A living reference,
built from real conversations.

Claude is the product; Anthropic is the company behind it. Within Claude there are multiple models, each a different level of intelligence and cost:

• Sonnet: good for most creative and writing tasks. Cheapest of the three, satisfies most day-to-day AI needs.
• Opus: more powerful than Sonnet, more expensive.
• Fable: the newest, most powerful, most expensive model. Usage resets weekly. Handles more complex builds "more intelligently" than Opus or Sonnet, and was used for the Daena webpage build.

You can also select effort level (e.g. "max effort"). Higher effort means more compute and faster responses, but burns tokens faster and costs more.

• ChatGPT: OpenAI's product.
• Grok: X (Twitter)'s AI.
• Gemini: Google's model. Called out as especially strong for deep research.

Platforms like Perplexity and Lovable don't have their own foundation model: they call Claude or other models via API in the background. Perplexity is search-oriented; Lovable takes a build spec or file (often written by Claude) and turns it into a live, working website in minutes.

The goal over time is to build a feel for what each model is uniquely good for, and to default to the cheapest model that gets the job done rather than always reaching for the most expensive one.

API = Application Programming Interface. In plain terms: it's a connector. It exposes part of one application's code so another application can "handshake" with it: each side says "this is how you talk to me," and then data moves between them, often near-instantly.

Example: Amove is built entirely API-driven. It connects to any cloud storage provider through handshake agreements and moves data straight to the user's desktop.

"Max effort" settings buy more compute (faster machines, faster responses) but burn through tokens, and therefore money, faster. Every AI action has a real, calculable cost. One free automated site-fix example ran roughly $0.50 in API cost: trivial at low volume, material at scale (1,000 free users = $500).

The framing used by AI thought leaders: intelligence itself is becoming a commodity everyone has access to. The differentiator isn't having AI; it's how you configure it, direct it, and turn it into something that generates value.

Real example: the Daena local-business audit tool above was designed, planned, and built end-to-end in under two hours, for less than $40, using this exact loop.

The holding structure sitting above the individual ventures. Model: run 15–20 businesses at once, each its own legal entity, with CIG holding a partial (sometimes full) equity stake in each. Some ventures arrive with outside investors already in them; some come from founders bringing ideas to CIG, which then builds them out.

Bobak and Siamak (concept originators) hold 35%. CIG, which builds the business out, holds 65%.

CIG currently owns 10% of Amove. The near-term goal is to get Amove to a stable, revenue-producing state (or sell it), then treat it as one venture inside the CIG portfolio rather than the main focus.

Where no outside investor is involved, the working default between PK and a co-founder (e.g. John) is a straight 50/50 split, with a deliberate policy of not diluting it unless someone brings meaningfully large capital to the table.

Beachhead market is local Monterey County / Carmel small businesses: cafes, restaurants, service businesses (example used: Rise and Roam). Outreach plan: identify business owners, email them, offer the free scan and fix, don't stop until they opt out.

• ~4.3M small businesses in California; ~36.2M in the U.S.
• 10,000 customers × $100/mo = $1M/month run-rate, used purely as a scale reference point, not a near-term target.
• Realistic early goal floated: 500 businesses out of the local base as a starting point; even 100 customers × $100/mo = $10,000/month is framed as "a great start."

Rough margin benchmarks discussed:

• Software businesses: ~80% margin, typical.
• Pure intellectual-services businesses (no office, no equipment, no staff): ~95–98% margin.
• Physical/storefront businesses (rent, staff, inventory, equipment): ~40–50% margin.

The main variable cost in an AI-driven business is "intelligence" (i.e. API/token spend), which scales with usage and needs active budgeting, especially anything offered for free.

Framing for investor money: it's the budget that lets you actually test an idea at real scale. It pays for the "materials" and the "intelligence" (AI cost) needed to run the experiment properly, rather than a self-funded trickle.

Daena's local-business tool was explicitly modeled on an existing, well-funded competitor (Owner, ~$170M raised). The point: a legitimate, working idea copied and rebuilt fast with AI is a perfectly good starting point; it doesn't need to be novel to be worth building.

Comfort makes difficulty feel harder over time, not easier. The counter is deliberate, repeated practice on things that are uncomfortable at first, the same logic as building calluses for guitar.

The original intent: record these conversations as they happen, and keep feeding them into this wiki as a standing knowledge base, not a one-time snapshot.