Skip to content

Start here / Step 3 of 7

Start here ยท step 3 of 7

What AI can do on its own

Many people expect AI to simply do the work. Some of it can now run on its own, from reading a contract to drafting the reply, and knowing where it stops is what makes that safe.

Start here

  1. 1What an AI chat tool is
  2. 2The tools and their features
  3. 3What AI can do on its own
  4. 4The words you will hear
  5. 5Your first prompt
  6. 6What is safe to paste
  7. 7How to check an answer

What an agent is

An agent is an AI setup that takes several steps toward a goal without you typing each one. Given a contract and your positions, an agent can read the contract, compare each clause, rate each difference, draft reply wording and assemble the result, from one instruction.

In plain terms

A chat answers one message at a time. An agent works through a sequence of steps, and a well-built one pauses at set points for you to decide.

Three levels of automation

Level 1

Saved instructions

Your instructions and positions are saved once in a project or custom instructions. Every new document gets the same treatment from one short message.

Level 2

Plugins and skills

A packaged set of instructions added to the tool, such as a legal plugin loaded with your playbook. It applies your positions every time without being asked.

Level 3

Agents

A multi-step job run from one instruction, with review points where it stops and waits for your decision before continuing.

What can run on its own

  • Reading a document against a standard and listing every difference, with the words it relies on quoted.
  • Filling a template from answers you have already collected, and asking about anything missing.
  • Turning notes and updates into a report with each figure tied to its source.
  • Sorting incoming requests by urgency using rules you set.

Where you still decide

  • Whether a difference matters for this deal, and what to accept.
  • Anything that goes out under your name or your organization's.
  • Anything that depends on facts the tool cannot see.

This is called keeping a human in the loop: a person reviews and approves at set points before anything is relied on or sent. Every ProvisoIQ product says what it runs on its own and where you step in, at the top of its page. The agent kits run the furthest on their own.

When the model changes

AI providers release new models every few months, and a setup that worked on the old model can behave differently on the new one. ProvisoIQ retests every playbook, template and kit when a supported model releases a major version and emails buyers the updated version, so you do not need to track releases or rewrite your setup yourself.

PreviousThe tools and their featuresNextThe words you will hear