Delivering what was promised

Before AI became something you paid for, it was something you were promised. The early language (the public charters, the founding documents, the talking-head appearances on stages around the world) said that this work was too important to be controlled by any one company. It was going to be science delivered for everyone. Open access, broad benefit, accountability to humanity rather than to shareholders.

Most of the people who said it at the time meant it. The pitch itself wasn't dishonest. What's dishonest is what the pitch turned into. The drift isn't hidden, and it isn't controversial: the same companies who said the words above filed regulatory paperwork, restructured their entities, and watched their pricing pages drift in a different direction over a roughly seven-year window. The dates below are the public ones.

A short timeline, on the record

1. June 2018. Google publishes its AI Principles, a public commitment to specific norms (no autonomous weapons, no surveillance violating international norms, accountability for impact). The principles were the strongest version of the “benefit to humanity” framing committed to paper by a major lab.

2. March 2019. OpenAI restructures from a pure nonprofit into a “capped-profit” entity called OpenAI LP. The capped-profit structure was framed as a way to attract investment without abandoning the original mission. The original 501(c)(3) board retained nominal oversight.

3. January 2021. Anthropic incorporates as a Public Benefit Corporation, a legal structure that requires the board to weigh stakeholder benefit alongside shareholder returns. The PBC framing is closer to what the original promise actually meant in legal terms.

4. February 2023. Getty Images files suit against Stability AI in both US and UK courts, alleging the training data for Stable Diffusion included millions of Getty-watermarked images without license. The first major copyright case to put the training-data question in front of a judge.

5. September 2023. The Authors Guild and seventeen named authors file a class action against OpenAI in the Southern District of New York, alleging large-scale unlicensed use of copyrighted books in training. Followed within months by parallel actions from publishers and other writers' groups.

6. December 2023. The New York Times sues OpenAI and Microsoft in the same district. The complaint includes side-by-side examples of model outputs reproducing Times articles in substantial part, escalating the legal stakes from “general training” to “specific recall.”

7. November 2023. OpenAI's nonprofit board removes CEO Sam Altman, then reinstates him within five days under a reconstituted board. The episode demonstrated, in public, that the nonprofit oversight structure that anchored the 2019 capped-profit framing was thinner than the original documents described.

8. 2024. OpenAI publicly discusses plans to convert from capped-profit to a conventional for-profit structure with no profit cap. The conversion (still in progress as of this writing) would unwind the structural commitment that the 2019 announcement was built around.

What changed in concrete terms

Three things drifted, fast, across that window. Who got to access the work: pricing pages that started at “free for research” ended at per-token consumer billing. Who got to set the prices: investors with profit motives replaced foundations and academic boards as the dominant shareholders. Who got to decide what the technology was used for: revenue contracts, including some with categories the original principles documents had named as off-limits.

The drift wasn't a betrayal in any single moment. It was a series of small steps that, taken together, produced a posture the original mission documents would have called something other than democratization. None of the dates above are the cause; they're the receipts.

The other half of the pivot is the training-data question above. Name the transfer plainly. Value flowed from the people who made the work, to the platforms that scraped it, to the customers who now pay to access models built from it. The people in the first column did not consent and have not been paid. The people in the third column are sometimes the same people in the first column, paying for access to a service whose underlying material they themselves produced. That is a strange shape for an industry that was, on paper, about benefit.

The “utility” framing, and what utility actually requires

The current pitch is that AI is the next utility. The next electricity. The next inevitable layer of infrastructure that everyone uses without thinking about it. There's a real argument for that future, and the part of the argument that's honest is worth taking seriously: ubiquitous AI may be coming whether or not anyone wants it to.

The absurdity is what's being skipped over in the framing. Utilities come with norms. A water company that resold your usage patterns to advertisers would be sued out of existence. A power company that locked your stove behind a sign-up page would be the subject of a Senate hearing. An AI company that does the rough equivalent of either is described as innovative.

For the “utility” framing to be honest, four conditions would need to hold. They're short and falsifiable, on purpose:

• Universal access at the basic tier. No per-token consumer pricing for the entry-level model; the equivalent of a public-utility lifeline tier.
• No surveillance built into the meter. No cross-session profile of what an individual used the tool for, ever, regardless of business need.
• Open weights or a genuinely free tier with comparable capability. Not a free trial that lapses; a permanent free tier whose ceiling is documented.
• Compensation paths for training data. Some defensible answer to the question of what the people in column one are owed, beyond “the courts will sort it out.”

Those four are a high bar. The companies in the timeline above are mostly not clearing them. The door is still open, though. Free, no-login, no-tracking is unusual because the easier revenue paths involve doing the opposite.

What icandothat uses AI for, and what we ruled out

We're a tiny site, not a movement. We just want to build a tool that doesn't take from the people using it. icandothat uses AI to identify items from photos, to look up sales comps, and to draft a starting version of your listing. Pretending we don't would be silly.

What we don't do: sell your data, share it with anyone outside the AI providers in the request loop, build a profile of you across visits, run a free trial that turns into a paid one, or put the tool behind a sign-up.

What we do: keep your inputs (photos, descriptions, notes) on our storage for up to six months so we can make the tool better for the next person. After that, they're deleted. Nothing is tied to a person, because no account exists to tie it to.

For context against the timeline above: icandothat LLC was founded in September 2025. Everything you've read on this page came after the dates listed, not before. The values position isn't a retrospective justification of older choices; it's the founding stance, written down before there was a product to defend.

We'd rather you know the real shape of that than read a vague reassurance. We're not anti-AI; we're against how it got sold.