Six weeks in, the person who is now our top contributor was ready to quit using AI. He had never been a developer, and while he’d already had a few wins, the results were inconsistent. Something that should have taken twenty minutes could take three hours.
The valley is the stretch after the first win, when the tool’s limitations and the person’s learning curve show up at the same time. Results get inconsistent, tasks that should be faster aren’t, and it is easy to wonder whether the first win was a fluke. It is stage two of five, and it is where most AI adoption quietly dies.
Six weeks in, the person who is now our top contributor was ready to quit using AI. He had never been a developer, and while he’d already had a few wins, the results were inconsistent. Something that should have taken twenty minutes could take three hours, and sometimes the result was worse than if he’d just done it himself.
What made it harder was watching everyone else seem to figure it out. People were sharing what they had built and discussing how much faster they were working, while he was still fighting with the tool. It was easy to assume the problem was him.
But looking back, the period where he was most ready to give up wasn’t evidence that AI wasn’t working for him. It was a stage almost everyone at CampusIQ would eventually go through.
We started calling it the valley.
When we first started pushing AI adoption across CampusIQ, we assumed the hard part would be getting people to start. Once someone saw what the tools could do and experienced a meaningful win, we thought adoption would naturally build from there.
Instead, we kept seeing the same pattern. The first win created excitement, but it was often followed by a period when the tool’s limitations and the person’s learning curve showed up at the same time. Once we saw it happen across different people and roles, we started thinking about AI adoption in five stages.
You haven’t really started yet. The cost of learning a new way of working still feels higher than continuing to do things the way you’ve always done them.
Something works, and usually better than you expected. You start to understand what everyone has been talking about and imagine all the other things you could do with it.
The early excitement wears off. The tool’s limitations become more obvious at the same time that your own skill gaps start showing up. Results become inconsistent and tasks that should be faster aren’t.
Your mental model starts to change. You understand what to give the tool, where it needs more direction, and when you need to step in. You get better at working with it.
Eventually, the tool becomes less visible. You’re no longer thinking about whether you’re “using AI.” You’re solving problems and creating things that make other people faster too.
Stage two can be particularly frustrating because the experience doesn’t match what people expect from AI. They’ve already seen it work, so when bad outputs, wasted time, and unexpected mistakes start showing up, it can feel like they’re going backward.
It’s even harder when everyone else’s successes are more visible. Your Slack is full of what someone built in an afternoon while you’ve spent three hours on something you could have done manually in twenty minutes. One mistake can quickly undo the trust built from that first win.
We initially treated the stall as something we could fix with better tools or more documentation. That framing put the problem in the software, or in the person, and it sent us looking for a fix that was never going to land.
Eventually, we realized the experience itself wasn’t unusual. The problem was that nobody knew to expect it. When you know the wall is coming, a stretch of bad results feels less like proof that AI doesn’t work and more like part of learning how to use it.
Simply naming the valley became one of the most useful things we did for adoption.
Knowing the valley existed helped, but people were much more likely to move through it when they weren’t figuring everything out alone. Often, the most helpful person wasn’t the most technical. It was a peer who had already been through the same frustration and could explain what finally clicked for them.
The person who gets someone through the valley is usually someone who has already been through the same frustration, and who can explain what finally clicked for them.
We started working directly with teams instead of simply giving them tools and documentation. Over three days, we helped people set up and use AI to solve real problems in their own jobs, not hypothetical exercises. That meant we were also there when they hit the valley and could work through the bad output before frustration turned into giving up.
Once members of the executive team went through the same process, adoption became easier. Seeing someone in a similar role successfully use AI in their actual work built more confidence than another presentation about what AI could theoretically do.
One of the more interesting things we learned was that technical ability wasn’t the clearest indicator of who moved through the stages quickly. Having someone nearby who had already been through the process mattered more.
That changed how we approached AI adoption. Giving people the right tools and training wasn’t enough. We also had to support them when the initial excitement wore off, and the work got harder.
We learned that moving backward is normal, too. A new model or unfamiliar problem can expose new limitations, but that doesn’t mean someone is starting over. They carry what they learned the first time with them, making the next climb through the stages easier.
Most AI adoption stories focus on the beginning and the end. Someone starts experimenting with AI, and before long they’re automating work, building things they couldn’t build before, and getting dramatically more done.
What usually gets skipped is the period in between, when the results are inconsistent and the person who was excited two weeks ago starts wondering whether they’re actually any good at this.
That’s the part companies need to prepare people for. If someone struggles after their first win, it doesn’t necessarily mean the tool isn’t working or that they aren’t capable of using it. They may simply be experiencing a predictable part of learning a fundamentally different way of working.
For us, recognizing that changed the question from “Why aren’t they adopting AI?” to “What do they need to get through the valley?”
And once we started asking that question, adoption became much easier to support.
As CEO and Founder of CampusIQ, Aaron leads the vision to transform how colleges and universities understand and use space. A mathematician and data scientist, he’s built analytics platforms and teams across higher ed. Off the clock, he’s a former Academic All-American lacrosse goalie, coach, and relentless builder of things.
Connect on LinkedIn →
Aaron Benz : Sep 9, 2026, 11:32:34 AM
We set out to see if four engineers, working with AI, could produce like a team five times the size. What we built along the way changed who can...
Veena Vadgama : Nov 20, 2025, 12:12:20 PM
by Veena Vadgama, CampusIQ Chief Marketing Officer
Veena Vadgama : Nov 12, 2025, 6:00:01 AM
By Veena Vadgama, Chief Marketing Officer, CampusIQ
Aaron Benz