We spend a lot of time talking about AI taking jobs. We spend almost none asking whether they are the right ones.

Before automated switching, there were telephone operators, plugging and unplugging cables into boards while supervisors timed their bathroom breaks. When the technology caught up, nobody mourned the loss of that task.

This is the best application of automation: remove the tasks most tedious, backbreaking, and uninspiring.

It’s a comforting promise. But based on the data, it’s not necessarily what’s happening.

Our relationship with work, today

PayScale surveyed 2.7M workers across 500+ occupations with two questions:

1. Does your job make the world a better place?

2. Are you satisfied with your job?

A majority of Americans say yes.

The chart below shows occupations across the high meaning, high satisfaction range, weighted by median salary for that occupation.

Meaning and satisfaction are only lightly correlated with salary (r=.32, r=.42).

Both clergy (98%, 90%) and surgeons (96%, 83%) rank among the most fulfilled by their work. Meanwhile, the former earns $46K and the latter, $365K.

Compare that to airline pilots who also rank among the highest paid at $281K but find their work less fulfilling (53%, 74%) than personal care and service workers who make $38K (83%, 81%).

There are also a multitude of low fulfillment occupations, like parking lot attendants (5%, 41%), that feel akin to a modern switchboard operator. Should this be the work we focus on automating?

Where we see AI today

In March 2026, Anthropic published Labor Market Impacts of AI, introducing observed AI exposure, a measure of where LLMs are actually used in real work. It asks: of the tasks AI could speed up, which are seeing adoption in practice? They then mapped this usage across tasks and linked it to 756 occupations.

The headline: the vast majority of work (and workers) remain untouched.

But, of the work that is impacted by AI, a lot of it is that low-fulfillment work.

(Fulfillment here is a function of meaningfulness and satisfaction.)

The largest cohort of impacted workers is customer service representatives. 70% of their tasks covered by AI. Not surprising, as we’ve been moving in this direction for some time.

There’s even a slightly negative correlation between AI-impacted work and job fulfillment (r=-0.21). That sounds like good news.

It isn’t. The relationship is weak. And it reflects current model capabilities… It’s not that targeting unsatisfactory work on purpose.

To understand AI’s real impact, we’ll need to fast forward and look at where the technology is going.

Where AI is headed

There’s a second measure worth looking at, theoretical AI coverage. It starts with the tasks that define each occupation and asks: could an LLM do this at least twice as fast? Add those up and you get a ceiling for how much of the job is automatable.

The gap between today and that ceiling is large but informative.

For example, ~95% of business and financial operations tasks are theoretically automatable while only ~18% are today. Conversely, building and grounds maintenance are largely unaffected. Only 2.5% of tasks today and up to ~4% at full potential.

There’s no strong correlation between AI-ification, fulfillment, or even population size.

But patterns emerge when we group by nature of work:

• Digital/Knowledge — Work that is analytical, text-based, or computational

• Relational/Professional — Work defined by judgment, relationships, or perspective

• Care/Service — Work centered on serving or caring for people

• Physical/Manual — Work that requires building, moving, or maintaining

Here, we compare % jobs at risk, job fulfillment, and salaries.

What sticks out: the labor pool in physical/manual work is largely untouched.

Meanwhile, digital/knowledge workers see the largest potential impact, while also being paid well and pretty fulfilled.

This table summarizes. Each category is weighted to the highest value across each variable.

It’s worth noting, the most fulfilling work, taking care of others, is not at risk of total automation but is likely to feel impact. Hopefully AI improves these workers’ experience (and pay!).

What this means

AI follows the money, not the misery

The US services economy is largely a digital one, well suited to AI applications.

Within digital and service jobs, AI coverage and salary are correlated. The business case is simple: automating an hour of work that bills out at $200 is a lot more attractive than the one that pays $20.

Accordingly, AI is doing little to assist rental counter clerks or parking lot attendants, among those with the least fulfilling and lower paid jobs. Instead, it’s concentrated in higher-paid, more fulfilling knowledge work.

As we do not yet know if AI will displace digital jobs or level them up, there’s potential for the fulfillment (and salary) gap to widen further.

The Physical AI gap is real but temporary

There’s some bias in this analysis.

The underlying research, much produced by Anthropic, is scoped to digital AI/LLM capabilities and so the conclusions follow.

For the same reason, you wouldn’t see research from Skild or Pi describing their impact on accountants.

That said, Digital AI is far ahead of Physical. It’s faster, cheaper, and easier to spin up a software product with an LLM than to teach a robot a task; building an agent to hail a Waymo is trivial compared to installing Waymo’s system in your own vehicle. (When I say trivial, one is possible and the other is not.)

It’s not just that the work is harder to automate, but the economics are tougher (real capital costs in hardware, deployment, maintenance) and the failure modes are critical too.

Digital diffuses faster.

So, in the near term, this research is the relevant research, but it won’t always be.

Where AI meets reindustrialization

At Grid, we get a lot of inspiration from where supply can’t meet demand across the industrial landscape. Specifically, where there exists acute labor shortages and stalled productivity gains. We look to technology as a bridge.

But what we see in this research is that today’s reindustrialization efforts depend on the physical/manual workforce, where fulfillment is lower and AI has barely arrived.

The question becomes: where should we focus?

We still see a lot of opportunity in vertical AI solutions across the industrial sector. Low hanging fruit like replacing outdated systems or digitizing manual workflows to start. And that’s before considering labor shortages and upcoming baby boomer retirements. But we also focus on outsourced services, vertical integrators, and physical automation.

Importantly, these markets work differently, and workers view their work differently than digital-first ones. A strong business case is critical to access a market. Adoption depends on understanding the incentives of the people currently doing the work.

If that excites you, we’re excited to talk.