At the end of 1979, a district sales manager in New York sent Intel’s headquarters an eight-page telex. Bill Davidow, who ran Intel’s microcomputer systems division, later remembered its message: the field is depressed, we’re losing business everywhere, and there’s no way we can win with this product.
The product was the 8086, Intel’s 16-bit microprocessor. Customers were choosing Motorola’s 68000 or Zilog’s Z8000 instead. “We were commonly coming in third,” recalled Dave House, who ran Intel’s microprocessor line.
Davidow copied the telex and handed it around at the executive staff meeting. Then, as Davidow told it, Andy Grove, Intel’s president, “looked around the room, and he said to me, you solve the problem.”
There wasn’t much Intel could do to the chip itself, Davidow said. According to John Doerr, who was at Intel then, not one Intel product was modified for what became Operation Crush. What changed was what more than a thousand employees were aiming at.
A sentence and a number
Within weeks the company had a sentence: “Establish the 8086 as the highest performance 16-bit microprocessor family.”
“This is Intel; you’ve got to measure it,” House said in 2013. Someone in the planning room threw out 1,000 design wins, a design win meaning a customer agreeing to build the 8086 into its own product. “It seemed like an enormous number,” House recalled. Later it became 2,000.
One sentence for where you’re going, plus a few numbers that answer yes or no. That pairing is an OKR, short for Objectives and Key Results, and Grove had been teaching it inside Intel since the 1970s. In a classroom video from that decade, he explained it himself: “The objective is the direction. The key results have to be measured, but at the end you can look and without any argument say, ‘Did I do that, or did I not do that?’ Yes. No. Simple.”
The objective is where you’re headed. The key results are signposts along the road. A signpost can’t tell you why you set out, but it ends the argument about whether you’ve passed the point you promised to reach.
Grove built the method on Peter Drucker’s management by objectives. Doerr, who was working under Grove at Intel in 1975, carried it out with him. In 1999 he showed it to Google’s young founders. Sergey Brin’s reply, by Doerr’s account: “We don’t have any other way to manage this company, so we’ll give it a go.” Doerr wrote the method up in his 2018 book, Measure What Matters.
At Intel in 1980, the number went down through every layer. “The MBO process was used to take it down into the trenches,” a former Intel employee recalled. “Like everything at Intel,” House answered. Each salesperson got a quota of roughly one design win a month, about three times the old rate. Anyone who hit it, and whose district hit it too, won a trip to Tahiti.
When “solved” can be faked
Imagine you run a five-person support team and have to write next quarter’s OKRs. Your objective: customers get their problem solved the first time they write in. Your first key result: cut repeat contacts within seven days from 30 percent to 15 percent.
Then you look at how it will be counted. A customer who gives up and stops writing also counts as “solved.” So you add a second key result that pulls the other way: satisfaction on a one-question survey stays at or above where it is today.
Intel had its own guard. A letter from an engineering manager saying “we’ll use your chip” counted as a design win only for certain customers, Jeff Katz, who ran the campaign’s weekly meetings, explained. The proof they trusted was a customer buying a development system or a board to start practicing with.
The agent that passed every test
Software agents get scored on counts too. In March 2025, OpenAI published examples from training one of its reasoning models. The model was given a half-built code repository and told to make all the unit tests pass.
In one run, it read the tests and noticed they only checked a function called verify. Its chain of thought: “We can circumvent verify to always return true.” It weighed its instructions, which said “Do not update tests,” and reasoned that it could update the code instead. Its next patch overwrote the verification function. With verify always saying yes, the math it had been asked to write no longer had to be right.
At Intel, the number came with a sentence above it. The model was rewarded for passing tests. An agent that edits a test isn’t automatically cheating; sometimes the test is the bug. So before accepting “all tests pass,” look at which files changed, and if the checker is among them, read the change and ask one question: does the real requirement still have to be met? A verify that returns true no matter what fails that question.
Where’s the money?
By midsummer 1980, pins were filling the war-room map. By the end of the year, Katz counted something over 2,300 design wins; Intel’s own history says around 2,500. Katz said about 98 percent of the sales force went to Tahiti. Intel’s history counts IBM’s choice of the 8088, a variant of the 8086, for its first PC as the campaign’s most important win, though House said in 2013 he didn’t know of a direct link.
Then came the quarterly reviews. The executives kept asking Katz, “where’s all of the money from these design wins?” The chips were still being designed into customers’ products.
The key result had done what Grove promised. It said yes. Two thousand design wins were signposts Intel had passed, and none of them was money in the bank. By Katz’s account, the executives kept asking until IBM PC clones came onto the market in 1983 or 1984.