AI agents don’t free us from work. They make us do even more work to keep up with competition and customer expectations.
Quarterly releases? Gone. Sales reps getting back to you in a few days? Over. The customer being responsible for their own deployment? A relic.
Every AI vendor is selling you the same dream: do more with less. Fewer employees. Less effort. Same output. Maybe even better output.
It’s a compelling pitch. And it’s mostly wrong.
Not because AI agents don’t work — they do. We run SaaStr with 3 humans and 20+ AI agents generating over $1M in revenue. But here’s what nobody tells you: we’re not working less. We’re doing dramatically more than we ever did with 20+ employees. And we have to. Because the old baselines are dead.
This isn’t a bug. It’s a pattern that’s over 100 years old.
The Vacuum Cleaner That Didn’t Save Any Time
In 1983, historian Ruth Schwartz Cowan published More Work for Mother, documenting one of the most counterintuitive findings in the history of technology: labor-saving household appliances didn’t actually save any labor. Economist Joel Mokyr later dubbed this the “Cowan Paradox” — and it explains exactly what’s about to happen with AI agents.
The vacuum cleaner is the clearest example. Before it existed, cleaning a carpet was a brutal, seasonal project. You had to move all the furniture, roll up heavy wool rugs, drag them outside, hang them on a line, and beat them with a paddle until the dust stopped. It took multiple people. It happened once or twice a year. That’s literally where “Spring Cleaning” comes from.
Then the electric vacuum arrived in the 1910s and 20s. Suddenly, you could clean a rug right where it lay. No heavy lifting. No help needed.
So what happened? Did housework go down?
No. Three things shifted simultaneously:
- The frequency spiked. Cleaning a rug was now “easy,” so it was no longer acceptable to have a dusty floor for six months. Advertisers weaponized germ theory. The standard moved from annual deep cleans to weekly — or daily — vacuuming.
- The labor got reassigned. Beating rugs required brute strength, so it was often done by men or hired help. Vacuuming required less physical effort, so it got reclassified as “light housekeeping” — the wife’s job. The work didn’t disappear. It moved.
- The helpers got fired. Middle-class families had employed domestic servants for the heavy work. The “electric servant” replaced the human one. The housewife now did it all, alone.
The result? Time-use studies from the 1930s through the 1950s show housewives still spent roughly 51 hours a week on housework — barely changed from before the appliances arrived.
The tools changed. The hours didn’t.
Cowan called this the paradox of labor-saving technology: it doesn’t save labor. It raises the standard of what’s expected and creates new categories of work that didn’t exist before.
If this sounds familiar, it should. The Cowan Paradox is a cousin of the Jevons Paradox, identified by economist William Stanley Jevons in 1865. Jevons observed that when coal engines became more efficient, total coal consumption increased — because efficiency lowered the cost per unit, which drove more usage. Build more highway lanes, traffic increases. Make cars more fuel-efficient, people drive more miles. Make a task easier, people do more of it.
The pattern is 160 years old: efficiency gains don’t reduce consumption. They get absorbed by demand expansion.
Cowan applied this to household labor. Now it’s playing out with AI agents in B2B.
If this sounds familiar, it should. Cowan’s insight is a cousin of the Jevons Paradox, discovered over a century earlier. In 1865, economist William Stanley Jevons observed that as coal engines became more efficient, total coal consumption increased — because efficiency lowered the cost per unit, which drove more usage. Build more highway lanes, traffic increases. Make cars more fuel-efficient, people drive more miles. Make housework easier, cleanliness standards go up. Make knowledge work faster with AI, output expectations explode.
It’s the same pattern every time: efficiency gains get consumed by demand expansion, not by reduced effort.
The difference with AI is the speed. The vacuum cleaner took decades to reset household expectations. AI agents are resetting B2B expectations in months.
This Is Exactly What’s Happening With AI Agents
Weve watched this pattern repeat in real-time at SaaStr.
Two years ago, we had 20+ employees running events, content, community, and a fund. Today we have 3 humans and 20+ AI agents. Our revenue trajectory flipped from -19% to +47% year-over-year. SaaStr.ai hit 500,000+ users in 45 days. We process hundreds of thousands of startup valuations every month. I personally built 10+ production applications using AI that have generated 750,000+ uses.
