TL;DR. Half your AI seats are idle six months in. The reflex is to fund a training program. Training won’t change the underlying gap: disposition, workflow fit, and manager signaling. Adoption improves when you replace the bad approved tool, fund the champions and workflow designers you already have, redesign and schedule three checkable workflows, make managers use the work visibly, and audit the flat-priced workspace seats quarterly. A team without a champion can still gain from one well-designed scheduled workflow.

You ran the audit from Evaluating Spend. The shape of your usage report was uncomfortable. Six months later you ran it again. Same shape. Maybe ten percent better at the top. The bottom thirty percent is still bottom thirty percent. The new seats you handed out after the last all-hands have already gone quiet.

Your head of HR has a proposal on your desk. It is a thirty thousand dollar AI enablement program. Six modules. Mandatory completion. A vendor with a deck full of “AI-fluent workforce” pull quotes. A certificate at the end. There is a similar memo from a department head asking for an “AI champions network.” There is a third asking to stand up an internal Center of Excellence.

None of this will move the number you actually care about.

Training won’t resolve the adoption gap

The default leadership move when usage stays flat is to schedule learning. It feels safest in front of a board. It is also the most reliably wrong response to an adoption gap in enterprise software I have watched leaders reach for.

The reasoning behind the training reflex goes like this. Usage is low. People don’t use what they don’t know how to use. Therefore teach them. Therefore the gap closes. Each step in that chain is intuitive. The problem is the second one. The premise that low usage is caused by low knowledge is empirically wrong for AI tools, and has been from the beginning.

The consumption gap between the top of the distribution and the median, better than tenfold in the telemetry, is not a knowledge gap. Bick, Blandin, and Deming’s St. Louis Fed paper found that people who had used generative AI in the prior week self-reported saving 5.4% of their work hours, about 2.2 hours a week.¹ That average hides the shape underneath it. LayerX’s enterprise telemetry found half of users had twelve conversations or fewer while the top 5% had at least 144.² BCG’s 2025 AI Radar surveyed 1,803 C-level executives and found that fewer than one-third of companies had managed to upskill even a quarter of their workforce on AI.³ Their headline finding was not “train more.” It was the opposite: their leading firms ran on a “10-20-70” principle in which 10% of effort goes to algorithms, 20% to data and technology, and 70% to people, processes, and cultural transformation. The training inside that 70% is a small piece. The bulk is workflow redesign and changes to how teams work.

The pattern across credible 2025 and 2026 datasets is the same. The companies extracting value picked a small number of workflows, redesigned them, and assigned the work to people who were going to do it well regardless of the training hours delivered.

Curriculum is downstream of the gap, not the cause of it.

Three forces create the adoption gap

Three things, in order of how much budget should go to fixing each.

Disposition. The people in the top 5% of the usage curve are willing to delegate work to a model, iterate on a wrong answer, and treat the chat tab as a workspace rather than a search engine. That posture isn’t taught in a session. It is a temperament. The people who have it figure the tool out in their first week. The people who lack it don’t develop that reflex through curriculum. Recognizing Leverage covers the shape of this distribution and Selecting Talent covers how to identify the people who sit at the head of it. That person is a champion. Champions determine whether an adoption program gets traction. The disposition gap is the largest and most durable gap.

Workflow fit. Most of the broad middle of your org opens AI once a week, asks it to summarize a document, copies the output into an email, and closes the tab. They get marginal value because their workflow has not changed. They are using a 2026 tool to do their 2022 job. The model only sees a small part of the work. The other ninety percent sits in steps the user never thought to surface because the workflow was not designed to surface anything. The workflows your champions did compress all had a cheap way to check whether the output was correct, which is the underlying property that decides which work yields to AI. The leftover workflows the broad middle faces usually lack that check. Process design is the missing piece.

Manager signaling. Your director-level managers need to use AI visibly. That means pasting drafts into the team channel and saying “here is the first cut Claude gave me, fix it,” asking in one-on-ones what a report’s prompt setup looks like, and modeling the behavior. Without that signal, the team correctly reads the situation as “the official line is to use AI, but the actual review and promotion incentives haven’t changed.” Adoption of a new working practice in a knowledge-work org follows the practices of the manager two layers up. AI is no exception.

The three failure modes interact. A leader without disposition can’t sense the workflow that needs redesign. A redesigned workflow that the manager doesn’t visibly use doesn’t stick. A manager who signals usage in a workflow that was never redesigned looks like theater within two weeks. All three need attention. Training belongs below them.

Idle seats belong to three different groups

Your idle seats fall into three groups.

The won’t. People with the wrong disposition for this tool, in this role, at this stage of their career. They are doing fine work. They won’t be your AI leverage story. Treat that as a staffing insight rather than a reason to make the program about them. Selecting Talent calls this group out at length, because pretending it doesn’t exist is the leadership move that costs the most.

