how to build an agent that finds you customers:
If you have something to sell then there’s multiple customers across the world that are looking for exactly what you offer all the time. The problem is they just don’t know you exist yet... YET!
Let’s change that with either Hermes or Grokbot when you have some time this week (or even this evening if you have the time).
This is the normal way to find customers:
But there’s a better signal… people are literally posting “does anyone know a XYZ?” or “a good alternative to X?” or “how are you solving X problem?”… they’re customers for people waiting to be found by YOUR product… you don’t believe me?
the posts are endless tbh and cover any category…
why are they so important?
79% of B2B buyers initiated the first engagement with vendors themselves in 2025.
6sense’s study covered nearly 4,000 B2B buyers. it also found buyers waited until roughly 61% of the way through their buying journey before first engaging sellers.
the point is they have a problem and they’re looking for a solution so the agent we’re about to make will follow a simple workflow:
find signal → qualify buyer → score intent → save lead → draft outreach
this article is going to go through the following contents:
then you review it and make it personal before sending.
here’s how you go from old prospecting to building the entire buyer intent agent with either hermes or grokbot, here is how:
1: HOW
Step 1: Tell the Agent Who You Want
before searching anything, give the agent enough context to know what a good lead looks like.
tell it:
for example:
this gives the agent two things it needs:
intent: does this person have the problem?
fit: are they actually somebody you could sell to?
you need both.
a perfect ICP with no intent is just another cold lead.
a strong complaint from somebody completely outside your market is not much better.
Step 2: Tell It What Buying Intent Looks Like
do not ask the agent to find “mentions”.
you are not building a social listening dashboard.
you want behaviour that suggests somebody is:
good searches include:
then give each signal a simple intent score:
this does not mean a 9/10 lead has a 90% chance of buying.
it is simply a way of telling the agent:
show me the strongest opportunities first.
recency matters too.
someone asking for a recommendation yesterday is much more interesting than somebody complaining about the same product a year ago.
Step 3: Search Where Buyers Reveal Problems
start with X and Reddit.
then expand into other public places where your buyers spend time.
X is useful for:
do not stop at the original post.
read the replies.
one person might ask:
“what are people using instead of X?”
then underneath you find:
one post can reveal several potential buyers.
Reddit is useful because people often explain the problem in much more detail.
search:
Search the Wider Web Too
depending on what you sell, also look at:
different sources reveal different things:
the main rule is:
search for evidence of a problem, not simply mentions of your category.
Step 4: Research and Qualify the Buyer
once the agent finds a good signal, it needs to answer two questions:
who is this person?
could their company actually buy?
research:
then look for useful company context:
this helps separate a random complaint from a genuine commercial opportunity.
and give the agent one hard rule:
if something cannot be verified, write Unknown.
never let it invent:
unknown is better than fake enrichment.
Step 5: Filter, Score and Save the Best Leads
this is where the agent becomes useful instead of noisy.
tell it to reject:
then save every qualified lead somewhere persistent.
that could be:
for each lead, save:
also keep a Seen Leads list.
every new search should check against it.
at minimum:
same person + same signal = do not return again
but if the same person produces a genuinely new buying signal later, the agent can surface them again.
Step 6: Draft Outreach From What They Actually Said
this is the important part.
most AI personalisation tries to manufacture relevance.
you do not need:
“saw you went to Oxford and support Arsenal…”
if the prospect already wrote:
“we’re looking for an alternative to X because the data is terrible.”
the signal is the personalisation.
tell the agent to:
for example:
“saw your post about struggling with [specific issue]. we built X around that problem. happy to show you how we’re handling [relevant outcome] if useful.”
then keep one important boundary:
the agent drafts. you approve. ideally you make it your own.
automate:
keep human approval for:
now you have the actual system.
search → qualify → score → save → draft → approve
the only question left is where you want to run it.
2: HERE’S HOW YOU RUN IT WITH HERMES
Step 1. Enable the Tools
open: hermes tools
enable: x search, web search, browser tools where needed.
if you want qualified leads saved directly into another service, connect your CRM, spreadsheet or database through MCP where supported.
X Search needs the relevant xAI credentials before it can be used.
