Short answer: An outcome-based agent is judged on a finished result, not on how busy it looked. You name the result, the limit of what it may do, and the person who owns the result. Activity, tokens and "tasks completed" are not the outcome.
I'm Vinay Pasricha. I run GoodSpace AI in Noida, and I wrote AI for Business Leaders. The phrase is now a sales line. It is worth keeping, but only if you can point at the result. Otherwise it is a new name for the same retainer.
A result, not a motion
AI, in the way I use the word in the book, does five kinds of work. An agent is the kind that takes a step, not only a draft. An outcome-based arrangement says: we will look at the step's result. A reconciled invoice. A quote that went out and matched the price list. A shortlist prepared for a person, not a hire made in the dark.
The result has to be checkable by someone who was not the vendor. If only the vendor can say whether it worked, you do not have an outcome. You have a claim. Write the result so a colleague could mark it pass or fail on a Friday without a meeting.
This is also how you should hear a pricing proposal that says "pay for outcomes". Ask which outcome, checked by whom, with what happens when the outcome is wrong. A bonus tied to a vague success is still a vague contract. A fee tied to a pass or fail you both can see is a different thing, and it is the one worth discussing.
What the person still owns
- The name of the result: one sentence, with a number you already understand.
- The boundary: actions the agent may take alone, actions it may only prepare, actions it may not take.
- The signature: the person who owns the result if the agent is switched off tomorrow.
- The record: what the agent read and what it did, kept where you can look without asking the vendor.
Automate the reversible work inside that boundary. Assist where a judgement is needed: the agent prepares, the owner decides. Escalate anything that would hurt a customer, a colleague or the company's word. An "outcome" that skips this sort is just a target stuck on top of an unbounded tool.
The owner is the point of the whole arrangement. In a larger argument I make elsewhere, one person holds one result and uses agents as the means. The agent does not become the owner because the invoice says "outcome". If you cannot name the person, do not sign the contract. Name the person first, even if the person is you.
How to test the phrase
Ask the seller to finish three sentences. The result is …. We will know it happened because …. If it does not happen, the consequence is …. If any sentence needs a demo to make sense, it is not an outcome yet. It is a feature.
Then run it on a small batch you can check yourself. Ten invoices. Ten quotes. Ten drafts. Compare them to the result sentence, not to how impressive the interface was. Keep the ones that pass. Stop the ones that only look busy.
Use the free Risk Ladder before the agent is allowed to act, and Automate, Assist, Escalate to sort the tasks around it. The eight steps are in AI strategy for business. Collapse or Collapse is the longer argument: one person owns one result and runs agents to deliver it.
One thing to do this week
Take an agent proposal on your desk. Cross out every sentence that describes activity. If nothing remains, you do not have an outcome-based offer. Send it back with the three sentences above.
Where this comes from
This answer draws on The Five Things AI Actually Does and Automate, Assist, Escalate in AI for Business Leaders.
Your company will have its own details. To ask a follow-up in your own words, try Ask Vinay's books.
Vinay Pasricha is the founder of GoodSpace AI and the author of six books, including AI for Business Leaders.
Go deeper in AI for Business Leaders
This answer is the short version. The full argument, and the method behind it, is in the book.
