AI Is Humanizing and Dehumanizing Organizations
At the Same Time.

May 14, 2026 · 11 min read

A friend recently told me her AI assistant — a voice model, not text — was “the best listener in her life right now.” She meant it. It pays attention. It remembers what she said last week. It does not check its phone, does not compete for airtime, does not make her feel like she is wasting its time.

I had two reactions at once. The first was a quiet agreement. Most humans she talks to, including me, are worse at this. By every measure of what a good conversation is supposed to feel like, the bot delivers.

The second was a small, specific kind of dread, because the thing she was experiencing as more human than her actual humans was being delivered by something that has none of the qualities we mean by the word. There is nothing on the other side. No listener. No attention. The interaction is humanizing. The other party is not human. And once you notice that, the whole language we have for this technology starts to slip.

That is the question this post is about. Not whether AI inside a company is good or bad, but whether the word “humanizing” — which I will defend later in this piece — survives the case where the most human-feeling interaction you have all week is with a machine that has no inner life. The honest answer is: only barely, and only if we are very specific about what we mean.

Two definitions worth fighting over

Most arguments about AI at work fall apart because nobody agrees what they are measuring. Humanizing is when an organization uses AI to shift work toward judgment, empathy, presence, creativity, and hope (MIT Sloan's EPOCH cluster); democratize expertise gate-kept by seniority; shrink toward Dunbar tiers where trust replaces hierarchy; and let machines absorb surveillance and drudgery instead of administering them.

Dehumanizing is the operational opposite: AI tooling that strips judgment, context, and recourse out of decisions; commoditizes the human side of human contact; moves accountability into opaque systems; and shifts cost and cognitive load onto invisible offshore workers, customers debugging chatbots, or colleagues cleaning up your AI-generated documents.

Both are happening at the same time, often inside the same company. Neither definition is about the technology. They are both about deployment — which is the part leaders actually control.

Where AI is making work more human

BCG's 2024 survey of 13,102 employees found roughly half saved five or more hours a weekwith generative AI, and reinvested it into strategic work, experimentation, and new tasks at almost identical rates. Anthropic's Economic Index — admittedly a measure of Anthropic's own product, but the largest sample of real LLM use anyone has published — found augmentation finally overtook automation in January 2026:

Treat the number as directional. But it is a better dashboard than the ones most enterprises currently track, which is none.

Every employee gets a thought partner — and a quiet bill

Microsoft's 2025 Work Trend Index found 46% of employees already use AI to brainstorm, stress-test, and challenge their thinking. Reid Hoffman calls this superagency.

The catch arrives in two ways. Brynjolfsson's work on customer service agents found gains concentrate at the bottom of the skill distribution — 34% for novices, essentially zero for experts. And an MIT Media Lab EEG study found writers using ChatGPT showed significantly lower prefrontal cortex activation than those using search engines, with the gap persisting across sessions — what the authors call “cognitive debt.” If AI's biggest gift is to make junior people sound senior while the senior muscle atrophies, who is doing the actual thinking five years from now?

Where AI is making work less human

Most knowledge workers are not yet under per-minute surveillance, but the same dashboards exist in the seller productivity tools my industry buys every quarter. UnitedHealth's nH Predict tool allegedly drove premature termination of post-acute care; internal data in the lawsuit suggests the model was wrong about 90% of the time it was appealed. Air Canada's chatbot fabricated a bereavement-fare policy and the tribunal explicitly rejected the airline's argument that the chatbot was a “separate legal entity.”

“AI cannot do your job, but an AI salesman can 100 percent convince your boss to fire you and replace you with an AI that can't do your job. The point of AI isn't to make workers more productive, it's to make them weaker when they bargain with their bosses.”— Cory Doctorow, 2025

The middle of the org chart is being deliberately hollowed. Manager engagement, per Gallup, fell from 31%22%2022 → 2025 over three years — the steepest decline of any cohort, and the proximate cause of a global engagement collapse.

The quiet lie about AI layoffs

The inconvenient number is small and rarely cited.

What is actually happening is something else. IBM's CEO announced a pause on 8,000 back-office roles in 2023 citing AI; subsequent reporting found IBM's Bengaluru and Hyderabad job postings surged from 173 in January 2024 to 3,866 in early 2025. The work did not get automated. It got offshored, with an AI story bolted on. Sam Altman has publicly called the pattern “AI washing” — and it matters that it is the CEO of the company selling the cover story.

The argument I want to make: smaller, more human, but barbell

Coase wrote in 1937 that firms exist because using the market has costs — search, negotiation, contracting, monitoring — that are sometimes higher than the cost of doing the work in-house. The boundary of the firm is the line where those two costs meet. The NBER and Berkeley have now formalized what happens when AI agents push the market-side costs toward zero: the boundary collapses inward. But at the same time, AI raises the minimum efficient scale of the platforms underneath those firms.

The clearest way to see what those two forces produce together is one number — revenue per employee — across two firms in the same product category, separated by a generation of operating model:

One comparison can be cherry-picked. So here is the same metric across forty-five firms — public, private, hyperscale, mid-market, AI-native, solo. The shape stops being a coincidence:

The platforms below are bigger than they have ever been; the operators on top are smaller than at any point in modern industry; the mid-market — generic agencies, sub-scale SaaS, mid-tier BPO, body-shop consulting — has no home. Hyperscalers have $660-690B of capex committed for 2026. OpenAI has $1.4T in long-term infrastructure commitments. Anthropic alone is in the middle of a $50B American data-center buildout. And on top of that substrate, Midjourney crossed $300M in revenue with about 40 people, Cursor went from $0 to $1B ARR in under two years with about 150 people, and solo founders are now 36.3% of new US startups.

For an executive, “is AI making my firm shrink?” is the wrong question. The right one is which end of the barbell are we on, and what does the answer mean for the next ten people I hire?

What “more human” actually feels like — honest version

Small teams are more human in the ways founders care about — autonomy, signal, ownership — and less human in the ways employees care about — security, predictability, peer support. The Sifted 2025 founder mental health survey found 54% of founders reported full burnoutin the prior year. TechCrunch's reporting found the first burnout signs of this cycle are coming from the heaviest AI users, not the holdouts. The thesis is true from the founder's chair and ambiguous from the IC's.

The lever an executive actually controls

Daron Acemoglu frames the choice this way: AI can either deliver machine usefulness — making workers more capable — or machine substitution — replacing them. The lever has four indicators I track inside my own org:

• The augmentation-to-automation ratio — the way Anthropic publishes theirs externally. When a tool comes in, is it making my best people better, or am I using it to do their job without them?
• The shape of the time saved — does it become EPOCH work (judgment, empathy, hard customer conversations) or does it become more meetings and more email?
• The apprenticeship pipeline — if no one in the organization is mentoring the analysts and reps who will be seniors in 2030, the senior shortage of 2028 is a bill I am quietly running up.
• The workslop tax — am I receiving more documents from colleagues that I have to verify before I can trust them?

The cut

The right question was never whether AI is humanizing or dehumanizing organizations. The technology is doing both, and which one wins in any specific company is decided by deployment choices made in the next eighteen months — choices that will calcify into culture for a decade.

The bet I am making is that a meaningful number of organizations make the same choice for the same reason. Whether you are one of them is, today, still up to you.