Two scenes from a typical mid-market company in 2026.
In one, a 23-year-old analyst spent her weekend wiring up Claude to her CRM via MCP, then built a custom agent that summarizes account health and drafts outreach emails. She demos it on Tuesday. Her boss is impressed. He doesn't fully follow how it works.
In the other, the same boss is in a leadership meeting deciding whether to invest a hundred thousand dollars in an enterprise AI platform. The vendor's deck is convincing. He approves it. The platform never reaches production because the team that would actually use it builds something better with off-the-shelf tools and a weekend.
This is the awkward shape of AI fluency in 2026. The people who understand the technology have the least decision rights. The people with the decision rights have the least understanding. Most mid-market companies are running a quiet structural failure mode in their AI investment, and the failure is showing up in budgets, in tool sprawl, and in the gap between what the org is buying and what its teams actually use.
Why fluency clusters at the bottom
AI fluency in 2026 isn't a generational thing exactly, but it's strongly correlated with how recently you've been hands-on with new tools. Junior people are hands-on by default. Senior people are hands-on by exception.
The gap compounds because the tools are moving fast. An LLM that did one thing six months ago can now reason about its own work, call tools, and integrate with operational systems via MCP. Keeping up with that requires actually using the tools every week. Most VPs in mid-market organizations don't have the schedule to use a new tool every week.
That's not a criticism of the VPs. It's the structure of the role. They're paid to make decisions, not to write prompts. The result is that the most current understanding of what AI can and can't do lives in the heads of people whose job titles say 'analyst' or 'associate.' Those people are doing the experimentation, finding the workflows that work, and generally getting smarter about AI on a weekly cadence that the senior org can't match without changing how time gets allocated.
Why decision rights cluster at the top
Decision rights cluster at the top because authority and accountability follow seniority. That's not changing. The board doesn't fire interns when a six-figure platform rollout fails. They fire VPs.
The risk-aversion is calibrated correctly for almost every other category of decision. AI tooling is the rare category where it's calibrated wrong, because the most important input to a good decision is hands-on knowledge of what the tools actually do, and that knowledge is sitting two levels below the person making the call.
The outcome is what you'd predict. Mid-market companies in 2026 are systematically over-buying enterprise AI platforms and under-investing in the workflows their junior teams have already figured out work better. The platforms get bought because they have a polished sales motion, an enterprise compliance story, and a vendor relationship the VP feels comfortable defending. The workflows the team actually wants don't have a sales motion, so they don't get the same airtime in the budget conversation, and they end up underfunded relative to what they're producing.
Mid-market companies in 2026 are systematically over-buying enterprise AI platforms and under-investing in the workflows their junior teams have already figured out work better.
What this looks like in practice
We had a conversation last quarter with the CEO of a $120M revenue services firm. He'd just approved a six-figure annual contract for an enterprise AI platform that promised to handle proposal generation, client communication summarization, and knowledge management across the firm.
A week later, his head of analytics quietly mentioned that two of the senior associates had built a working version of all three workflows over a weekend, using Claude, the firm's existing document storage, and a few prompt templates. The associates had been using the workflows for three months. They'd shared them with about a dozen people across the firm, all of whom found them more useful than anything else the team had tried.
The CEO did not know any of this when he approved the platform purchase. The associates hadn't shared their work because they weren't sure who to share it with, didn't think it was 'serious' enough to surface to leadership, and assumed the firm would do something more enterprise-grade if it wanted real AI capability. The platform purchase happened in a vacuum the firm had created by not having a knowledge-capture loop.
The end of the story is more common than it should be. The platform got delivered, never reached full adoption, and the associates kept using their weekend stack. The firm paid for both. Eighteen months later, the platform got quietly retired and the firm finally formalized the workflow the associates had built, two years late.
The org-design problem this creates
The fix isn't to give interns purchasing authority. The fix is to design a knowledge-capture loop that closes the gap before decisions get made.
The failure mode we see most often is silent. The junior analyst builds something useful. She doesn't share it because the demo culture in her company isn't strong, or because she assumes leadership wouldn't be interested. The VP makes a decision without knowing the better option exists. Six months later, the analyst has either moved on or stopped innovating with company time.
The pattern repeats. Each cycle the organization is slightly more out of date than the cohort it just hired. After two or three cycles, the gap between what the org has officially deployed and what its junior teams actually use is wide enough to show up as a productivity drag. The org thinks it has a $400K AI investment. The team thinks it has whatever its associates built last weekend.
How to capture knowledge at each level
Three practical moves we've seen work in mid-market organizations.
First, run a recurring AI demo session. Once a month, anyone in the company who built something useful with an AI tool gets fifteen minutes to show what they did. Leadership attends. The format is informal. The point isn't training. The point is that the knowledge can travel from the person who has it to the people who need to make decisions with it. After three or four months of this, leadership stops being surprised by what the team has already figured out.
Second, treat the AI buying decision as a build-versus-buy decision the same way you treat any other technical investment. Before approving a six-figure platform purchase, the team that would use it gets two weeks to build a working alternative with the tools they already have. Half the time the platform purchase becomes obviously unnecessary. The other half the time the platform decision improves because the team understands the alternative.
Third, write down the AI capabilities you actually have. Most companies have no inventory of which functions are using which tools, what those tools do, what they've learned. A simple shared document, updated quarterly, surfaces the capabilities that already exist and stops you from buying something you've already built. The document doesn't need to be sophisticated. It needs to exist and to be updated.
Where the AI fluency gap is heading
The companies that capture this knowledge well in 2026 will look like the companies that captured analytics knowledge well in 2014. The ones who don't will spend the next three years buying platforms their junior teams could have built, then wondering why their AI investments aren't compounding.
The gap will widen before it closes. The tools are not slowing down. The hands-on time required to keep up with them is going up, not down. The structural reasons fluency clusters at the bottom of the org chart aren't changing in the next eighteen months. The companies that recognize this and design knowledge-capture into their operating model will look very different from the companies that don't, and the difference will compound quietly until it shows up in margins, in retention, and in the speed at which the company can ship new internal capability.
The leadership move worth making in 2026 isn't to learn the tools yourself. It's to design the operating loop that brings the team's tool knowledge into the room where the buying decisions happen. That's the cheapest, highest-impact adjustment most mid-market companies can make this year.
