AI agents for business
Almost everything sold as an AI agent is a thing you build. You describe the steps, connect the tools, test it, and own it afterwards. That is a real option, and for some teams it is the right one. It is worth knowing which kind you are buying before you start, because the work of keeping it running lands on somebody either way.
Concretely
- Qualifies an inbound lead, asks the qualifying questions, scores it and routes it to the right person.
- Handles customer questions, resolves the common ones, and escalates the problems that need a person.
- Cleans and merges CRM records, chases the missing next steps and close dates, and flags deals that have stalled.
- Books and reschedules appointments, sends the reminders and handles the cancellations.
- Processes email, updates the spreadsheets, and puts the weekly report together.
What an AI agent actually is
An AI agent is software that takes an instruction, decides the steps itself, and uses your tools to carry them out. That last part is the whole distinction. A chatbot answers in the chat window; an agent goes and does something in the systems you already run, then comes back and tells you what it did. Everything else in this category, the builders and the hires alike, is a different answer to the same question: who decides the steps, and who is responsible when they are wrong.
Most of the category is a builder, not a hire
The market splits cleanly in two, and the words on the homepages do not tell you which half you are looking at. A builder gives you a canvas: you design the agent, wire the triggers, define the handoffs, test it, and maintain it as your tools change. A hire arrives with the jobs already described, and you tell it which ones it is taking. Builders offer more control and cost you an owner. Hires offer less control and cost you less attention. Neither is the better product in the abstract, and the honest question is whether anyone on your team wants to be the agent builder.
The questions worth asking before you buy
Five, in the order that matters. Whose credentials does it use, and can one person see another person's data through it? What can it do without asking, and who decides that? What happens in the first week, before you trust it? Where does the work happen, in a tool you have to open or in the place your team already talks? And how does the bill behave when the whole team starts using it? Most buyer regret in this category traces back to the second and the fifth.
What it costs to run, not just to buy
Agents are priced three ways: per seat, per plan, or for the work done. Per seat punishes you for giving it to more people, which is the opposite of what you want from something that works across a business. A plan is predictable until you outgrow it. Paying for the work is honest but only if the rate is legible. Whichever shape you pick, the real cost is the one that never appears on the pricing page: the hours someone spends building, fixing and re-testing the thing after the tools it depends on change.
Where Sanaf fits, and where it does not
Sanaf is the hire, not the builder. One AI employee for the workspace, hired into jobs that are already described, working in Slack and Microsoft Teams across the apps you already run, with rules you set per kind of action and a shadow period where it drafts and you approve before it does anything on its own. If you want to design a multi-agent system yourself, or you need control over every individual step, a builder is the better buy and we will say so on the comparison pages.
The jobs this comes from
Sanaf holds thirty five written job descriptions. These are the ones behind this page.