I Found the First Vibecoded Business That Makes Real Money

Business outcome (what worked)

• Kristoff, founder of an organic craft soda business, used AI-assisted “vibe coding” to build an outbound lead-generation “AI salesperson” that:
  • Continuously finds, enriches, and emails prospects in batches (no manual daily outreach).
  • Scores leads for fit against the customer’s ICP.
  • Generated meetings/negotiations with major hospitality brands (e.g., Four Seasons, Belmond, InterContinental) and other B2B targets (restaurants/chains), including negotiations with a retail chain in Mexico.
• Early proof-of-value:
  • After pressing “start,” he received notification emails the next day showing positive response activity—validating immediate demand.

Key context & strategic rationale

Pre-AI period (2019–2020 to 2023)

• Ran the soda business without AI.
• Struggled with sales workload and generic cold-email approaches.

AI adoption trigger/timing

• The “ChatGPT” era led him to experimenting about 3 years prior.
• “Vibe code”/agent development began about 1.5 years ago.

Why outbound + why B2B

• Outbound required adapting messages per account/person, often involving research.
• He believes the approach works best for B2B, not B2C (he notes his father’s realtor use-case didn’t fit as well).

Core operating model (how the AI salesperson works)

Lead pipeline automation

• The agent runs continuously and “never stops emailing” while active.
• It generates leads in batches, then scores them for closeness to the customer’s ICP.
• Reported 3,284 leads currently in the pipeline that have not yet been contacted.

Data inputs

• Can ingest:
  • Upload your own Excel lead lists.
  • Upload leads from your existing customers (so it avoids contacting them).

Filtering/compliance deliverability controls

• Requires correct email authentication to reduce bounces/spam risk:
  • SPF, DKIM, DMARC
• Supports opt-out handling (prospects can reply “Please stop”).
• Does not contact in countries where it’s not legal (high-level compliance rule).

Humanization + messaging

• Uses an AI copywriting layer to sound natural.
• Optional tactic: occasionally “make spelling mistakes” to appear more human.
• AI disclosure is usually omitted unless a customer requests it (one customer wanted “Hey, I’m an AI agent” in first contact).

Outbound safety throttling

• Learned deliverability limits after early failures (spam listings).
• Guidance/best practices:
  • Don’t send through the main domain.
  • Use multiple domains.
  • Up to 3 inboxes per domain.
  • About 75 emails/day per inbox, then scale via domain/inbox rotation.

Frameworks / playbooks mentioned or implied

Outbound vs. Inbound (GTM framing)

• Inbound = marketing funnels.
• Outbound = proactive research + tailored messaging.

ICP scoring

• The agent scores leads based on fit to the customer’s Ideal Customer Profile.

Deliverability “warm-up + validation” playbook

• Early stage pattern:
  • High volume → high bounce → spam listing.
• Then implemented:
  • Domain warming (send slowly to improve reputation).
  • Validation/filters before sending (ensure emails/people are real).

Iterate-and-improve loop

• Weekly analysis of results.
• Agent updates copy/angles over time based on performance.

“Agency setup” onboarding playbook

• Initially treated it like DIY SaaS.
• Shifted to an agency model:
  • Deeply dig into the client’s business/processes.
  • Build/encode the agent with customer context and target-specific strategy.
  • Run campaigns for the client.

Metrics & KPIs (explicit numbers and targets)

Deliverability / quality

• Early bounces: 30–40% bounce rate.
• After validation + warming + filters: 2–5% bounce rate.
• Early risky behavior:
  • Sent 200–300 emails/day from a main domain/inbox (flagged as a red flag).

Pipeline & lead throughput

• Leads in pipeline: 3,284 uncontacted leads.
• Early positive replies:
  • 1–2 positive replies per day once it started working.
• Early lead generation for other clients:
  • About 20–40 “good leads” in the first 1.5 months (average stated range).

Customer acquisition examples (qualitative but specific)

• Conversations/advancement with:
  • Four Seasons Hotel (1)
  • Belmonds (2)
  • InterContinental (1)
  • Aman (negotiations / getting into)
• Negotiations with Liverpool (large retail chain) in Mexico.
• Additional restaurants/chains (a few).

Pricing / commercial metrics

• Subscription tiers:
  • Basic tier: $4.99
  • Includes 1 domain and up to 2 email inboxes
  • Tier differentiation based on volume (higher volume = higher tier)
• Early V1 cost (prototype test):
  • Around $18/month, plus mention of an additional ~$20 item (context unclear), described as “absurdly low” relative to value.

Labor cost comparison (ROI framing)

• Mexico outbound labor estimate:
  • 30–35,000 pesos/month, plus benefits/insurance/taxes/fees/commissions
  • All-in estimate: ~50,000 pesos
  • Compared to roughly $2,500/month+ in US terms
• This frames the AI agent as replacing paid sales labor rather than only competing with SaaS features.

Concrete examples / case studies

His own soda business as guinea pig

• Used his own company to test risks:
  • Mistakes led to spam listings.
  • He fixed deliverability and stopped damaging behavior.
• The system is tailored to his company and:
  • Converses after prospects reply.
  • Guides them toward accepting free samples.

Client industries and outcomes

• Client types included:
  • Marketing companies
  • Other CPGs (consumer packaged goods)
  • Ticketing services for events/venues
• One industry was implied to be harder (cold-email fit issues), but most produced “great results.”

Operational lesson learned from early experiments

• Problem:
  • Too much volume too fast + invalid emails → spam issues.
• Fix:
  • Validation + warming + strict sending limits → dramatically improved bounce rate.

Actionable recommendations (extracted from what he actually did)

• Start with your own business to validate deliverability and targeting risk before selling externally.
• For cold outbound agents:
  • Prioritize email authentication (SPF/DKIM/DMARC) early.
  • Warm domains and throttle sending (multi-domain/inbox strategy).
  • Validate contacts and emails to avoid bounce spikes.
• Encode domain knowledge into the agent:
  • Provide decks, presentations, price lists.
  • Use a questionnaire to extract product differentiators and sales talking points.
• Plan for the real bottleneck after automation:
  • The system can generate leads faster than many founders can qualify and follow up, so align lead volume with sales capacity.

Revenue / business model (high level)

• Monetization:
  • Subscription tiers (volume-based) starting very low (basic tier $4.99).
  • Transitioned from DIY SaaS experiments to a boutique agency onboarding approach (done-for-you setup and agent programming).
• Positioning:
  • “Pricing vs. labor” ROI argument: instead of hiring outbound staff, clients can deploy an agent that reduces manual research and copywork.

Presenters / sources

• Kristoff: founder of the craft soda business; creator/operator of the AI outbound “salesman” agent.
• Video narrator/interviewer: unnamed in the subtitles; asks questions and summarizes/frames the story.