In the spring of 2023, a mid-sized SaaS company I know well locked in their product roadmap through Q2 2025. They were methodical about it. They surveyed customers. They mapped competitor moves. They hired a VP of Product who had done this exact exercise at two prior companies and both times it worked. Eighteen months felt like a reasonable horizon. It always had been.
By October 2024, three of their five roadmap pillars were already obsolete. Not behind schedule. Not deprioritized. Obsolete. The capabilities they had planned to build as differentiators had been folded into foundation models as commodity features. The problem they were solving had either been solved upstream or dissolved entirely because the underlying workflow it depended on no longer existed in the same form. They shipped anyway, because what else do you do when you've already allocated the engineering budget.
This is not a story about bad strategy. It's a story about a broken instrument. The strategic planning frameworks most companies operate on were built for a world where 18 months was visible. That world is gone.
What the Planning Horizon Actually Assumed
The math works like this: traditional strategic planning borrowed its logic from capital allocation cycles. If you're building a factory, you need 24-36 months of visibility because the physical infrastructure commits you. Software companies inherited the cadence without the physical constraint, but they kept the timeline because it matched investor reporting cycles, hiring pipelines, and the natural rhythm of how product teams actually build things. Eighteen months became the default not because anyone proved it was optimal but because it fit the organizational calendar.
Underneath that timeline sat a quiet assumption: that the competitive and technological landscape would shift predictably enough that you could extrapolate from current trends. You didn't need perfect information. You needed enough signal to make reasonable bets. And for most of the 2010s, that assumption held. The cloud transition was fast but it was directional. Mobile was disruptive but it was legible. You could watch the trendlines and triangulate.
What most people miss is how much of strategic planning was actually pattern matching dressed up as analysis. Executives weren't predicting the future. They were betting that the future would rhyme with the recent past, and they were usually right enough that the framework survived. The 18-month horizon wasn't a window into what was coming. It was a confidence interval built on the assumption that change happened at roughly the same speed it always had.
That assumption collapsed somewhere around late 2022 and has not recovered.
The pace of capability change in foundation models specifically has broken the extrapolation logic. When GPT-4 shipped in March 2023, the professional consensus was that reasoning tasks and multimodal work were 18-24 months away from being practically useful. That consensus was wrong by roughly 18 months in the optimistic direction. When people make predictions now, even careful, well-informed people, they are operating with error bars wide enough to drive a logistics fleet through. The signal-to-noise ratio in technology forecasting has deteriorated to the point where the forecasts are more useful as documentation of current assumptions than as actual guides to future decisions.
The Mismatch Nobody Wants to Name
Here is the uncomfortable part. Most organizations know this at some level. Ask any product leader off the record about their 2026 roadmap and they will tell you it's essentially fiction beyond the next two quarters. They know the assumptions are fragile. They know that a single model release could reframe the entire competitive landscape before their next board presentation.
But they are still running the same planning process. Still producing the same 18-month documents. Still hiring against headcount projections that assume a competitive environment that may not exist by the time those hires are fully ramped. The process continues not because anyone believes it produces accurate outputs but because the organization doesn't know what to replace it with. And because the people who run the planning process have careers built on running the planning process, which creates its own kind of institutional inertia.
Stewart Brand had a useful framework for this. He wrote about how different layers of civilization change at different speeds... fashion fast, governance slow, with culture and infrastructure somewhere in the middle. The implication was that stability comes from the slower layers anchoring the faster ones. What's happening now is that one of the fast layers, technology, is moving faster than the organizational layer can process. The org chart is a slower layer. The planning calendar is a slower layer. They cannot update fast enough to track what the technology layer is doing, and the gap between them is where strategy goes to die.
The companies that are quietly doing well in this environment are not the ones who solved the forecasting problem. Nobody solved it. They're the ones who stopped pretending the 18-month window was real and reorganized around shorter feedback loops. Notion's product team has talked publicly about running what amount to 6-week decision cycles for anything AI-adjacent. Linear, the project management tool, ships so continuously that the concept of a roadmap is almost architectural... it describes values and directions rather than features and dates. These aren't post-hoc rationalizations of chaos. They're deliberate responses to a planning environment where the old instruments don't work.
The analogy I keep coming back to is dead reckoning. Before GPS, sailors navigated by estimating position based on known speed, heading, and time elapsed. It worked well enough when conditions were stable and the errors were small. When conditions became unpredictable, errors compounded and dead reckoning became actively dangerous because it gave you false confidence in a position that was wrong. The answer wasn't better dead reckoning. It was a different navigation system entirely.
What most planning teams are doing right now is sophisticated dead reckoning in a storm. The math is careful. The process is documented. The outputs are plausible. And the actual position of the ship is somewhere nobody predicted.
What Comes After the Blindspot
I want to be careful not to turn this into a consulting pitch for agility frameworks, because most of those are just slower planning processes with better branding. The honest answer is that nobody has a clean solution to this problem yet, and anyone who tells you they do is selling something.
What I think is actually useful: separate the decisions that require long horizons from the ones that don't, and stop applying the same planning process to both. Infrastructure investment, key hires, core architecture... these require longer commitments and the uncertainty is acceptable because you're buying optionality, not predicting outcomes. Feature roadmaps, pricing strategy, positioning relative to AI-native competitors... these are operating in the fog and should be treated accordingly. Short cycles. Small bets. Reversible decisions where possible.
The companies that will look smart in 2027 are not the ones making the most accurate 18-month predictions right now. They are the ones who have quietly admitted to themselves that the predictions don't work and have built organizations that can respond rather than anticipate. Patti Smith didn't plan Horses. She paid attention to what was alive and followed it. That's not a strategy. But right now it might be more useful than one.
The math works like this: if your forecast error is larger than your planning horizon, you don't have a strategy. You have a story you're telling investors while you figure out what's actually happening. Most companies are at that point right now. The ones who admit it first are the ones with a real chance to adapt.