Before you jump into an unfamiliar industry, have your AI agent scout it first

What you get

You want to move into a new industry for a job, or start work or a business in a new field. The biggest risk is jumping in based only on hearsay. It looks busy from the outside. Once you are inside, you find that someone else is making all the money.

This tutorial breaks "mapping an unfamiliar industry" into four parts: research, talk to people, try it yourself, and decide. Your AI agent does the research and the fact checks. Below are six prompts you can copy as they are. Replace the text in brackets with your own details.

Let's be clear about who does what. Your AI agent searches public sources thoroughly and double-checks the conclusions that matter most. You talk to people and try the work yourself. You make the final call. The part that public sources cannot answer is often the part that decides whether you should go in.

Here is the full process. First, your AI agent searches the web for the industry, and every number comes with a source. Then you open a fresh chat and use it only to find errors. It checks the claims that matter most against the original text. Next, you take the questions the web could not answer to people and talk with them. Then you pick a small task and try it yourself. Finally, you put together what you found, what people told you, and what you built. Work out how much you can afford to lose, and pick your first small step.

This tutorial uses one example throughout: you want to run a business offering AI customer support for local pet shops. Swap in your own industry.

Why you should not decide right after AI finishes the search

This method borrows from three well-known ideas. Michael Porter (Harvard Business School) created the five forces model. It shows who ends up with the profits in an industry. Steve Blank's customer development rests on one idea: there are no facts inside the building, so get out of the building. Saras Sarasvathy's effectuation logic starts from what you already have, and asks how much you can afford to lose first. You do not need to study them before you start. The prompts already show how to use them.

AI has three weak points in this work. Watch for them early:

• If you don't know the industry, everything it says sounds right. It can make an unreviewed report sound like settled fact.

• Public sources only tell half the story. How prices get negotiated, what customers really care about, and how competitors actually work are mostly not online.

• It does not know who you are. Your savings, your contacts, and your skills decide whether the same industry is an opportunity or a trap for you.

Step 1: Have your AI agent research the industry

Run this step with an AI agent that can search the web. Send it this prompt:

Why this works: the final table matters most. It pulls out the conclusions with the biggest impact, so the next step only checks those few instead of everything. Keep the line "write 'not found' if you cannot find it." Without it, the AI will fill the gaps with numbers that look plausible.

Step 2: Open a clean chat and look only for errors

After the research, open a new chat, or switch to a different AI agent. It did not take part in the research, so it has no reason to defend those conclusions. Paste the table from the previous step into it:

You are a fact checker. Your only job is to find errors. Below is a table of industry research conclusions written by another AI:
[paste the table from step 1]

Check only the rows that matter most for "go or no go":
1. Open each row's source, find the sentence in the original text that matches, and copy it word for word.
2. Judge whether the conclusion matches the original. Label each one as: Accurate / Needs correction (write the correct version) / Exaggerated or taken out of context / Source cannot be opened or the original text cannot be found.
3. Watch for these common mistakes: calling a "share of organizations" a "share of projects," treating a preliminary report as a final result, treating the number of job posts as the number of people hired, and mismatched years or regions.
4. Last, tell me: after the corrections, which conclusions changed, and whether those changes affect "go or no go."

Why this works: point 3 lists the most common ways facts get swapped. Without that reminder, a checker can easily accept everything as it is. It is normal for many items to be corrected or downgraded after checking. That means this step did its job.

Step 3: Ask peers and customers

Some questions can only be answered by people. Each person knows one part of the industry. Do not expect one person to explain the whole thing:

• Sellers know why customers buy and how prices are built. But they usually only talk about the deals that closed.

• Buyers know how their last purchase went and what they regret. But they only speak for their own company.

• Doers know which step, from signing to final acceptance, is the hardest.

• Newcomers know where their first job came from. But they usually only share the wins.

• People who left know why they quit. When you listen, watch for strong emotions or another business they want to sell you.

Ask your AI agent to build a list of people to contact and draft what you will say:

I am researching [industry name]. Below are the questions I still cannot answer from public sources, and need to ask people:
[paste the questions marked "not found" in step 1 and the open questions left after step 2]

Please help me:
1. For five groups (sellers, buyers, people who do the work, newcomers, and people who left), list the 2 most useful questions for each group. Each group should only be asked about the part they know best.
2. For each group, suggest where to find them (industry events, communities, platforms, job pages). Mark which places you are sure exist and which ones I need to check myself.
3. Write a chat outline of about 15 minutes. Open with one sentence that explains why I'm asking. The main part should focus only on one recent project the person finished. Ask them to walk through it from the start: who handled it, what tools they used, where it got stuck, and how long they waited. Do not ask vague questions like "what do you think of the industry?"
4. If I ask about money, only ask about amounts that already happened, ask for rough ranges only, and first say, "Feel free to skip this." If they decline, do not rephrase and push again.