None of that was happening when we had 20+ people.
The AI agents didn’t let us do the same work with fewer people. They let us do entirely new categories of work that were previously impossible. And now that we can do them, we have to do them. Because if we don’t, someone else will.
That’s the Cowan Paradox in B2B: AI doesn’t reduce work. It resets the baseline of what’s expected.
You’re Seeing This Everywhere If You’re Paying Attention
Aaron Levie, CEO of Box, said it plainly at SaaStr AI Annual: “I’m just finding more stuff to have the AI do — and then I end up doing more work as a result.” He called it the death of the four-day work week: “The company that implements the four-day work week, they’ll just have a competitor that says, ‘Well, we can work five days, but also with this productivity gain.'”
That’s the Cowan Paradox in one quote from a CEO running a billion-dollar company.
Or take coding. Everyone brags that 50% of their code is now written by Cursor or Claude. But does that mean your engineers only work three hours a day? Of course not. They ship more. They take on more projects. They build features that would have been deprioritized six months ago. The AI didn’t buy them free time. It raised the bar on what “shipping fast” means.
At SaaStr, our Chief AI Officer Amelia spends an hour every morning managing our 20+ AI agents — checking Qualified for overnight meeting bookings, reviewing Artisan’s outbound sends, QA’ing support conversations, monitoring Momentum’s deal summaries. That same hour used to be spent in 1:1s with two human reps. The hour is identical. But the output is 10x. And she keeps finding new things for the agents to do — deanonymizing website traffic, running follow-up campaigns, expanding into use cases we never would have staffed for. The work compounds. It never contracts.
The Research Now Backs This Up
This isn’t just a theory from a 1983 history book or my experience running SaaStr.
HBR published a piece this week — “AI Doesn’t Reduce Work — It Intensifies It” — based on an eight-month study by Berkeley Haas researchers Aruna Ranganathan and Xingqi Maggie Ye. They embedded at a 200-person tech company and tracked what actually happened when employees adopted AI tools.
Three findings, none of them surprising if you’ve lived through this:
Task expansion. Workers didn’t just do their own jobs faster. They started doing other people’s jobs. Product managers started writing code. Researchers took on engineering tasks. People absorbed work that previously would have justified hiring someone new. Sound familiar? It’s the vacuum cleaner replacing the hired help.
The boundaries between work and non-work dissolved. Workers started prompting AI during lunch, in meetings, before leaving for the day — “just one more quick prompt.” The conversational feel of AI tools made it not feel like work. Until it was 9 PM and they were still going. As one engineer told the researchers: “You had thought that maybe, oh, because you could be more productive with AI, then you save some time, you can work less. But then really, you don’t work less. You just work the same amount or even more.”
Multitasking exploded. Workers ran multiple AI agents in parallel, revived old projects because AI could “handle them in the background,” and managed several active threads at once. The feeling was momentum. The reality was cognitive overload.
The researchers’ conclusion: without intentional guardrails, AI doesn’t contract work. It intensifies it. The productivity surge is real, but so is the burnout that follows.
This is the Cowan Paradox validated by an eight-month field study, 40+ interviews, and published in HBR. In 2026.
How the Paradox Is Killing the Old B2B Playbook
The baselines that defined B2B for 20 years are collapsing in real-time. Not slowly. Right now.
- Quarterly releases are over. When AI coding agents let a team of 3 ship what used to take 15, your customers don’t wait 90 days for the next release. They expect continuous deployment. Weekly updates. Instant bug fixes. If you’re still on a quarterly cycle, you feel broken compared to the competitor shipping daily.
- “We’ll get back to you in 48 hours” is over. When AI agents can respond to a prospect in 30 seconds with full context on their company, their tech stack, and their use case, a 2-day response time from a human SDR isn’t “standard.” It’s a lost deal.
- Customer-managed deployments are over. When AI agents can handle onboarding, configuration, and integration automatically, asking the customer to do their own deployment feels like asking them to beat their own rugs in the yard. The expectation shifted. You deploy it for them, or they find someone who will.
