In 1982, General Motors closed its assembly plant in Fremont, California. The workforce there had a reputation that the man who ran the UAW’s western region later called “well earned”: chronic absenteeism, sabotage on the line, cars leaving the factory with defects built in.

Two years later the plant reopened as NUMMI, a joint venture with Toyota. More than 85% of the workers were the same people. However, one rule shows the difference. At GM, you never stopped the line. At NUMMI, any worker could pull a cord strung along the line, and if the problem could not be fixed in time, the line stopped. Within three months, according to auto analyst Maryann Keller, the cars coming off it were earning near-perfect quality ratings.

It took GM years, a failed attempt at one plant, and a massive loss in 1992, to take NUMMI’s system company-wide. When it finally installed cords in other plants, some workers who pulled one got yelled at. A few plants cut the cords down. Everyone NPR’s Frank Langfitt interviewed said it took about 15 years after NUMMI “for change to even begin to take hold at GM.”

What made the cord work at NUMMI was having the permission to pull it.

Publishers are hanging cords all over the building right now, approving AI tools for newsroom, commercial and operations teams. Inside the organizations I work with, who is allowed to act on what those tools produce, and who answers when the output is wrong, have barely moved. Michelle Manafy wrote here in April that the leadership test is now “where, how and to what end.” Those questions should be answered by the person who has the authority to make the call, the one who will be accountable for the result. That authority should be established in advance, rather than requiring approval each time.

Trained, and still unsure

Tess Jeffers, the Wall Street Journal’s director of newsroom data and AI, described the problem here last October. After an enterprise rollout and company-wide training, “many reporters remained unsure how to use these tools effectively.” The Journal’s answer was a summer series of lunch-and-learn sessions where colleagues showed how they already used the tools. Jeffers writes that “with the editor in chief in attendance, the sessions delivered a clear endorsement that the newsroom can and should be using these tools.”

That last part is permission, and it came from the top. Training covers how. On its own, it says nothing about whether you are allowed. Until someone with authority says so, people wait.

What stays human

I ask teams building AI tools one question. When your CEO sees this working, what reason do they have to keep you? A senior analyst at a professional services firm went quiet, then said the only part she would never hand over was the call she makes when the numbers look right but the client doesn’t.

I call that the Human Remainder. It is the part of the job that matters more once the machine has taken the rest. In a newsroom it is the call that depends on knowing a source, the decision about whether a story is fair, and the name on the byline. That is where human authority and accountability belong. Most of the rest can move without anyone asking.

The name on the byline

In September, as Futurism reported, Cleveland.com government reporter Kaitlin Durbin posted that a story had run under her byline while she was on her honeymoon. “I did not write or review a word of it and it does not pull info from past reporting,” she wrote. “Does Cleveland.com think they just own my name now?” The story now carries only an Express Desk credit, a desk the paper says uses AI tools to assist in producing content that is reviewed and edited by staff. Editor Chris Quinn called it “a simple mistake, the result of a miscommunication between a couple of editors,” and said it was corrected as soon as the paper learned of it from her post.

The paper fixed it within hours of her post. But until then, the name on the story was hers.

This is one reason adoption stalls. The organization keeps the speed and the savings, and the person whose name is on the work keeps the mistake.

Who decides

If management does not decide who owns these calls, a union or an arbitrator will. Beginning in August 2024, Politico’s union filed grievances over two AI tools launched without the advance notice and human oversight its contract required. In November 2025 an arbitrator ruled against the company, and in May 2026 the union announced that Politico had agreed to shut down one tool and not revive the other. At ProPublica, unionized staff walked out for a day in April, in what Nieman Lab called the first strike at a major U.S. newsroom fought at least in part over AI protections. Their first contract, ratified in July, creates an AI committee where, in the union’s words, bargaining unit members “will have a say in issues related to technology.”

At The New York Times, Semafor reported in August that the paper is testing AI search summaries with a small subset of visitors, calling it “the news arm’s first test of AI-generated text that is unmediated by a Times journalist or editor.” The Guild, which protested outside the paper’s 175th anniversary gala in September, is pressing to get AI guardrails, including the paper’s existing policies, into its contract. Managing editor Marc Lacey told the Guild by email in March that management believes “it’s imprudent to focus on adding restrictions into a contract given the rapid and constantly changing nature of this technology and its use cases.”

A few publishers are working this out themselves. The Guardian’s editorial code says significant use of generative AI in journalism “must be explicitly approved by the relevant senior duty editor.” At the BBC, Responsible AI Lead James Fletcher points to how editorial decisions already work: “thousands of people make editorial decisions every day about reasonably significant risks,” by and large on their own, turning to a central team when they need support. That, he says, is the model the BBC wants to extend to AI.

To give people a cord they will actually pull, start with four moves:

• Put the sign-offs on the Human Remainder, and document the uses that do not need one.
• Name, by role, who owns each call that remains.
• Move accountability with permission, so the organization owns the first good-faith mistake with a tool it approved.
• For every new tool you approve, remove one existing approval step, so permission gets simpler as tools multiply.

The fight inside the building

High-stakes work will still need an editor. NUMMI’s lesson holds all the same. A cord nobody is allowed to pull is decoration. Publishers have spent nearly three years fighting over who gets to use their journalism. The next fight is inside the building. It may seem smaller, but it has enormous consequences: Who owns each decision the machine touches? The publishers that answer it first will be the ones whose tools get used.

---

About the author

Ramy Nassar is a futurist, speaker, and strategic advisor who works inside organizations on how AI gets adopted, across financial services, life sciences, retail, and professional services. He was previously Head of Innovation at Mattel. He is also the father of three boys, so the question of what AI does to work and agency is one he brings home every night. His new book, The World After AI: Why the Future Is Still Ours to Shape, will be published in April 2027.

More about Ramy at ramynassar.com