The shift worth recording is not that companies adopted AI. It is that adoption and productivity have come apart. Business Insider's report on PwC's 2026 Global CEO Survey says 56 percent of CEOs had not yet seen revenue or cost benefits from AI, while only 12 percent reported both higher revenue and lower costs. Deployment can be visible before return is measurable.
The clearest measure of the gap comes from individual work. In METR's 2025 randomized study of experienced open-source developers, developers using AI tools believed the tools made them 20 percent faster. Timed task completion showed them spending 19 percent longer. The spread between what the tools feel like they save and what they save runs in the wrong direction.
Validation is the hidden cost. AI output tends to arrive almost right rather than right, and the work of finding the difference does not register as the tool's cost. A worker drafting an email remembers the five minutes of generation and not the fifteen minutes spent checking tone, detail, and fact before sending. The generation is visible; the verification is absorbed.
The deeper division is between purposes. AI used to open new revenue, enter new markets, and build new products has a different return profile from AI used to do existing work somewhat faster. The first purpose can compound. The second yields a one-time gain in throughput and then flattens.
This is where small companies face a sharper version of the choice. A 1,000-person enterprise with a reconciliation team can convert AI into labor savings by applying it to a fraction of that team. A company with one finance person cannot; the tool moves that person to higher-value work without removing a salary. The efficiency path, which underwrites most large-company AI spending, is largely closed to small operators. The expansion path is not.
The historical rhyme comes from William Baumol, the economist whose work on cost disease showed that productivity gains concentrate where there is process to compress, while some work resists compression by its nature. Small, high-leverage companies have little back office to optimize. Their return on AI comes from widening what they can offer rather than shrinking what they spend, and partnership-shaped economics rewards the company that reinvests a saved hour into a new line rather than banking it against headcount.
The Almanac will watch the divergence between adoption and return surface in third- and fourth-quarter 2026 earnings from small- and mid-cap software and services companies. The early signal points toward higher software spend against flat or thinner margins. The question for 2027 is whether the winners are the companies that changed what they do, not only the companies that did the same work faster.