In October 2024, a mid-market SaaS founder I know was in a board meeting when someone asked a reasonable question: "Where will AI capabilities be in 18 months?" The room went quiet. Not awkward quiet. The quiet of people realizing they had no framework for answering.
He told me this story over coffee last month. Not as a war story. As a confession.
The Setup
Meet Sarah. Not her real name, but real founder. Series B, 40 people, $8M ARR, operating in the workflow automation space. Her product sits between human workers and AI execution... which means every quarterly planning cycle has become a game of chess against an opponent that rewrites the rules between moves.
Sarah's company does something specific: they help mid-market operations teams route work through AI systems. Think intake forms that get smart-sorted, documents that get auto-categorized, approvals that get flagged by exception. Boring work. Essential work. The kind of work that was always going to be the first thing AI ate.
For three years, Sarah's positioning held. "We're the human filter. We keep AI honest." Her margins were 72%. Her customer retention was 94%. She had visibility into 2024, reasonable sight lines into 2025. Then late 2023 hit, and the ground shifted so fast that by mid-2024, every strategic assumption she'd built the company on was in question.
Not wrong yet. But questioned.
The Problem
Here's what most people miss about AI adoption in B2B: the problem isn't that AI is advancing too fast. The problem is that visibility into what you should build has become impossible to price.
Sarah went into 2024 with a roadmap. Feature releases planned through Q4. A hiring plan based on the assumption that her core value (human review layers, exception handling, workflow optimization) would remain defensible for 18-24 months. Investors were happy. The math worked. She had time.
Then three things happened in rapid succession.
First, in spring 2024, Claude 3.5 and then GPT-4o showed that context windows and reasoning had jumped a tier. Her customers started asking: "Can we just... feed the whole document set to Claude and let it learn?" The honest answer was yes. But yes meant her review layer might not be necessary for simple cases.
Second, three competitors in her space pivoted to "AI-native" positioning. They stopped selling the human-review angle and started positioning as orchestration layers... which is what Sarah was building. But they said it louder.
Third... and this is the hard part... her largest customer (14% of ARR) hired an in-house ML person and started building their own version of what Sarah was selling. They were still paying her. But they were also hedging.
Now Sarah faced a structural problem that no amount of roadmap planning fixes: the 18-month vista had collapsed to 6 months. And even that 6-month view was cloudy.
When visibility collapses, everything that depends on it gets fragile. Hiring plans. Feature priorities. Pricing strategy. Go-to-market angles. Sales cycles that used to close in 90 days started stretching to 150 because customers couldn't articulate what they actually needed. (How can they? They don't know what AI will do next quarter, either.)
Sarah's board wanted to know: What's the plan? And the honest answer was... there isn't one that survives contact with reality.
What They Did
Here's what Sarah did, and it's almost boring because it's the opposite of what every AI-era founder tries to do.
She stopped planning 18 months ahead. She planned 8 weeks.
Not from fear. From clarity. If the future is unknowable, the rational response isn't to predict harder. It's to build the thing that remains valuable regardless of what the future does.
She cut the roadmap in half. Literally. And rebuilt it around a single question: "What would we want to sell if Claude and GPT both became free tomorrow?"
That question changed everything.
It forced her to separate the "AI capabilities we're bundling" from the "actual value we're delivering." And when she separated them... the answer became clear. The AI part was commoditizing fast. But the workflow routing, the exception detection, the integration layer, the institutional knowledge about which processes could even be automated... that was still scarce. Still valuable. Still defensible.
So she made three moves.
First, she reframed her product positioning from "human-review layer for AI workflows" to "process intelligence platform." Not hiding the AI. Just centering what was actually hard to replicate: the mapping of real business processes, the taxonomy of exceptions, the institutional memory of how work actually gets done at her customers' companies.
Second, she changed how she built. Instead of trying to stay ahead of AI capability waves (impossible), she optimized for flexibility. Her code architecture got refactored to swap out AI providers like lightbulbs. Claude for this. GPT for that. Open-source models where they fit. No religious attachment. Just... what executes best for this customer's specific problem?
Third, she restructured how she sold. Instead of selling a product, she started selling outcomes. "We map your process, we instrument it, we handle the exception cases, and we give you visibility into what the AI actually does." The price moved from per-user to per-process. Her largest customer? Still paying. More, actually, because the model changed from licensing to outcome-based pricing.
The hiring plan stayed flat. She wanted senior people (people who could think in systems, not just code features). She bet smaller on product roadmap, bigger on customer implementation and process mapping.
All of this feels conservative. It is. That was the point.
What Happened
Sarah's 2025 numbers came in stronger than the board's 2024 projections. Not wildly. But solidly. ARR grew 34%. Churn dropped to 91% (retention improved). Margins actually grew because outcome-based pricing meant she was capturing value she was already creating.
More important: she regained sight lines. By playing for defensibility instead of velocity, by centering process intelligence instead of AI capability, she could see 12 months ahead again. Not clearly. But clearly enough to make hiring decisions. Clearly enough to push back on what customers asked for. Clearly enough to say no.
Her largest customer didn't leave. The ML hire they brought in? The person actually started using Sarah's platform more, not less, once they understood what it did. Because what it did wasn't "be AI." It was "make our actual process legible."
One of her competitors who went all-in on "AI-native" positioning? Raised their Series B, but at a flat valuation. They were more efficient. But they had no defensible positioning. They were riding the wave. Sarah was building the shore.
What I Learned
The first thing: AI visibility collapse is real. It's not hype or founder anxiety. It's a structural condition where the pace of capability change exceeds the pace of predictable business utility. When that happens, planning beyond the near horizon becomes speculation dressed up as strategy.
The second: the winners in this era won't be the ones who predict AI right. They'll be the ones who build things that are valuable independent of which AI wins. Defensible positioning isn't about being smarter than the next model. It's about being wedded to the customer's actual need... which rarely changes as fast as AI capabilities do.
The third: outcome-based pricing isn't clever. It's honest. When you can't predict where capability lands, pricing by capability (per-user, per-call, per-token) becomes arbitrary. Pricing by outcome... that scales with actual value. Sarah discovered this by accident. She might have stayed on the feature treadmill if she'd had board support for it.
The fourth, and maybe the hardest: the default move for founders in uncertain times is to accelerate. Ship faster. Move quicker. Get ahead of the curve. Sarah did the opposite. She slowed down, deepened her customer work, and bet on defensibility over visibility. It felt like losing. It wasn't.
The math works like this: if you're planning for a future you can't see, you're going to spend money on the wrong things. Sarah spent 30% less on product roadmap execution and got better outcomes because her roadmap aimed at things that remain valuable regardless of what AI does next. That's not a strategy for abundance. That's a strategy for surviving fog.
In 2026, when someone in the room asks "where will AI be in 18 months," Sarah will give the same answer she gives now: "I have no idea. But here's where our customers need us to be." That's visibility worth something.