Internal · Pelikan / Ludia · agentic company · noindex

The know-how is the agents. The trip is what they produce.

This is not a villa website with some bots on the side. It is a company made of agents. Each agent is a named asset, owned by a team, with a learning loop. Multiple teams work different sides of the same firm — supply, media, demand, OS — and every cycle that agent gets better. The villa inventory already proves one of them works.

Open the roster Scout already live Decisions
147Named villas
9Agents to stand up
4Teams / sides
89Outreach-ready
1Live scout OS
8Still to create

01 — The company

Nine agents. Four teams. One critic.

We create each agent as part of this project — not as a borrowed chatbot. The IP is the spec, the evals, the taste loop, and the data that agent touches. Teams do not share a generic “AI assistant.” They own a side, ship a better version, and hand a cleaner interface to the next agent in the chain.

Supply

Scout · Outreach · Site intel · Editor. Find, contact, score, keep true.

Media

Film factory. Extract the house, rate the site, shoot wow. Pelistar team.

Demand

Concierge · Trip packer · Ads. Sell the trip, not the night.

OS

Vision critic + vault. Brief, bar, kill, compound. Never solo-hero the craft.

02 — What the agents produce

One flywheel. Agents on every step.

Video SaaS and a world concierge page are not two startups. They are outputs of the same chain. If an agent is missing, that step is still a human hero — and it will not compound.

01

Scout finds the house

Fellner, live. Named villas 5–20, value vs Mekkala. This is the only agent that has already shipped a body of work.

02

Outreach earns the listing

To create. WhatsApp/mail as Pelikan.sk. Drafts by the outreach agent. Send stays Henry. Follow-up weeks later with a film from the film agent.

03

Concierge + trip packer sell the trip

To create. Best-of-best page, flights, cars, eSIM. Commission on the stack. Ads agent only after an owned Pelikan door exists.

04

Every cycle, the agent gets better

Henry Good/Bad → taste patterns → next scout batch. Film scores → next brief. Outreach replies → next script. That loop is the moat. Empty today on scout. First compounding asset we are not using.

03 — Proof the first agent works

Scout already produced 147 named houses.

Canon inventory v2.1 (2026-08-26). This is not a pitch. It is the output of one agent. The other eight do not exist yet as products of this company.

What Scout shipped

Named bookable only. Capacity 5–20. Ban Mekkala as value benchmark. Explorer + map live (ops, noindex). 127 floorplans. 113 WhatsApp, 64 email. Website UX already scored (mean 7.52) — that is Site intel v0 sitting inside Scout’s file.

Explorer · Map

What is not yet an agent

  • Editor — AIPali 300 is a static package. No CMS. Colleagues bootstrapped the factory floor, not the editor agent.
  • Outreach — zero messages sent. No spec, no evals, no two-variant loop.
  • Film — zero Pelistar villa films on this project.
  • Concierge / trip packer / ads — no customer surface.

04 — Go to market · owned by agents

Cheap tests. Each test has an agent owner.

We do not “run Facebook.” The ads agent runs a capped test after the concierge agent has a URL. Kill dates belong to the agent’s eval suite.

Owner: Concierge agent · Demand team

8–12 villas on a Pelikan collection / VIP WhatsApp. SK/CZ groups, Samui + Bali first.

Kill if 0 quote requests in 30 days live.

05 — Money

The company sells trips. Agents make that cheap.

P&L

Trip concierge

Villa + flights + cars + eSIM. Commission on the stack. This is what Demand team agents exist to produce.

Not the firm

Video SaaS / one-time clips

Film agent’s output is a gift and an ad. Login SaaS only if three owners pull. Commodity renderers already exist.

Shape, not a forecast · prices in file are low-confidence

Monthly contribution before ads

€4,500

06 — 90-day MVP

Stand up the agents. Then the tests.

9 specs — Purpose, envelope, evals, human owner, learning loop. No spec, no agent.

Scout loop closed — Henry rates 20 cards. Taste file stops being empty.

Outreach v0 — 20 drafts, Henry sends, replies scored.

Editor v0 — SOA on AIPali 300. Inventory without hand-JSON.

Film v0 — 3 houses, critic above adequate.

Concierge v0 — 1 trip page, owned Pelikan door.

Day-90 kill: if we still have one hero doing five jobs, we did not build a company. We built a memo. <3 owner replies and 0 owned quotes → keep Scout as internal tool, freeze the rest.

07 — Cofounder decisions

Mark these in the room.

The first decision is the company shape. The rest are which agents we create this quarter.

Objections

Why not just use the existing fleet chat?

Because a chat is not an asset. An agent with a spec, a team, evals, and a learning file compounds. A chat resets. This project needs nine compounding assets, not nine prompt windows.

Why create agents we will not fully run in 90 days?

So the interface between teams is real. Film cannot wait on a hero to “also do video.” Outreach cannot live in Scout’s leftover context. Empty slots are honest. Borrowed chatbots are not.

Is the editor already in AIPali?

No. Project 300 is the static scout package. The editor is the next agent the product team creates on that repo.