Short answer: Start with one result the company already measures, one person who owns it, and a small store of memory the agent can actually read. Do not start with a platform. Run it for a fixed period, then decide to scale, fix or stop.
I'm Vinay Pasricha. I run GoodSpace AI in Noida, and I wrote AI for Business Leaders. A small company does not need an agent strategy. It needs one job that is late, repetitive and reversible, and a way to tell whether the agent helped.
Memory before the agent
Every company already runs on memory, reasoning, action and feedback. An agent is an action layer. If the memory is in one person's phone, the agent will act on a gap and sound sure of itself. That is the failure mode I see most often, and it is not fixed by a better model.
Before you connect anything, write down the few facts the job needs. Prices. The rule for a discount. The last answer you gave this customer. Who is allowed to promise a date. A shared folder with those facts, kept current by a named person, is enough to begin. A large system can wait.
This is unglamorous, which is why it gets skipped. The agent demo looks like progress. The folder looks like filing. The folder is the part the agent cannot invent. If two people in the company would answer the same customer question differently, the agent will pick one, or blend them, and you will not know which. Settle the answer first.
A 90-day shape, kept small
- Days 1 to 15: name the result, the owner, and the number you already track. Read the last twenty times the job was done by hand.
- Days 16 to 30: write the memory the agent needs, and the list of actions it may not take.
- Days 31 to 75: run it on real work, with a person checking every result at the start. Loosen the check only when the errors are boring.
- Days 76 to 90: scale, fix or stop. Write the decision down. Do not let the pilot become the process by default.
One result, not a tour of the company. If you cannot finish the sentence "this agent exists so that …", you are not ready to start. The sentence should name a customer, a document or a decision, not a department.
The owner is not a committee. In a small company it is often the founder for the first month, then the person who already does the job. The agent does not replace that person in this window. It does the preparation they can check. If they cannot check it, the memory is not ready, and the right move is to stop and write the memory, not to buy a second tool.
What not to start with
Do not start with an agent that talks to customers on its own, pays a supplier, or hires someone. Those are later, and only after you have rated the risk. Do not start with five agents because a vendor bundled them. One, finished, teaches you more than five, abandoned.
Do not announce it to the whole company on day one. Tell the two people who touch the job. They will find the cases your sentence missed. Add those cases to the "may not" list before you widen the circle.
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
Pick the job that was late most often last month. Write the result, the owner and the number on one page. If you cannot find a number you already track, you are not ready for an agent. You are ready for a count.
Where this comes from
This answer draws on The Company Brain and The 90-Day AI Leadership Playbook 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.
