Two numbers came out this month that I haven't stopped thinking about. Sixty-nine percent of workers say their organization has "taken some action" on AI agents. Sixteen percent actually use one at work.

(Section's AI proficiency report, via The AI Daily Brief.)

Sit with that spread for a second. More than two-thirds of organizations have done something — bought licenses, run a pilot, sent the memo. One in six people is actually working differently because of it. And fewer than one in ten can define an AI agent in their own words. At organizations that already use agents, only 30% of employees have received any training on them.

That 53-point spread between "we took action" and "people use it" is the gap this newsletter is named for. It finally has numbers.

Here's what the gap costs while it sits there.

Workers who adopt AI more than doubled their time in email and chat, increased business-software use by 94%, and lost 9% of their focused, uninterrupted work — a state researchers are now calling "AI brain fry" (ActiveTrack analysis of 10,000+ workers). Glean found workers spending 6.4 hours a week "bot sitting" — feeding agents context, checking outputs, rerunning them. And in a GoTo survey, 43% of workers admitted submitting AI-generated content they suspected was low quality.

None of that is an AI capability problem. Every bit of it is an implementation problem: tools arrived, work wasn't redesigned, and people are quietly absorbing the difference.

Now the part I found genuinely hopeful.

Uber ran an experiment they call Agentic Pods: pair one AI-proficient engineer with one domain expert — a real person who does the real work — for a two-week sprint. Days one and two, the engineer just shadows the expert. Then they prioritize together, build alongside the worker, validate with peers, and ship on day ten. Sixteen pods across sixteen functions in two months. One result: capital allocation across 150 cities went from 15 hours to 30 minutes (Uber CTO Praveen Nepali).

Nepali's biggest lesson from the whole program: "The workflow becomes the unit of automation, not the individual task." And the best opportunities were invisible from the outside — "you find them by sitting next to the people doing the work."

That's a CTO with 30 engineers describing the exact method that works at any scale: don't audit the tools, sit with the people. Interview the team. Map the workflow they actually run, not the one in the process doc. The gap doesn't close from the software side. It closes from the work side.

Run your own 69/16 check. Ask five people on your team — not the champions, the regulars — two questions. "What AI tools has the company rolled out?" and "Which ones did you personally use this week?" The spread between those answers is your real adoption number, and it's a better diagnostic than any dashboard you're paying for.

The Implementation Lane is a weekly newsletter about making AI work inside real organizations. Written by Amanda Crawford, an AI Implementation Specialist who builds systems in the gap between configuration and engineering. If someone forwarded this to you, subscribe here.