A pointer, not a product

Starburst AI™

Next Great AI Innovation Is Here!!

Harvested or Produced

The candy in front of you is not the offer. It is a claim about who the provider thinks you are.

The structure of your AI use determines the nature of benefit derived from — or for — your human brain.

Harvested or Produced. Structure yields results.

Read the format, not the feature list

The candy is not the product. The candy is the question.

Every provider hands you a juicy square. Glossed. Saturated. Rounded. Dripping. A name you choose, a face you design, a colour you pick — Meta's Muse, OpenAI's dots, and every agent that arrives grinning.

You are told to read the feature list. Read the format instead. A format is a sentence about the reader. A juicy square placed in front of a person is a claim that the person is a mouth — that attention is the input and behaviour is the output, and that the transaction runs one way.

This is what behavioural targeting looks like before the targeting starts. The icon is not decoration on top of the business model. The icon is the business model, rendered in the only language that survives a two-second glance.

Formats are chosen. Someone picked the square. Someone chose the drip. That choice is deliberate — and it points.

It points at a customer structure: the shape of the person the provider has already decided to serve. And it points at a literacy: whether you know which door you walked through. The candy is a signpost. It is aimed. The only question is whether you can read where it is aimed.

The variable is not the candy

Two people open the same product. Same icon. Same gloss. Same juice.

One is harvested. One produces. The candy is identical in both cases — which means the candy cannot be what differs. If the picture is the same, the picture is not the mechanism.

The difference is what the AI use sits inside. A personal account, opened in a personal name, holding a personal identity, wired to a personal inbox — or a container: an entity that contracts, a tier that does not train, a contract that says what may be done with what you supply.

Structure is the only variable that moves. Not the icon. Not the model. Not the prompt. The container.

The fork

Same candy. Different structure. Different harvest.

Two doors into one building. Same provider. Same model. Same interface. Two entirely different regimes.

Everything below is available to you today. The only thing separating the two columns is which door you chose — and whether anyone told you there were two.

Harvested

You are the inventory

  • You contract personally, in your own name
  • Your conversations train the model by default
  • Ads are drawn from your past chats and memory
  • Your mail, files and repositories are read-scoped to your identity
  • Your behaviour becomes the company's own first-party record — no third party needed
  • No DPA. No indemnity. No audit. No retention control.
  • Choosing to see no ads costs you capability

You pay a subscription and you supply the substrate. Value accrues away from you.

Produced

You are the customer

  • An entity contracts — the counterparty is not your name
  • Training is off by default on business and API tiers
  • No ads on the tier, and none drawn from your history
  • Connectors are org-scoped, revocable and auditable
  • Retention is contractually set; zero-retention available by agreement
  • DPA, indemnity, SOC 2, audit logs, regional processing
  • Your IP is the entity's asset, held under a duty of confidentiality

You pay, and the contract says what may be done with what you supply. Value accrues to you.

Exhibit A · the harvest, in the wild

This is what the Harvested column looks like when it runs.

No hypotheticals. A company used its own customers' records to train a model that finds the people most likely to lose — and then sent them candy.

Source

Electronic Frontier Foundation · Deeplinks

DraftKings Is Using AI to Supercharge the Harms of Online Behavioral Advertising

eff.org/deeplinks/2026/09/draftkings-using-ai-supercharge-harms-online-behavioral-advertising ↗
01

…using its customers’ betting records to train a machine learning model to find losing gamblers. Once found, DraftKings sends these customers targeted advertising designed to lure them back to the site to place more bets—bets that DraftKings thinks will be losing ones.

— EFF Deeplinks
02

DraftKings seems to be using solely “first party data” to target their ads, meaning that they’re using only the data they collect directly from their users and are not buying any additional data from third parties to fuel their machine learning model. This highlights how policy solutions that only limit third-party data sharing and selling would not be enough to prevent these predatory advertisements. Rather, policymakers must ban online behavioral ads.

— EFF Deeplinks
The ledger is the target.Betting records are a ledger of losses. Whoever holds it can aim at you. That is Action 04, below.
No third party required.The model ran on first-party data alone. That is the Harvested column's "own first-party record," in practice.
Policy lags; structure doesn't.Rules that only police data sales miss this entirely. The container you choose is the protection you control today.

Action-based literacy

This is not a setting you toggle. It is an action you take.

Literacy here has nothing to do with technical skill. It is knowing that consumer and business are two different legal regimes wearing the same interface — and choosing deliberately instead of by default.

The person who types the same prompt into a personal account and into an entity-held API key has performed two completely different acts. Only one of them is a gift.

Action 01

Never sign commercial AI terms as a natural person. Let the container be the counterparty.

Action 02

Do commercial work where training is off by default and a contract governs the data.

Action 03

Keep your mail, files and repositories out of a consumer agent. The intake is the exposure.

Action 04

Never let someone else's server hold the ledger of your losses. That record is the target.

Action 05

Audit what you already connected. Disconnecting does not un-take what was taken.

Action 06

Build defaults that are safe. Architecture beats willpower — you will not out-discipline a reward loop.

Harvested or Produced.

The candy is the same in both hands.
Structure yields results.