Why this works: asking only about "the last job they finished" gets you facts, not opinions. Asking about money with a range and an easy way out makes people willing to answer. It also keeps the conversation friendly.

After the talk, give the chat notes to your AI agent to compare:

Below are my chat notes with [their role], and the research conclusion table from earlier:
[paste chat notes]
[paste conclusion table]
Please compare them:
1. Which claims match the research, and which do not? For claims that do not match, put both original quotes side by side. Do not pick a side for me.
2. What new information came up in this chat that the research did not cover?
3. Which one of my earlier judgments did this chat change?
4. Who should I talk to next, and what should I ask them?

Step 4: Pick a small task and do it yourself

Research and interviews show other people's experience. You only find out whether you can do the work by trying it. You do not need a client to start. Make your own practice task:

I want to get into [industry name]. My current skills: [e.g. I can build simple automations with my AI agent, and I have done e-commerce customer support].
Here are the job requirements and public project examples I collected:
[paste]
Please help me:
1. Work backward from the most common type of delivery in this industry. Tell me which skills it requires, and mark the ones I already have and the ones I am missing.
2. Propose a small project I can finish in one weekend. Its input and output should be very clear, and when it is done, someone should be able to test it and say whether it works well. For example, "merge the appointment records from three spreadsheets at a pet shop into one customer list." Do not start with a full system.
3. Use simulated data instead of real customer data, and tell me what the simulated data should look like.
4. Define what "done" means: someone else can understand what problem it solves, can run it once by following the instructions, and the known limits are written down.
5. Remind me of things that job postings usually leave out but that I must handle when taking projects myself: finding clients, pricing, signing contracts, and collecting the final payment.

When you finish, have your AI agent look for errors first. Then find two people to test it. One should be someone who might use it. The other should be someone who has done a similar project. Ask them to follow the instructions on their own while you watch where they get stuck. Do not rush to explain.

Step 5: Put the evidence together and pick a first small step

Your AI agent can help you organize everything in the last step. But you make the decision yourself:

Here are all my materials for researching [industry name]:
- Verified research conclusions: [paste]
- Conclusions from the chat comparison: [paste]
- Results from my practice project and the problems other people found: [paste]
- My own situation: my savings can last [how many months], my fixed monthly costs are [amount], my skills are [list them], and the people I know are [list them].

Please help me:
1. Put the evidence into one table with three columns: what I found, what I learned from people, and what I built. Mark any places where the three contradict each other.
2. Check five business gaps for me: where the first orders will come from, how much work it takes to clarify what a customer needs, how much it costs to win one job, how payment is collected, and who handles maintenance after delivery. If you cannot answer one, mark it "to be verified."
3. Based on "how much I can afford to lose" rather than "how much I could earn," give me one small first step for each of three ways in: join a company to do this work, take on a small project on my own, or enter from the area closest to my current field. For each first step, write how much time and money it takes, how long before I see results, and what result tells me to stop.
4. Do not decide for me. Only lay out the options and their costs clearly.

Why this works: point 3 changes the question from "is this industry good?" to "how much can I afford to lose?" That is the most useful idea from effectuation. However hot the industry is, the first step should be small enough that you can absorb the loss if it fails.

You can change your decision too. If you work at it for a while and find your first conclusions were wrong, go back to Step 1, add the new evidence, and run the process again.

Notes

• Check the communities, events, and platforms the AI lists yourself. Some may come from memory and no longer exist.

• Only trust numbers you have checked against the original. If a number has no source, or only a secondhand source, treat it as "to be verified."

• Do not sell when you first talk to people. Say clearly that you want advice. Thank them after the talk, so you can ask again later.

• Start with a field you already know. When you know the material and can make the decisions yourself, your first delivery will go much more smoothly.

References

• The Five Competitive Forces That Shape Strategy (Harvard Business Review, Porter)

• Steve Blank's blog (customer development)

• Effectuation (official site)

• New to writing prompts? Read the prompt handbook first.