pro-human output ceilings

I have been a tech worker for most of my working life. I started intuiting early that designing systems to free us from grudge work would reclaim time for the parts I actually love — collaborative workshops, sitting with the people on the end of the experiences I was hired to design, learning what impact my judgment calls had on their one precious mortal life. I got into design systems out of a deep love of colleagues and the people who would use what I made. I love humanity and I love tools. And we humans love tools that help us solve real human problems.

Today AI workflows could, with better leadership and a more considered timeline, hold the promise to relieve us from so much labor. A future of three-day work weeks. Spaciousness and inventiveness. As a tech optimist with integrity, I hold that vision and work backwards from it.

But that's not what happened. The pitch was simple: same output, less grind, more room for the thinking that actually matters. What happened instead is that the time "back" was immediately absorbed. Not by rest or deeper thinking but more output. Congrats, your reward for working faster is more work. When a team demonstrated it could produce a deliverable in two days instead of ten, leadership didn't return anything to the team. They just asked for five deliverables instead of one.

This pattern is documented. An eight-month ethnographic study from UC Berkeley, published in Harvard Business Review in February 2026, found that AI intensified work in three ways: people expanded their scope into tasks that weren't originally theirs, work seeped into natural pauses, and workers ran multiple AI threads simultaneously with no recovery time between cognitive loads. The most unsettling part: much of this was self-imposed. Nobody mandated it. People voluntarily filled the space AI opened — and then that expanded output became the new baseline. The ratchet turned without anyone pulling the lever. I think the levers are implied...everyone is competing assuming their AI-enhanced best is the new floor to maintaining their job.

Meanwhile, 96% of C-suite leaders expect AI to boost productivity. Only 13% have an AI strategy. And 77% of employees using AI say it has decreased their productivity and added to their workload, according to the Upwork Research Institute. C-Suite leaders need to touch some grass with the rest of us.

Intuit laid off 17% of its organization. I worked there and was trained in "Ethical" AI usage before we were allowed and later encouraged to use their sandboxed platforms. 40% of my Mailchimp colleagues, some of which had over a decade of tenure there...were let go. I know that some of their livelihoods were cut to make room for more AI investment. Honestly that makes me want to throw up. I feel heartbroken. I feel such a sense of loss. I was laid off from Mailchimp in February of this year, along with a few others, and expected more layoffs were coming. But inside of Intuit none of the other brands have the heart and humanity and business intelligence from executing on those well like Mailchimp. What they just threw out is of particular value in this period of workslop.

As a design systems practitioner, I genuinely enjoy what these tools have done to parts of my practice. Scaffolding the basics is faster than it has ever been. But if I can go from ninety days to a solid v1.0 in five days, my first instinct is not to ratchet up output. The conversations, the deep embodied understanding, the building of trust, the serendipity that leads to co-ownership — those fall out of scope when the calendar fills back up.

I started building design systems in 2014 and it wasn't because I was excited about documentation sites. I was interested in transforming confidence. I wanted people to lean into their work knowing the ground beneath them was solid. I have always approached design systems as a human and organizational transformation tool, not an npm package. The documentation is part product, part field notes from collective decision-making.

We've seen this play out before. Google's 20% time was, in practice, an output ceiling — organizational permission to not be maximally productive, to follow a hunch without justifying it in sprint review. Gmail & Adsense came out of that slack.  And then Google killed it. The products born from spaciousness became among its most valuable assets. The spaciousness itself was eliminated as an inefficiency. The ratchet turns in one direction.

I think we need an output ceiling. A deliberate decision that when AI compresses the time required for existing work, you do not proportionally increase the volume expected. You cap it — and you direct the remaining capacity somewhere intentional. Depth over breadth. Protected experimentation. Skill development in judgment: reading a room, synthesizing conflicting stakeholder needs, recognizing when a technically correct solution is wrong for the context. These skills atrophy when every hour is optimized for higher output of assets.

An output ceiling is not really a go-slow policy. It means designing the organizational response to AI with the same intentionality you'd bring to any other strategic decision, rather than letting it emerge by default, which is always "more." I know my own tendency when I feel myself making more progress than I expected is always to overload my plate, and this is something I've had to actively hedge against. Mindful productivity always beats the bingy bursts I've tended towards in my career.

I have long been annoyed that honed instincts get penalized under hourly pricing. I moved to value-based pricing years ago because if I finish what I committed to on time, I'm accountable — and I'm allowed to move between producing actively and reflecting spaciously. This makes my work sustainable. It keeps me in this industry long enough to know whether AI output is useful or just overconfident. That judgment doesn't auto-generate.

The people best equipped to verify AI output, direct it toward real problems, and recognize when it's confidently wrong are experienced practitioners with years of lived consequence. They watched real users struggle. They developed instincts that can't be prompted into existence. These are also the people most likely to have options. When the pace becomes unsustainable, they leave. The loss isn't immediately visible — the deliverables keep shipping. But eventually the people capable of distinguishing velocity from value are gone, and the organization discovers that judgment was never the expensive part. It was the thing making the rest of the process worth doing. I have loved mentoring people in my career because I love people. It really sucks to imagine people graduating into an industry where Claude is being trusted by their leadership to take on the role of coach, mentor, principal and implementer...Humans toiling want their toil to have meaning.

Andrej Karpathy recently described looking at AI-generated code and feeling a "mass heart attack coming on" — even though the code technically works.  The structure is wrong in ways that take experience to detect. When the people who can tell the difference leave, who remains to notice? The companies that build in spaciousness aren't being generous. They're being strategic about a scarce resource that takes years to develop and can leave in two weeks.

I searched for a named organization that has publicly implemented output ceilings as a deliberate AI strategy. I couldn't find one. Cal Newport recently wrote about the Jevons Paradox: when you make a resource more efficient, demand for it increases. Company A invests in sustainable AI integration. Company B squeezes maximum output from every AI-augmented hour. In the short term, Company B wins every quarterly review. It's a classic collective action problem. The individually rational choice — maximize output now — produces a collectively irrational outcome: burnout, talent flight, degraded quality that everyone suffers from but no single company can fix alone. It makes me queasy watching it happen.  David Dylan Thomas puts it well: if you pay a bounty on finding snakes, you don't reduce the snake population — you create snake breeders. When companies track token usage or prompt volume as productivity proxies, they're paying the bounty. The metric goes up. The actual goal drifts sideways.

The AI consultants will sell you acceleration. But..who's gonna design the brakes?