The can’t. People in roles where current AI tools don’t have product-market fit. The relational sales lead in a six-quarter cycle. The senior litigator. The field service tech. The line worker. Recognizing Leverage names these. They define the floor of current adoption.

The won’t-bother. People with the disposition, in a role with fit, who have looked at the approved tool, found it inferior to what they used at home, and decided it isn’t worth the effort to swim against the procurement current. This is the group that responds to what you do this quarter. It is also the largest. Once the default improves, these people will follow your champion.

If you can’t tell which group each idle seat belongs to, you have a visibility problem. The audit in Evaluating Spend resolves it.

Five changes move adoption

Five moves. Order matters.

The wrong approved tool suppresses adoption

If the audit shows that your headline AI tool gets shallow use while your power users are paying out of pocket for something else, the tool is causing your adoption problem.

Every additional month it sits in front of your broad middle as the “official AI” is a month they form the conclusion that AI does not work. The tool they were given does not work for the job they have. They are correctly inferring its quality from the experience. They will then resist the next tool you put in front of them, because they have learned that your IT-procured “official” anything follows a vendor relationship rather than their actual workflow.

The default needs to be the tool your power users have already chosen. Move the seats and absorb the procurement awkwardness. The shadow AI line in your expense report is the user research.

Champions need concentrated funding

Take the dollars freed by cutting inactive flat-priced workspace seats and concentrate them on the people in your org who already have leverage. The shorthand is champion, and the long version is in Selecting Talent: the person on each team who has already built something nobody asked for, who keeps the chat tab open while they work, who has an opinion about the next tool and the next workflow change. Give them the higher tier. Give them the API budget. Give them a small operating budget for tools, prompt libraries, and time on the calendar. Tell them, in writing, that the expectation is they produce reusable artifacts (prompts, workflows, scripts, internal tools) that the rest of their team can run.

The usual move spreads the budget evenly across a broad cohort and produces a flat usage curve. Concentrate it on the steep part of the curve, where leverage spreads from. Your champions are the channel through which good practice reaches the broad middle. Their work is visible, so the practice can travel. A few people with budget and a mandate produce that effect.

If a function doesn’t have a champion you can name, write a hiring brief. Selecting Talent has the interview questions and the assessment exercise. A workflow designer elsewhere can still lift that team now: give them its clearest recurring, checkable job and have them build the standing workflow. One designed and scheduled job can change a team that has no personal power user. The hiring loop runs alongside it.

Three redesigned workflows beat thirty partial ones

Three workflows are enough. Choose three specific recurring workflows in three different functions, each consuming meaningful weekly time and each bottlenecked at a step a model can plausibly do. This does not require an enterprise-wide process re-engineering effort.

Examples that work: monthly close reconciliation in finance, account research and call prep in sales, first-draft contract redline in legal, internal report writing in operations, and customer ticket triage and response drafting in support.

For each, assign the champion on that team to redesign the workflow with AI in the loop, on a thirty-day clock, with a specific output: a written description of the new process, the prompt and tool stack used, and the new cycle time compared to the old. Where a workflow has a recurring, checkable step, include the scheduled job too.⁴ This is process-engineering work. The team learns the new shape by running it.

A scheduled task sits above the manual prompt stack. A champion or workflow designer writes the prompt once, picks a cadence, and lets the agent produce recurring, checkable work without anyone reopening the tool. Only schedule steps two people on the team can mechanically agree are right. An unattended job needs a verifier that runs faster than the work. The proactive agent pattern and this quarter’s state of AI cover where this is landing and how to govern it.

BCG’s data on this is direct. Companies that focus on 3.5 use cases expect 2.1 times the ROI of companies spreading themselves across 6.1 use cases.³ The discipline is depth, not breadth. Three workflows redesigned and scheduled to completion will move your usage numbers more than thirty workflows lightly touched.

Manager behavior carries more weight than policy

Your directors and VPs need to use AI visibly and repeatedly in front of their teams. Visible use means actual artifacts: the first draft of the strategy memo, with a note that it came from Claude; the market analysis with the prompt attached; the “here is what I asked the model and here is where it got it wrong” post in the team channel.

This is the strongest move on the list. Adoption of a new working practice in a knowledge-work organization is set by the manager two layers above the IC. If your director uses AI visibly, the team uses AI. If your director does not, the team does not, because the manager’s behavior predicts what gets rewarded at review time.

The implementation is small and uncomfortable. Tell the top two layers of management that visible AI use in their own workflow is part of how you evaluate their leadership over the next two quarters. Evaluate their own usage, not their team’s usage statistics.

