The fine print is the product now

Every product decision is a terms-of-service decision now. The feature that ships matters less than what you agreed to by using it.

This week made the point three different ways. The Writers Guild of America and the AMPTP reached a tentative four-year deal that expands the AI protections writers won during the 2023 strike, including new provisions around licensing writers' work for AI training. The deal landed a full month before the existing contract expires. No brinksmanship this time. The WGA showed up early, knew what it wanted, and got it on paper.

Compare that with what happened to developers. The Register reported that starting 24 April, GitHub will use code snippets, prompts, file context, and navigation patterns from Copilot Free, Pro, and Pro+ users to train AI models, unless those users actively opt out. The announcement drew 59 thumbs-down reactions and three positive ones. Business and Enterprise customers are excluded, which tells you everything about who GitHub considers a customer versus a product.

The contrast is sharp. Hollywood writers secured explicit protections through collective negotiation. Individual developers got a default flipped on them. Same underlying question (who controls the training data?) but wildly different outcomes depending on whether people organised before the terms changed.

Meanwhile, the scale of what's at stake keeps growing. Bloomberg reported that 92% of US developers now use AI tools daily and 41% of global code is AI-generated. The vibe coding wave has gone from novelty to default workflow. But here's the number that should make every developer read the fine print: 74% report productivity gains, while 63% have spent more time debugging AI-generated code than writing it themselves. The tools are useful enough to be indispensable and unreliable enough to demand constant attention, which makes the question of who owns what you produce with them all the more pressing.

The negotiation gap

The pattern here isn't really about AI. It's about who shows up to negotiate.

The WGA had leverage because it could credibly threaten to stop working, as it proved in 2023. Individual Copilot users have no equivalent. They can opt out, sure, but opting out of the tool that 87% of Fortune 500 companies have adopted means opting out of the default workflow at most workplaces. GitHub knows this. The opt-out mechanism is technically available but practically decorative.

I think the lesson for anyone building on AI platforms is uncomfortable but clear: the fine print is the product now. The model weights, the latency, the context window are table stakes. What actually determines whether a tool works for you is buried in the terms of service. How your data gets used. Whether your inputs train the next version of the thing you're paying for. Whether the vendor can change those terms with a blog post and a 30-day countdown.

The WGA understood this. Its members negotiated training data licensing into the contract alongside compensation and credits. They treated the legal terms as seriously as the creative ones.

For the rest of us building on third-party AI, the question is whether we'll do the same, or whether we'll keep scrolling past the terms update email until the defaults have already decided for us.

---