Agents for High Intent Leads

Why this exists

Cold outreach got a bad name because most of it is spray and pray. Buy a list of 10,000 "SaaS founders," blast the same email, and almost all of them mark you as spam. That is gambling with your domain reputation.

High-intent outreach is the opposite. You find the handful of people who are feeling your problem right now, and you show up while it still hurts. Send 20 of those and get 5 replies. Not 2,000 to get 1.

Building complex software has become accessible with Replit, Lovable, Emergent, and similar tools. Customer-finding has not. The builder still has to understand ICPs, signals, lead quality, outreach, email validation, copy, follow-up, and reply handling.

That gap is absurd. The goal with Autark is to make customer-finding tooling feel as accessible as building on the new AI app builders.

The core principle

Every outreach tool fails the same way: it tries to do everything. Finding the right people and writing to them are different jobs. Mash them together and you get a mediocre version of both.

• Signal engine — find who is worth your time, and why now.
• Conversation brain — turn the signal into a natural message, then handle the reply.
• Human control — keep judgment, taste, risk, and serious buyers with you.

The mental model is: intent source is the radar, Claude/Codex is the local sales operator, and you are the decision-maker.

Step 1: intent

This is the half where the money is, and it is not "send more emails." It is knowing who is worth writing to before you write a word.

A high-intent lead is already standing next to the problem:

• complaining about the tool they use now
• asking in public for the thing you built
• hiring for a role that only exists because of your problem
• engaging with a competitor or category conversation
• shipping something that only makes sense if they need you next

The exact signal depends on your buyer. A B2B tool and a consumer app leave different fingerprints. Find yours and you stop guessing.

Option 1 — use Autark locally. If you already pay for Claude or Codex, use the subscription you already have. Autark runs in the background on your laptop and gives your agent the structure to do this properly. It learns your product, makes one sharp guess about who the customer is, hunts public buying signals for that guess, builds the list, then sharpens the next guess.

Option 2 — use paid lead tools. If you prefer a hosted signal engine, tools like GojiBerry, OpenFunnel, and Instantly can help collect intent and lead data.

A paid tool is a finished machine you rent. Autark turns the agent you already have into the machine, and it gets sharper because it runs on your own memory and context.

Step 2: conversation

Claude or Codex should not guess intent. Its job starts after Step 1 has already said: this person is worth your time.

Then your agent handles:

• making sense of the signal
• turning the signal into a natural opener
• managing tone and pacing
• holding memory across the conversation
• handling objections and hesitation
• knowing when to slow down or advance

This is where Claude Cowork or Codex becomes useful: not as a random message generator, but as a calm sales operator sitting next to you.

Set up the skills

If you want Claude or Codex to handle the conversation side better, copy this block into your agent.

Set up these sales skills for me.

Primary install path:

npx skills add kiluazen/kstack -s outreach,email-finder -y -g

After installing:
- use outreach when writing LinkedIn DMs, cold emails, follow-ups, or reply drafts
- use email-finder before sending email to find and verify the right address

If npx is not available, use the raw skill files instead:

outreach
https://raw.githubusercontent.com/kiluazen/kstack/main/skills/outreach/SKILL.md

email-finder
https://raw.githubusercontent.com/kiluazen/kstack/main/skills/email-finder/SKILL.md

Human oversight

Autark handles the repeatable work, then brings you back in when judgment matters: weak signals, risky messages, objections, serious replies, and meeting asks.

Humans should still own:

• reviewing first messages to leads you really want
• reviewing call-booking messages
• stepping in when emotion appears
• handling serious replies
• adjusting tone for enterprise leads
• deciding when not to send
• refining ICP and signals

You are still the boss. Volume is the machine's job. Trust is yours.