a browser and operates it like a person: signs in, searches, clicks, copies data across. An old supplier portal or an in-house app with no API therefore does not block the work. Beyond that there are close to a thousand ready integrations — the browser is the emergency exit, not the rule.
How many agents should we build to start?
None — to start you have Wise, the most versatile agent in appto: it knows the context of the whole company, reaches for every connected tool and comfortably covers the first weeks. You build your own once one area starts repeating: one of them. Ten agents in one evening is the shortest path to nobody using any of them, because nobody remembers which is for what. The next one makes sense when the previous one entered circulation without reminders.
Does an agent work outside appto?
Yes. The same agent answers in appto, in a Slack channel and in Microsoft Teams — it reads the thread and acts in it on the same rules. You do not have to open another tab to hand it something.
