OP #324: Predicting Is Hard. Deciding Is the Job.

Happy Tuesday, OP Community.

Two weeks ago, I published a piece on MidlifeCrisisWatches.com titled "The Watch Internet Has One Prediction for 2026 That Should Make You Nervous." I tracked 12 major prediction outlets ahead of Watches & Wonders, tallied where the crowd was converging, and staked my own contrarian hot take. The show happened last week. I want to tell you how the predictions held up, because what I took from the exercise has less to do with watches than with how we all make decisions at work.

Stay with me through the short watch scoreboard. The second half is about the job.

A quick housekeeping note. Silicon Alley Sports Flagship is exactly one month out (May 18th). Golf is sold out. Pickleball has a handful of spots open. Schmoozer tickets are still available if you want in for food, networking, and the afternoon. Sandy Hook Promise is back as our 2026 Impact Partner.

A few links to get us started this week...

• The Watch Internet Has One Prediction for 2026. That Should Make You Nervous. (MidlifeCrisisWatches)

• Why this playoff run matters more than last year's and next year's (Posting and Toasting)

• Agent-to-Agent VC Meetings Are Coming (GTMnow)

• What the Best Private Equity-Backed CEOs Do Differently (Harvard Business Review)

• The CMO Survival Guide for 2026 (CMSWire)

I hope you enjoy this one. It weaves business with watches - two areas that I have lots of interest.

Be well, do good.
Darren

Predicting Is Hard. Deciding Is the Job.

If you read my prediction piece, you know I did the exercise of cataloging every major Rolex call across the watch internet ahead of Watches & Wonders. Nine sources predicted a Rolex Milgauss return. Seven predicted a Coke GMT. Seven predicted a Pepsi discontinuation. Six predicted a Day-Date 70th anniversary edition. Five predicted an Explorer II refresh. My own contrarian call, loudly made: a moonphase complication on the Rolex 1908.

Here's the scoreboard. Milgauss: no. Coke GMT: no. Explorer II: no. 1908 moonphase: no. Day-Date in Jubilee Gold: partial credit. Yacht-Master II return: nobody saw it coming.

The people I was quoting are pros. They track patents. They have dealer relationships, forum sources, and CEO interview access that most of us don't. They had every data point available. They still called it mostly wrong. So did I.

That's not a story about bad analysis. It's a story about what happens when you try to predict a decision being made by other humans inside an opaque organization. No amount of information closes the gap to certainty. The ecosystem looks smart right up until the moment the actual decision lands. Then suddenly nobody called it.

Same dynamic in our world.

In every portfolio company I spend time with, there's a version of this exercise running. Will this GTM motion work. Will this price increase hold. Will this campaign land in the category we think it will. Will this senior hire stick after 90 days. Will agentic AI adoption stall or compound. The honest answer to all of these is: you cannot know. There is no additional slide deck that will get you to 100%. The one you already have is probably close to the ceiling.

The most common failure mode I see at senior levels isn't bad decision-making. It's deferred decision-making. Someone asks for one more analysis. One more data point. One more stakeholder conversation. I shake my head. I’m sometimes guilty of it too. The deck keeps getting polished while the window closes, and the decision gets made for you by default.

Colin Powell had a rule. Make the call when you have somewhere between 40 and 70% of the information. Under 40 and you're gambling. Over 70 and you're late. The opportunity has moved. The market priced in the move you were considering while you were asking for another tab on the spreadsheet.

Jeff Bezos drew a related line. Most decisions are Type 2: reversible, two-way doors. You walk through, you see what happens, you walk back if it didn't work. Those should be made fast with partial information. A small number are Type 1: irreversible, one-way doors. Those get the full treatment. The trap is treating every Type 2 decision like it's a Type 1, which makes us slow at exactly the moments where speed is free.

The executive skill isn't being right more often. It's calibrating your own confidence honestly, knowing which kind of door you're walking through, and being willing to move before the picture is complete. Good operators are comfortable being wrong 20 or 30% of the time because the alternative is being paralyzed 80% of the time.

I was wrong about the 1908. The watch internet was wrong about the Milgauss. None of us had a Yacht-Master II on our board. That's how this goes, in watches and in business. The job isn't to predict. The job is to decide.

I'll take a 60% call made now over a 95% call made six months from now, every time.

OP Links

The Beginning of Scarcity in AI (Tomasz Tunguz)
AI compute scarcity could emerge as early as 2026, with potential to double operational costs for developers and force project prioritization.

Retail Media. Commerce Media. Financial Media. The $200B Shift Third-Party Cookies Made Inevitable. (Abhi Yadav)
Abhi Yadav is a previously cited known voice with genuine practitioner credibility, and the macro reframing of cookie deprecation as a $200B structural shift offers a non-obvious, operator-relevant angle that goes well beyond typical programmatic trend coverage.

Why We're Removing Our Programmatic Ads (Prospect.org)
The contrarian first-person decision to exit programmatic entirely provides a genuinely non-obvious POV on the industry's structural flaws that a marketing operator at a PE-backed media or content business would find thought-provoking, even if the author is a journalist rather than a seasoned operator.

AI-enabled GTM is the new value creation lever for private equity (Clay)
AI-powered GTM tools are shifting how PE firms approach go-to-market, enabling account scoring, personalized campaigns at scale, and measurable improvements in pipeline efficiency across portfolio companies.

Rising AI Adoption Is Driving Up Enterprise Costs (PYMNTS)
Enterprises face unexpected cost explosions from usage-based AI pricing as adoption scales, with changes potentially doubling or tripling costs for heavy users like Uber.

LLMs are becoming commodities (RunLLM)
As model quality converges across providers, LLM differentiation now depends on finding niches and specialized use cases rather than competing on general-purpose capabilities—a shift forcing startups to rethink their go-to-market.

2026 Private Equity AI Radar (FTI Consulting)
95% of PE funds report AI initiatives meeting or exceeding business cases, but adoption remains low and performance gaps are large, with most firms struggling to scale from isolated successes to enterprise-wide advantage.

YC W26 Batch Breakdown: Deep Dive on 199 Companies With Founder Data (Extruct AI)
YC W26 represents the deepest-tech batch in recent memory, with founders younger but more technical, drawn from Tesla and SpaceX, building genuinely hard-to-replicate businesses that signal a return to YC's historical roots of funding ambitious deep-tech founders.

Thank you very much for subscribing to the OP Letter. I hope you enjoyed this letter - and if you have any feedback, I’d love to hear it. Just hit Reply and drop me a note.