- Annual support contracts with email-only SLAs are over. When AI agents can provide instant, personalized support 24/7, charging extra for “premium support” that means “a human responds within 4 hours” sounds absurd.
None of these old baselines were unreasonable two years ago. They’re all unacceptable now. Not because customers got greedier. Because AI made better possible, and “possible” always becomes “expected.”
This is happening everywhere founders are deploying AI agents.
Customer Support
The pitch: AI agents handle 80% of tickets. You need half the support team.
The reality: Customers now expect instant, 24/7, perfect responses. They expect personalization. They expect the AI to know their account history, their plan, their last three conversations. The 20% of tickets that still need humans are the hardest, most complex ones — the ones that take real skill and time.
Paul Adams, CPO of Intercom, shared at SaaStr AI London what happened when they deployed Fin, their AI agent that now resolves over 1 million customer problems per week. Did Intercom’s team shrink and relax? No. They parted ways with ~33% of the company, had every designer start shipping code to production, Paul personally took over two-thirds of marketing and rebuilt it from scratch, and they now run hundreds of experiments on AI accuracy because every tiny improvement compounds. As Paul put it: “If it doesn’t feel brutal, you’re not going deep enough.” The work didn’t decrease. It transformed and intensified.
You didn’t reduce the work. You raised the floor of what “good support” means. And your competitors raised it too.
Content and Marketing
The pitch: AI can draft a blog post in 2 minutes. Your content team can produce 5x more.
The reality: If AI can draft a post in 2 minutes, so can every other company in your space. The volume of content explodes. The bar for what breaks through the noise goes up, not down. Now you need better distribution, better insights, more original data, more founder voice — all the things AI can’t easily replicate. The humans on your team aren’t doing less. They’re doing harder, higher-leverage work.
Sales and Outreach
The pitch: AI SDRs can send 1,000 personalized emails a day. You can cover 10x more pipeline.
The reality: Every company now sends 1,000 personalized emails a day. Inboxes are flooded. Response rates crater. The signal-to-noise ratio collapses. Now you need even more creative approaches to break through — warm intros, community, events, content, product-led growth. The work didn’t shrink. It shapeshifted.
Philip Laheurte, CRO of Personio, shared at SaaStr AI London what happened when they deployed AI research assistants across their 400-person sales team. Research time per rep dropped from 2 hours a day to 15 minutes. Pipeline per rep doubled. Sounds like the dream, right? Less work, more output?
Except they didn’t stop there. They built 400+ AI assistants in six months. Their AI chat agent booked 140 meetings in 7 days. Reps who used to spend 2 hours researching now spend that time working more pipeline, doing more outreach, handling more conversations. The 1 hour and 45 minutes of “saved” time didn’t become free time. It became more work. As Philip put it: “If you use ChatGPT to make a sentence sound better, you’re not doing AI transformation.” Real transformation means the work expands to match the new capacity.
Product Development
The pitch: AI coding agents let a team of 3 engineers do what used to take 15.
The reality: If your 3 engineers can ship what 15 used to, so can your competitor’s 3 engineers. The pace of shipping accelerates across the entire market. Quarterly releases become weekly. Weekly becomes daily. Customers expect faster iteration, more features, more polish. Your 3 engineers aren’t kicking back. They’re shipping at a pace that would have been physically impossible two years ago — and they have to, because everyone else is too. The quarterly roadmap presentation is a museum piece.
Everyone Is More Productive. That Means We All Work Harder. Or You Fall Behind.
Here’s the part nobody wants to hear.
If AI makes every company 5x more productive, and every company adopts AI, then no company has an advantage from AI productivity alone. The baseline just moved up. You’re on a treadmill that got faster.
This is exactly what happened with the vacuum cleaner. Once every household had one, a clean floor wasn’t a competitive advantage. It was the minimum expectation. The work didn’t go away. The standard just got higher.
In B2B, this means:
- If AI lets you publish 10 blog posts a week instead of 2, your competitors will publish 10 too. Now 10 is the baseline.
