I set out to test a simple idea: that as employer leverage returns, the well-being budget migrates from the system to the self. The clean version of that story — a measurable substitution of individual perks for structural ones — the data could not support; the enumerated-benefits signal on careers pages is too sparse to carry it. But something else was there, and it turned out to be more interesting.
The concession moment
In 2020, companies’ care talk — language about rest, balance, mental health, sustainable pace — jumped to roughly 3.7× its prior level. It did not rise alone. DEI talk rose in near-lockstep: within companies, the two move together at a median correlation of +0.53 (95% CI [0.27, 0.70]), positive in 15 of 16 firms. Care and diversity read like two faces of one concession.
And the concession tracked worker bargaining power. Overlaid on the JOLTS quits rate — how freely workers were leaving jobs — care and DEI talk rise into the Great Resignation and recede as quits fall. Standardized so all three sit on one scale:
The cleanest part of this is the recede: as the quits rate fell from 2.8 back to 2.1 (2022→2024), care talk fell with it. The 2020 spike itself is harder to read cleanly — it coincides with the pandemic, so some of that surge is lockdown wellness-talk rather than a pure labor-market response. The deflation is the part that isn’t confounded.
And the bundling isn’t carried by a few firms. Take it one company at a time — each firm’s own care and DEI talk, standardized — and the same paired rise-and-recede shows up in 15 of 16, from Salesforce (r = +0.79) down through GitLab, Amazon, and Palantir. Only HubSpot breaks the pattern. Pick a company:
The performance regime never flinched
If 2020 were simply a year companies got more values-forward across the board, every axis would have lifted. It didn’t. Measuring each theme’s 2020 deviation from its own historical average, only the worker-concession axes spiked. Performance, meritocracy, and control language barely moved.
That’s what makes “concession” the right word. The care and DEI surge was targeted — the language a tight labor market extracts — while the performance regime, the part that serves the employer, held perfectly steady underneath it.
Talk is cheap; benefits are slow
The rhetoric was an event — a sharp 2020 spike. The actual care benefits were a slow build: their prevalence climbed gradually from 2016 to 2019 and had already plateaued before the talk spiked. Two different clocks. The volatility lives in the framing, not in the obligation — a company can re-word a careers page in an afternoon; it cannot re-staff a leave policy that fast. Talk flexes with the moment; substance moves on its own slower schedule.
The care that survived
Here is the finding I didn’t predict. Not all care benefits behaved alike — they split by who absorbs the cost. Mental-health benefits (EAP, therapy stipends, wellness apps — things the employee uses on their own, coded individual-locus in 19 of 20 cases) climb steadily, up and to the right, still peaking at the end of the record. Family and caregiving benefits (paid leave, childcare — where the organization absorbs the coverage, coded structural in 105 of 125 cases) crest around 2019 and recede.
This is individualization — not as the aggregate ratio I couldn’t measure, but as divergent survival. The care that endures is the kind you carry yourself; the care the company has to absorb is the kind that fades. The well-being budget didn’t just shrink. What was left of it shifted from system to self.
The counts here are modest and this is self-presentation — what firms advertise, not what they provide. But the shapes are clean and they line up with the locus split almost perfectly, so I read it as a genuine illustration of the mechanism, not a fragile quantitative claim.
Whole-life benefits
There’s a second family of benefits worth watching: the ones that support your life outside work — the people you care for and the moments that don’t fit a calendar. Parental leave, childcare, elder and dependent care, fertility, lactation, identity-positive coverage, bereavement. Measured as a share of everything companies advertise, they rose through the mid-2010s, peaked around 2018 at roughly one benefit mention in seven, and have been a shrinking slice ever since — down to a low near 2024, with only a slight recovery after. That peak comes before the 2020 concession-talk spike: the family-benefits wave crested and began fading while the care-and-diversity language was still climbing — substance leading talk, and receding first. Parental leave is the sturdiest thread; lactation, bereavement, and identity-positive coverage barely register.
GitLab, up close
The public record rarely lets you watch this happen. GitLab’s is an exception — its handbook is version-controlled, so every edit to its Family & Friends Day (a company-wide paid day off) is timestamped and public. Reconstructing that history overturned the tidy story I’d been handed: there was no clean “pandemic measure → permanent benefit” arc. There was something more honest.
Two edits carry the whole thesis in miniature. In April 2021, for six days, the benefit was renamed “Pandemic Support Day” — the contingency made explicit in its name — and then quietly reverted, the ambiguity never resolved in writing. A company briefly said out loud that the commitment was temporary, recoiled, and buried it: legible-as-generous, obligation deniable. And in July 2022, the handbook added coverage requirements — some teams must stay back so others can rest. That is the exact moment a structural benefit privatizes: the synchronized day off only works if someone absorbs the cost, and the record shows GitLab naming who.
What I can’t say
I want to be precise about the limits. The aggregate individualization index — the headline this study was designed around — is too sparse to test: only 8 of 15 companies have enough enumerated benefits on both sides of 2022, at 2–5 items each. Fertility benefits appear in only four companies ever, so their trajectory is an anecdote, not a trend. The benefits data throughout is self-presentation, not provision. The quits rate is total-nonfarm, not the information sector specifically. And the leverage correlation rests on about ten annual points, with the 2020 spike confounded by COVID.
So the strong claim here is narrow and it is about language: a worker-concession bundle that inflated with the labor market and deflated as leverage returned, while the performance regime never moved. The benefits are texture — vivid where the shapes are clean, silent where the counts run out.
The care that’s left
Resilience, we’re often told, is a personal skill — something to build with an app, a stipend, a mindfulness subscription. What this record suggests is that the framing itself is the concession receding. When workers had the leverage to ask, companies talked about care as something the system owed them. When the leverage passed, the care that remained was the kind you were expected to manage on your own. Resilience is a systems problem; the language just stopped saying so.