Base research
The AI Admin Debt Study, 2026
Ungated. Every finding is on this page.
The finding
AI gives back the hours.
Managing it takes them back.
77% of the people we studied say AI makes them feel more productive. 68% say they have to adjust what AI produces most of the time or almost every time before they can use it. 49% told us both things at once.
That gap has a cost, and the cost has a name. AI Admin Debt is the work that arrives with AI: the setup, the prompting, the checking, the fixing, and the deciding which tool to use next. It does not show up on a calendar. It shows up in the evening.
72% of the operators we studied spend four or more hours a week on it. 34% spend eight or more. For most of them, nobody else is doing that work, because eighty percent manage their own AI workflows day to day. That holds for organizational leaders too: 71% of the respondents at companies of eleven or more employees are still the ones doing it.
Key findings
What we found
49%
Half feel faster and still rewrite nearly everything
In a 2026 study of 74 US-based founders and operators, 49% reported both that AI makes them feel more productive and that they must adjust AI output most or almost all of the time. The perception of speed and the hours actually spent have come apart.
91%
The problem is depth, not tool sprawl
91% of respondents use only one to six AI tools in a typical work week. Just one respondent used more than ten. Yet 34% spend eight or more hours a week on AI-related work. AI Admin Debt is not produced by a bloated stack. It is produced by how deep the work goes on a small number of tools.
80%
The most expensive person in the company is running the AI
80% of respondents said they personally manage their own AI workflows day to day, rather than a team, an ops function, or an assistant. 68% said they do not have anyone they fully trust to evaluate which AI tools their team should be using. The AI management job exists. In most organizations, no one has been given it.
71%
Company size does not change the answer
Among the 34 respondents at companies of 11 or more employees, 71% still manage their own AI workflows, and only 3 of them have a dedicated AI or operations person doing it. At companies of 201 or more, 73% still manage it themselves.
20% vs 36%
Executive support measurably reduces the load
Among respondents with full-time executive support, 20% spend eight or more hours a week on AI-related work. Among those with no executive support, 36% do. The work does not disappear. It moves to someone whose job it is.
Verbatims
In their words
"It is like a really powerful seven year old that requires all your attention."
Founder, AI and infrastructure
"AI doesn't know what it doesn't know about you. Every conversation starts from scratch. The person carries the institutional memory and AI borrows it. That's a real tax on the human."
Chief growth officer, professional services
"Equal parts transformative and exhausting."
Founder, professional services
"They are confidently wrong often. The confidence makes it harder to spot the mistakes."
Senior functional lead, software
"The AI usually gets about 80 to 90% of what I'm trying to do, but that last little bit is a lot harder than it should be. I'm not always sure it's faster or easier."
Founder, professional services
"I feel like I have superpowers and it has opened many doors for me. Also, it feels like a trap that is hard to escape."
Founder, professional services
Methodology
How we ran this study
What this is
A depth study of 74 US-based founders, operators, and senior functional leads on the time cost of working with AI tools. We set out to measure something that does not appear in productivity reporting: the hours that arrive with AI rather than the hours it returns.
Field window
April 29 to June 15, 2026.
Base
74 unique respondents. Duplicate submissions, Base employees, and incomplete records were removed before analysis.
How respondents were sourced
Direct outreach and professional networks in operator, founder, and executive support communities. Respondents self-selected into the study.
Who answered
45% founders or CEOs, 22% senior individual contributors or functional leads, 18% operators including COOs and chiefs of staff. 54% work solo or at companies of 2 to 10 people, and 46% at companies of 11 or more. 42% work in technology and software, 30% in professional services. 76% have no executive support today.
What we did not do
This is not a representative sample of any population, and the figures here should not be read as market projections. Respondents self-selected, and the sample skews toward small companies in technology and professional services. We report what these 74 operators told us. The value is in the depth and the consistency of the pattern, not in statistical generalization.
Questions and data
The full instrument and the aggregate response tables are available on request at [email protected].
Data tables
The data
Table 1: Hours per week spent on AI-related work (n=74)
| Hours per week | Respondents | Share |
|---|---|---|
| Less than 1 hour | 5 |
7%
|
| 1 to 3 hours | 16 |
22%
|
| 4 to 7 hours | 28 |
38%
|
| 8 to 15 hours | 14 |
19%
|
| More than 15 hours | 11 |
15%
|
Table 2: How often AI output needs adjustment before use (n=74)
| Frequency | Respondents | Share |
|---|---|---|
| Almost always | 24 |
32%
|
| Most of the time | 26 |
35%
|
| About half the time | 19 |
26%
|
| Almost never | 5 |
7%
|
Table 3: Who manages AI workflows day to day (n=74)
| Owner | Respondents | Share |
|---|---|---|
| The respondent personally | 59 |
80%
|
| Spread across the team | 8 |
11%
|
| A dedicated AI or ops person | 4 |
5%
|
| Nobody | 3 |
4%
|
Table 4: Number of AI tools used in a typical work week (n=74)
| Tools | Respondents | Share |
|---|---|---|
| 1 to 3 | 35 |
47%
|
| 4 to 6 | 32 |
43%
|
| 7 to 10 | 6 |
8%
|
| More than 10 | 1 |
1%
|
Table 5: Share who personally manage their own AI workflows, by company size (n=74)
| Company size | Respondents | Self-manage | Share |
|---|---|---|---|
| Solo | 19 | 19 |
100%
|
| 2 to 10 | 21 | 16 |
76%
|
| 11 to 50 | 12 | 8 |
67%
|
| 51 to 200 | 11 | 8 |
73%
|
| 201 or more | 11 | 8 |
73%
|
Subgroups here are small, so this table shows counts alongside shares. At companies of 51 to 200 and at companies of 201 or more, 73% is 8 of 11 respondents.
What this means
What we take from this
The market response to AI overload has been to consolidate the stack. Our data points somewhere else. People are not drowning in tools. They are drowning in the work that a few tools generate, and they are doing that work themselves because it was never assigned to anyone.
Headcount alone does not fix it. Once a company passes ten employees the share of people managing their own AI workflows stops falling, and it sits near seventy percent from there through the largest organizations we studied. Adding people to a team does not hand the AI management work to anyone in particular.
That work is absorbable. It requires a person who treats AI workflows as part of the job rather than a side project, who can tell what is worth automating, and who keeps the human work human. Base places those people. We built this study because we wanted to know how large the problem had become before we made any claims about solving it.
PDF download. Email optional.
Add your own numbers to the 2027 studyAbout the author
Sara Altuna
CEO @ Base
Sara Altuna, Chief Executive Officer. Sara joined Base as an executive assistant in 2018 and never left. She created Inbox Reset, worked across learning, talent, product, and strategy, became President in 2025, and now leads the company as CEO. She knows the EA role from the inside, and it shapes how Base serves both assistants and the leaders who rely on them.