The quarterly audit needs action

The seat-level audit in Evaluating Spend belongs in the operating cadence of an AI program that works. Every quarter, pull the report and sort activity on the flat-priced workspace SKUs. Cut the bottom third there, redirect the savings, ask your champions what they need, and act on the answer within thirty days. Metered work is different: manage consumption against named work and an accountable owner.

One thing has changed about the economics since this guide first ran. Seats still anchor the price on the business and workspace plans most companies buy. Metering has been added on top for agent work, APIs, credit packs, and a small number of enterprise plans. The Q3 2026 state of AI has the current vendor shape. The audit needs a second axis: surface the heavy-consumption workloads and ask whether that spend is landing on work that matters or on a workflow that should have been redesigned. Idle flat seats are visible waste. Mispriced consumption takes more work to find.

The hard part is cutting a flat-priced workspace seat from someone who said in a survey that they “find AI useful” when the audit shows they have not opened it in sixty days. The survey is polite; the audit is accurate. The second quarter is much easier than the first.

The adoption plan has five parts

The bad approved tool suppresses use. The “official” AI that your broad middle uses once and gives up on teaches them that AI does not work. Replace it with the one your power users already chose.
Champions deserve concentrated funding. Give the people who already have leverage tools, API, and a mandate to produce reusable artifacts. They are your distribution channel. Where you don’t have a champion, hire one, then let a workflow designer give the team a checkable recurring job it can run now. See Selecting Talent.
Three redesigned workflows beat thirty partial ones. Give each function’s champion or a workflow designer one recurring workflow, a thirty-day clock, and written outputs. Schedule the checkable steps. Depth beats breadth at 2.1 to 1.
Managers set the signal. Directors and VPs must use AI visibly on their own work. Make this part of how you evaluate them. Their behavior is the policy.
The audit needs action every quarter. Cut inactive flat-priced workspace seats, fund the active ones, and keep metered access broad while you manage spend against named work. Ask your champions what they need, then deliver in thirty days.

Mandatory training, corporate AI-champions networks with a logo and no authority, Centers of Excellence, enablement modules, and AI-fluency certifications have not moved a usage curve. They produce attendance and the appearance of action. They don’t produce adoption.

Training becomes useful once the first three workflows are redesigned and the manager signal is set. A tight, role-specific, hands-on session for the team running the new workflow is useful. Forty-five minutes, rather than six hours. Have the person who redesigned the workflow run it, anchored to the artifact rather than the abstract concept of “prompt engineering.” That costs a tenth of the program your HR head proposed.

A trusted function gives you a clean first test

The best first test is the function you trust most. Its champion is the person who has already built something nobody asked for, keeps the chat open while they work, and is named unprompted by teammates as the person to ask. If you can’t name one, Selecting Talent is the next read and the function needs a hire. In the meantime, give its clearest recurring, checkable job to a workflow designer from elsewhere in the org.

Where you can name a champion, identify the recurring workflow that consumes the most weekly time and is bottlenecked at a step a model can plausibly do. Hand it to the champion with a thirty-day clock, the budget for whatever tool they need, and a clear deliverable: a written description of the redesigned process, the prompt and tool stack, the scheduled job where the step is checkable, and the new cycle time.

Day thirty is the point to walk through the result with them.

You’ll get one of two outcomes. Either the workflow is meaningfully faster, in which case you have your first internal case study, your first piece of reusable IP, and your first concrete data point for the rest of the org. Or it isn’t, and you’ve learned something specific about where the current generation of tools doesn’t yet fit your business. Both are more valuable than another quarter of training attendance reports.

An adoption program finds the workflows where AI changes the work, gives them to the champion, and lets the manager signal carry the rest of the org.

Alexander Bick, Adam Blandin, and David Deming, “The Impact of Generative AI on Work Productivity”, Federal Reserve Bank of St. Louis, February 27, 2025. The November 2024 survey wave measured self-reported savings among people who had used generative AI in the prior week. Last verified August 10, 2026. ↩
LayerX, “State of AI Usage Report 2026”, May 28, 2026, enterprise browser telemetry. The top 5% had at least 144 conversations; half had 12 or fewer. Last verified August 10, 2026. ↩
BCG, “AI Radar 2025: From Potential to Profit”, January 15, 2025, 1,803 C-level executives across 19 markets. The 3.5-versus-6.1 use-case comparison is an executive expectation of 2.1x greater ROI, not a measured outcome. AI Radar 2026 is a separate 2,360-executive survey. Last verified August 10, 2026. ↩ ↩²
Anthropic introduced scheduled and on-demand Cowork tasks on February 25, 2026; Cowork reached general availability on desktop on April 9. Remote web and mobile sessions arrived July 7 and remain beta, rolling out from Max. Scheduled tasks are available on Pro, Max, Team, and Enterprise, and run remotely, including while the computer is asleep or Claude Desktop is closed. Last verified August 10, 2026. ↩

Driving Adoption