Step 2. Create a Buyer Intent Bot
open Bots.
create:
Buyer Intent Agent
give it:
then tell it:
never contact anyone without approval.
Step 3. Test It Manually
do not automate it immediately.
give it the Buyer Intent prompt below.
look at the leads.
if they are bad, explain why.
for example:
run it again.
keep correcting it until the results are genuinely useful.
Step 4. Save and Automate It
once it works:
save the process as a Buyer Intent Skill.
then schedule it.
for example:
“run the Buyer Intent workflow every Monday morning. search the previous 7 days. ignore Seen Leads. only return new prospects scoring 7/10 or higher.”
Hermes scheduled jobs run in fresh sessions, so make sure the task contains the information it needs or has the relevant Skill attached.
your Hermes workflow is now:
X + Reddit + web → find signal → qualify → score → save → draft → approve
3: HERE’S HOW YOU RUN IT WITH GROKBOT
the process with Grok Bot is almost identical.
Step 1. Create the Bot
create a new Grok Bot.
call it:
Buyer Intent Agent
give it:
Step 2. Connect What It Needs
Grok Bot can work through its cloud computer using:
connect your CRM or other sales tools if useful.
where a connector exists, use it.
where one does not exist, the Bot can use the browser.
Step 3. Run the Workflow Once
give it the same Buyer Intent prompt.
review the results.
reject weak leads.
explain why they were weak.
then run it again.
the important thing is to make the one-time task reliable before automating anything.
Step 4. Save It as a Skill and Routine
once the process works:
save the method as a Buyer Intent Skill.
then create a recurring Routine.
the distinction is simple:
Skill = how the task is done
Routine = when the task runs
for example:
“run Buyer Intent every Monday. search the previous 7 days. ignore Seen Leads. only return prospects scoring 7/10 or higher.”
Grok Bot can run the Routine through its cloud computer even when your laptop is closed.
your Grok workflow is now:
public conversations → find signal → qualify → score → save → draft → approve
4: COPY AND PASTE THIS PROMPT
you can use the same core prompt with either system.
You are my Buyer Intent Agent.
YOUR JOB
Find people and companies publicly demonstrating buying intent or a strong active problem that [PRODUCT] solves.
PRODUCT
[Describe what you sell.]
IDEAL CUSTOMER
[Describe the industries, company sizes, roles and geographies you want.]
COMPETITORS
[List competitors.]
PROBLEMS WE SOLVE
[List the specific problems your product solves.]
EXCLUSIONS
[List people, companies and situations that should be rejected.]
SEARCH WINDOW
Prioritise signals from the previous 7 days unless instructed otherwise.
SOURCES
Search public X conversations, Reddit and the relevant public web.
Also search useful sources such as Hacker News, GitHub, specialist forums, product reviews, Product Hunt, job adverts and company announcements where relevant.
BUYING SIGNALS
10/10: actively asking what product to buy
9/10: actively looking for an alternative
8/10: comparing products or vendors
7/10: strong complaint from a relevant potential buyer
6/10: actively asking how to solve the problem
5/10: explicitly says they have the same problem
Below 5: normally reject
FOR EVERY STRONG SIGNAL
1. Read the surrounding conversation.
2. Identify the person.
3. Verify their role and company where possible.
4. Research the company.
5. Check whether they fit my ICP.
6. Check whether my product genuinely solves the problem.
7. Score the buying intent.
8. Reject weak or irrelevant leads.
9. Check the result against Seen Leads.
VERIFICATION RULE
Never invent missing information.
If something cannot be verified, write "Unknown".
RETURN
Only return the strongest qualified opportunities.
For each lead include:
- Name
- Role
- Company
- Source
- Original signal URL
- Signal date
- Problem they described
- Why they qualify
- Relevant company context
- Intent score
- Reason for the score
- Draft public reply where appropriate
- Draft private message
APPROVAL RULE
Never contact anyone.
Do not send emails, DMs, public replies or add anyone to an automated outbound sequence without my explicit approval.hope this helps, I’m definitely interested in vibe coding solutions to problems and utilising this agent, or I’m even interested in doing affiliate marketing with this agent if the right offer came - the world is your oyster lads - enjoy!
Ya dig my Paul Allen business card looking header?
