The sales pitch for AI is simple: your team gets time back.
The research coming out of Berkeley Haas says something harsher: in practice, your team may just get more work.
In a February 2026 Harvard Business Review article, Berkeley Haas researchers Aruna Ranganathan and Xingqi Maggie Ye described findings from an eight-month observational study inside a roughly 200-person U.S. technology company. Employees were free to use AI tools if they wanted. What happened was not a neat story about efficiency.
Workers moved faster. They took on a broader range of tasks. They worked longer hours. They prompted AI during breaks and in meetings. Many ran several threads in parallel. The researchers describe the result as a self-reinforcing cycle: once AI made faster output possible, speed expectations rose, task boundaries expanded, and work started filling the space that efficiency was supposed to free up.
For small and midsize businesses, this matters because you do not have much slack to begin with. If AI quietly turns a five-person team into a seven-job team, that is not transformation. That is burnout with better software.
What the study actually found
The important point is that this was not about forced adoption. Employees chose to use AI because it helped them get more done.
That is exactly why the findings are useful. The trap was not bad intent from management. The trap was that useful tools changed the pace of work, and the organization adapted around that new pace.
According to the researchers, AI use led to:
- Faster work rhythms as people completed drafts, analysis, and responses more quickly
- Broader task scope as workers started absorbing responsibilities that might previously have gone to another role
- Longer workdays because AI made it easy to keep going
- Work bleeding into recovery time with people prompting tools during breaks and meetings
- Fragmented attention as workers managed several AI-driven threads at once
This is the part many business owners miss. Productivity gains do not automatically turn into lighter workloads. Very often they turn into higher throughput expectations.
If one account manager can now draft proposals, summarize calls, prep follow-ups, and build internal documentation twice as fast, the business rarely says, "Great, take Friday afternoon off." It says, often without saying it out loud, "Maybe that person can also own onboarding, reporting, and renewal prep."
That is workload creep.
Why SMBs are especially exposed
Big companies can at least pretend they have change management. Most SMBs do not. A small team adopts AI in the middle of regular work, without formal process redesign, role definitions, or workload controls.
That creates three problems fast.
First, AI hides overload. Output goes up before anyone notices energy, focus, and decision quality going down.
Second, the owner starts normalizing stretch behavior. If your team can answer more emails, ship more drafts, and handle more client requests this month, it is easy to budget around that temporary burst as if it is the new baseline.
Third, tool use spreads faster than operating discipline. One person finds a prompt workflow that works, another copies it, and suddenly everyone is multitasking across chatbots, documents, CRM notes, and meeting summaries all day.
The result looks productive on paper. It feels terrible to live inside.
The wrong takeaway: "Stop using AI"
That would be lazy thinking.
AI is genuinely useful for small businesses. It can reduce time spent on routine writing, first drafts, research prep, support triage, internal documentation, and repetitive admin work. The problem is not the tool. The problem is adding the tool without changing the rules of the work.
If AI makes a task 40% faster and you simply pile on 40% more tasks, you did not create leverage. You created a more efficient path to exhaustion.
How to avoid the trap
The Berkeley Haas recommendation is not anti-AI. It is pro-intentionality: build pauses for assessment, sequence work to reduce fragmentation, and preserve human grounding so people still have room for recovery and creative thought.
For an SMB, that translates into a few practical moves.
1. Decide what AI should eliminate, not just accelerate
Do not measure success by asking whether a task is faster. Ask whether a task disappears, shrinks, or becomes simpler enough to remove from someone's plate.
If AI helps your operations lead produce reports twice as fast, but they still own the same reports plus three new deliverables, nothing improved.
2. Put a ceiling on parallel work
Running multiple AI threads feels productive right up until quality drops and nobody can think clearly.
Set simple limits. One deep work block at a time. Fewer live threads. No expectation that employees monitor AI outputs while also sitting in meetings and answering Slack.
Fragmentation is not a side effect. It is the cost.
3. Protect recovery time aggressively
If people are using AI during lunch, after hours, or in the gaps between meetings, the business is borrowing energy from tomorrow.
Make it explicit that faster tools do not mean permanent availability. If you want sustained performance, recovery cannot be optional.
4. Re-cut roles every time AI changes the workflow
When AI makes someone more capable, responsibilities tend to expand informally. Do not let that happen by drift.
Review roles on purpose. What should this person still own? What should move elsewhere? What should stop entirely? If scope expands, name it and price it accordingly.
5. Build human checkpoints into important work
The researchers emphasize human grounding for a reason. AI can keep momentum high even when the work is getting sloppier, more fragmented, or less thoughtful.
For sales, marketing, operations, hiring, or client communications, define where a human has to stop, review, and make a judgment call. Speed without judgment is just faster drift.
The bottom line
The Berkeley Haas research is useful because it punctures a comforting myth.
AI does not automatically give your team more time. In many cases, it increases the amount of work your team is expected to absorb.
That does not mean you should avoid AI. It means you should stop treating AI adoption as a software purchase and start treating it as an operating model decision.
If you want the upside, redesign the work. Otherwise the machine will make your team faster, and the business will reward that speed by piling on more.
That is not efficiency. That is a treadmill.
Source: Aruna Ranganathan and Xingqi Maggie Ye, "AI Doesn't Reduce Work—It Intensifies It," Harvard Business Review, February 9, 2026.