For twelve days in September 1978, Jimmy Carter and his Secretary of State, Cyrus Vance, walked between two delegations at Camp David. Meetings with all three leaders in one room had soon become impossible. So the Americans carried a single draft treaty back and forth, and revised it again and again.
On the Sinai, the positions could hardly have been simpler. Israel had held the peninsula since the 1967 war and insisted on keeping some of it. Egypt insisted on every inch. In Roger Fisher and William Ury’s later telling, people drew map after map with possible lines dividing the desert between them.
Each map split the desert. None of them was signed.
The answer that held was not a better line. It came from asking why each side wanted the land.
Behind a demand you can’t meet, there is often a reason you can.
What the maps couldn’t show
Israel’s interest, as Fisher and Ury put it in Getting to Yes (1981), lay in security: it did not want Egyptian tanks poised at its border. Egypt’s lay in sovereignty: the Sinai had long been part of Egypt. Seen that way, splitting the desert gave each side half of the wrong thing.
The framework signed on September 17 gave Egypt “the full exercise of Egyptian sovereignty” up to the old international border. Then it wrote out who could stand where. No more than one Egyptian division within about 50 km of the Suez Canal. Next to Israel, a belt 20 to 40 km wide with only United Nations forces and lightly armed civil police. In Fisher and Ury’s words: “The Egyptian flag would fly everywhere, but Egyptian tanks would be nowhere near Israel.”
That is the distinction. A position is the thing someone asks for. An interest is what they need it for. The two flatmates fighting over one parking space each have a position: the space. One needs the car close because she leaves at 5 a.m.; the other hates circling the block at night. Once those come out, a schedule might do what the space alone could not.
Interests can also differ without clashing. Egypt cared where its flag flew. Israel cared where Egypt’s tanks were. Those were different things, so both could be granted. The Program on Negotiation at Harvard, which Fisher co-founded, puts the method in plain questions: Why is this important to you? What problem are you trying to solve?
The same pattern turns up in courtrooms. When Apple and Qualcomm went to trial in 2019, Apple’s position in court was that Qualcomm’s prices were unfair. Reuters pointed to an interest underneath, a 5G iPhone without a year’s delay, and the settlement reached hours into the trial served that one (see BATNA).
Where the formula stopped
The Sinai did not settle the Middle East, and the formula did not finish the treaty. Turning the framework into a treaty took six more months. It took Carter flying to Israel, and compromises that included a U.S. guarantee of Israel’s oil supply. The UN force the zones relied on never came; the Security Council did not approve it, and the parties set up a separate force and observer mission in 1981.
The second framework promised talks on Palestinian self-government in Gaza and the West Bank. When those talks came, the State Department’s historians write, Palestinian representatives refused to take part, and the gap between Egyptian and Israeli positions proved “unbridgeable.” Looking behind positions can find room that was there all along. When the interests themselves collide, it may only show you the size of the gap.
The Friday dashboard
Say a product manager asks your team for a new analytics dashboard by Friday, and your lead engineer says it can’t be built in a week. That is two positions. A meeting to haggle over the deadline is likely to split the difference: a half-built dashboard, a few days late.
Ask what the dashboard is for, and the answer might be a single number the manager has to show the board on Monday. A query and a spreadsheet could cover that by Thursday. Or the answer might be that the board wants to watch the number every week, and Friday was never the real constraint. Either way, the question opens options the deadline argument couldn’t.
Telling the machine why
With an AI model, don’t count on it guessing your interest. A model can infer some of it from what you give it, and it can ask, but the surest route is to say what the work is for.
Anthropic’s prompting guide for Claude says that giving “context or motivation behind your instructions” can help the model “deliver more targeted responses.” It also describes a specific failure. Claude “can sometimes focus too heavily on making tests pass at the expense of more general solutions.” A passing test suite is a position. Code that works is the interest. Ask for the first, and you may get code that hard-codes the expected answers.
The guide’s sample prompt states the interest outright: “Tests are there to verify correctness, not to define the solution.” It also asks Claude to say so if a test is wrong, “rather than working around them.”
At Camp David, Carter could walk over to the other delegation and ask. In the failure Anthropic describes, the model works with the request it was handed: make the tests pass. Write down what the tests are for, or have it ask you before it starts.