In April, Runpoint adopted EOS, the Entrepreneurial Operating System, a planning method popular with owner-run companies. Its basic unit is the “rock,” one of a handful of goals for the quarter, each owned by a single person and judged at the end of ninety days as done or not done. My partner Matthew Hall admitted in that planning meeting that he had “only heard this term four days ago.” In the same meeting I proposed that we fork it: “create our own version of that. That is not called EOS, but it’s the AI version of that.” When we asked an AI assistant to help restructure our goals, it kept rewriting our rocks as OKRs, a goal format it had presumably seen more of. “It doesn’t need to be like theirs,” I told it.
We set our rocks for the third quarter at a summer offsite. By August we had abandoned most of them. The ones that lasted changed shape along the way. A referral program turned into a client gift program, and a paper about owning your own data turned into a list of the companies we wanted to work with. By late September I was ready to admit the rocks hadn’t worked. EOS assumes the work under a goal takes most of a quarter. With AI, a first draft of almost anything takes a night, and the tools themselves changed several times while the rocks sat on the wall. We switched to monthly goals under quarterly themes.
On October 5 we dropped those too, in favor of what Matthew calls bets and commitments. A bet is a short written case for trying something. A commitment is a promise small enough to keep or break inside two weeks. You can keep the commitment and still lose the bet, and that’s fine. Breaking the commitment isn’t. The conversation later in this issue has Matthew explaining it better than I can.
My own last rock was writing the firm’s operating manual, the document that describes how we work with clients and with AI. I wrote it the day after it was assigned, and at the October retrospective I offered that as evidence that a quarter is too long. Matthew asked how much we cared about polish, since I had made it the next day but it had not been done the next day. He was right. Finishing it took three or four weeks, and at that same meeting none of us could say for sure whether anyone was reading it. The draft took a night. Everything that depended on other people took a month, and some of that still isn’t finished.
Which raises a fair question about the thing in your hands. We gave up quarterly goals and then launched a quarterly magazine. I think both decisions are right, for the same reason.
The quarter is too long for a commitment. A ninety-day promise mostly measures how long you can put something off, and by month two the promise is stale. A quarter is about the right length for judgment, though. An investor who advises our firm warned us in June that “you’ve got to stick with stuff long enough, and I don’t know when that is.” Matthew made a similar argument in August about holding the wheel long enough for our decisions to take effect. Neither of them was arguing for two-week thinking, and I’ve come around to the view that they were describing a different clock. Where I’ve landed is to commit in short, reversible pieces, look hard at our assumptions about every ninety days, and hold direction for years.
A quarterly is a filter for that middle clock. Anything printed here has to still be true in ninety days, after several model releases and a few hundred announcements. Most of what I read about AI in a given week wouldn’t pass, and most of it was never meant to. An announcement changes what a model can do. It rarely changes what a company should do this quarter, because the part that pays takes longer: getting the data into one place, writing down how the work is actually done, rolling a tool out carefully enough that people trust it. Our operating manual took a night to draft and a month to finish, and that ratio holds for most of what we build with clients. The trouble starts when an executive plans off the weekly material and ends up rewriting the plan every week. That is more or less what happened to our rocks.
This magazine began as a book. Late in 2025 I wrote one with AI, got seventeen chapters in, and realized nobody would ever read it, including me. I threw it away and we rebuilt it as a magazine. That first digest went out last winter, and when we talked about it on our company podcast in January, I said we would try to publish quarterly. The second issue made it onto our website and never onto paper, which is its own small lesson about quarterly commitments. So the number on this cover is two, though nearly everything inside it is new. The page is larger and the pieces are fewer and longer, each chosen because we expect it to hold up for at least a season.
The lead essay, “Half-Lives,” argues that the most useful executive skill right now is sorting what you know by how fast it goes stale, and checking each kind on its own schedule. Matthew and I then argue about planning in “Stale by Month Two,” including how short a cycle can get before it stops being useful. “Group Drive” goes back to factory electrification and the forty years it took to show up in the productivity numbers. After it we list what we are assuming for the next four quarters, with the evidence that would change our minds. “Arm the Few” makes the case for giving AI to a small number of people first, and Ryan Mish, an operator engineer at Runpoint, follows with “When Everything Becomes Buildable,” on why finishing is now the scarce skill. “Secondhand” starts from Andrej Karpathy’s observation that most executives know AI through other people’s reports, and asks what to do about it. The Ninety-Day Review is a set of questions for your leadership team to ask itself each quarter. “The Turning Radius” argues that mid-sized companies are better placed than anyone to rebuild around a fast technology: small enough to rewire themselves, large enough for it to matter. A short reading list follows it. In “The Down Payment,” Matthew talks with Alvin Wong, co-CEO of Advanced Aircrew Academy, about what AI amplifies and what it costs to keep what you build, and Questions We Get closes the issue.
We made our first two-week commitments on October 5. By the time this reaches you, several rounds will have come due. I don’t know yet whether two weeks is the right length or only the shortest one we could think of, and the investor’s question about how long is long enough is still open. I’m less worried about it than I was in August. Each version of our plan this year has been shorter and more honest than the one before, and the slow work underneath it has kept moving the whole time. In the next issue I’ll report which commitments we kept, and what we learned from the ones we didn’t.