- If AI lets you respond to support tickets in 30 seconds, customers will expect 30 seconds. The old “24-hour response time” SLA is dead.
- If AI lets you personalize outreach at scale, every buyer’s inbox is full of “personalized” messages. Personalization stops being a differentiator.
The work doesn’t decrease. It compounds.
So What Do You Actually Do About This?
Understanding the Cowan Paradox doesn’t mean you should avoid AI. That would be like refusing to buy a vacuum cleaner in 1920. You’d just have dirtier floors than everyone else.
But it should change how you think about deploying AI agents. Here’s the framework:
1. Don’t Deploy AI to Do Less. Deploy It to Do What Was Previously Impossible.
The biggest trap is using AI to replicate your existing workflows cheaper. That’s the vacuum cleaner replacing the carpet beater — same work, different tool, standards go up, net hours stay flat.
The real leverage is using AI to do things you literally couldn’t do before at any headcount. At SaaStr, we don’t use AI to write the same blog posts faster. We use it to process hundreds of thousands of startup valuations — something that would have required a team of 50+ analysts and would never have been economically viable.
Ask yourself: what could we do with AI that we would never have attempted with humans? That’s where the real value is.
2. Expect the Work to Increase, and Plan For It
Stop building your AI strategy around “we’ll need fewer people.” Start building it around “we’ll be able to do dramatically more with the same people, and we’ll need to.”
Wade Foster, CEO of Zapier, described the model that actually works: “The agent is basically doing 90 plus percent of the work and leaves the last mile to the account rep to make it happen.” That’s not fewer people doing less. That’s the same people doing 10x more volume with AI handling the repetitive 90% — and the humans handling the critical, high-judgment last mile that now comes at them 10x faster.
The companies that win with AI agents won’t be the ones that cut headcount the fastest. They’ll be the ones that redeploy their humans to the highest-leverage work while AI handles the new, expanded baseline of expected output.
3. Compete on Taste, Judgment, and Relationships — Not Volume
If AI equalizes volume and speed, the differentiators become the things AI can’t easily replicate: original insight, founder voice, authentic relationships, community, taste, and judgment.
This is why I spend 1.5-2 hours a day building apps by hand on Replit even though AI does most of the coding. The judgment about what to build and why it matters is the human edge. The AI is the tool. The taste is the moat.
4. Watch Your Team’s Cognitive Load
The Cowan Paradox has a human cost. When the vacuum cleaner arrived, nobody said “great, the housewife can now clean alone, without help, every day instead of once a year, in isolation.” But that’s what happened.
AI agents create the same dynamic. The Berkeley Haas researchers found that workers who adopted AI felt more productive but not less busy — and in many cases felt busier than before. The voluntary nature of the expansion made it invisible to leadership. People absorbed more work because AI made it feel easy, until the cumulative cognitive load caught up with them.
Your team of 3 is now expected to do what 15 did — and more. The cognitive load, the context-switching, the sheer volume of decisions doesn’t decrease just because AI handles execution. The HBR researchers recommend what they call an “AI practice” — intentional norms around when to stop, what not to expand into, and how to protect recovery time. That’s worth taking seriously. Make sure your humans aren’t burning out doing 5x the strategic work while AI does 10x the tactical work.
AI Agents Level Up The Work. They Don’t Eliminate It.
Ruth Schwartz Cowan figured this out in 1983 studying vacuum cleaners. The pattern hasn’t changed.
Labor-saving technology doesn’t save labor. It raises expectations and creates new categories of work.
AI agents are the most powerful labor-saving technology since electricity. And they will follow the same pattern. The companies that deploy them expecting to do less will be confused when the work expands. The companies that deploy them expecting to do dramatically more — things that were previously impossible — will win.
The vacuum cleaner didn’t free anyone from cleaning. It made cleaning a daily expectation for one person instead of a seasonal project for many.
AI agents won’t free your company from work. They’ll make 10x output the daily expectation for a fraction of the team. Quarterly releases, 48-hour response times, customer-managed deployments — those were the old standards. They’re already dead. And the new standards are being set right now by the companies that understood the Cowan Paradox before everyone else.
Plan accordingly.
