AI Without a Degree: What the Dallas Fed Found About Good Jobs

About half of American workers don't hold a bachelor's degree. Almost none of the AI-and-jobs research gets written about them. On August 12, 2026, a Dallas Fed senior economist stood up at a Texas Tribune symposium in Waco and said that out loud — then put numbers behind it.

Roberto Coronado's talk was billed as a workforce development update. Buried in the middle was the part that actually matters if you drive a truck, answer phones, or rebuild engines for a living.

The split inside "no degree required"

Coronado's colleagues looked at a specific slice of the labor market: jobs that don't require a bachelor's degree but still pay self-sustaining wages. The Dallas Fed calls these good jobs. [Fact] Their finding was that AI doesn't hit that slice evenly — it splits it. Administrative and IT roles may be greatly affected. Truck drivers and auto mechanics may see less immediate change.

Our own occupation modeling puts a size on that gap, and it is not subtle.

On the exposed side, administrative assistants carry a 60% automation risk in 2026, bookkeeping clerks 70%, and customer service representatives 72%. [Estimate] On the other side, truck drivers sit at 12%, auto mechanics at 15%, and electricians at just 7%. [Estimate] Between those two clusters sit computer support specialists at 38% — the IT half of Coronado's warning, and a reminder that "works with computers" stopped being a safe category some time ago.

The headcount behind those numbers is the uncomfortable part. Roughly 10.5 million Americans work in the four high-exposure office roles above. About 5.7 million work across trucking, auto repair, welding, electrical, industrial maintenance, HVAC, and construction labor combined. The exposed group is nearly twice the size of the sheltered one.

The credential math, and the trap inside it

Coronado spent more time on the fix than the problem. Short-term credentials — anything completed in under two years — are the Dallas Fed's leading candidate for connecting people to good jobs.

[Fact] Using Census data controlled for demographics, education, and years of experience, his colleagues found a credential adds roughly $5,000 per year in wages for someone holding a high school diploma or associate's degree. For someone who already has a bachelor's, the bump is about half that: $2,600. The return on a short program is larger for the people with less formal schooling, which inverts how most people assume education stacks.

Now run that $5,000 against real median wages, and something odd falls out.

For office clerks, whose 2024 median pay is $38,940, a credential is worth about a 12.8% raise. For customer service representatives at $39,680, roughly 12.6%. For electricians at $61,590, only about 8.1%. For industrial machinery mechanics at $60,000, about 8.3%.

The proportional payoff is biggest exactly where the AI risk is highest. That is not a coincidence — low-wage clerical work is cheap to improve on and cheap to automate for the same reason. It also means a credential earned inside an exposed occupation buys a raise on a shrinking base. BLS projects administrative assistant employment down 10% through 2034 and office clerks down 7%, while electricians grow 11% and industrial machinery mechanics 16%. A 12.8% raise in a role losing 10% of its positions is a worse trade than an 8.1% raise in one adding 11%.

Where we'd push back on the Fed

Coronado's prescription was soft skills. His words: jobs with high AI exposure scores still require work activities AI can't do well — active listening, establishing interpersonal relationships. Investing in those, he argued, could offset negative exposure.

We think that's half right, and the half that's wrong is worth naming.

Customer service representatives are the most interpersonal job on our entire high-exposure list. Active listening is the job. And they carry the highest automation risk of any occupation discussed here, at 72%, rising from 55% in 2023. [Estimate] Soft skills did not shelter them. What separates protected interpersonal work from exposed interpersonal work isn't warmth — it's whether the human context is scripted and remote, or situational and physical. A welder explaining to a foreman why a joint failed is doing interpersonal work AI can't reach. A rep reading from a decision tree is doing interpersonal work that already got reached.

There's a second wrinkle in the "trades are safe" read. Safe is a snapshot, and the snapshot is moving. Between 2023 and 2026 our modeled exposure for auto mechanics went from 8% to 20% — a 150% relative increase. Welders went 7% to 19%. Truck drivers doubled, 6% to 12%. Over the same window administrative assistants moved 45% to 64%, a relative increase of only 42%. [Estimate] The exposed jobs are further along a curve. The sheltered ones are climbing it faster from a lower base. Anyone treating a trade as a permanent hiding place is reading the level and ignoring the slope.

What this looks like on a shop floor

Two weeks before the Waco talk, Dallas Fed President Lorie Logan toured the GE Vernova Houston Learning Center. Trainees there learn to inspect and repair gas turbines — one machine can power 600,000 homes, or a data center. The facility runs without air conditioning through a Houston summer, on purpose, because field deployments mean ten-hour days in that heat.

An executive told Logan that a high schooler entering the program can go from working as a barista to field engineer in seven to eight years, if they stay with it. [Claim] That's one company's account, not an audited outcome, but it's the shape of the pathway the credential research describes.

The same trip produced a detail worth sitting with. Houston leaders described a pilot that uses AI chatbots as career counselors, matching job seekers to employers by skillset. [Fact] One participant framed it as AI helping to clean up its own mess. Ask the counselor what you're good at, and it names careers you didn't know existed. It's a small thing. It's also the first version of AI showing up on the helpful side of a workforce ledger it has mostly been debiting.

Texas is a reasonable place to watch this. State payroll employment grows about 1.9% annually against 1% nationally, unemployment sits near 4.5%, and the Dallas Fed forecasts 2.0% job growth for 2026 with an 80% confidence band of 1.5 to 2.5%. Growth this year concentrated in four sectors — professional and business services, leisure and hospitality, construction, and energy. Construction's surge is largely data center building, which is to say the AI boom is currently creating more jobs pouring concrete than it is destroying in the buildings it fills.

What to actually do with this

If you're in an exposed clerical role, the credential is worth taking, but aim it out of the occupation rather than deeper into it. The $5,000 average lands hardest as a percentage in the jobs with the weakest ten-year outlook.

If you're in a trade, the slope matters more than the level. Exposure doubling off a small base still leaves you far safer than a bookkeeper — but it means the tasks that get automated first are the diagnostic and documentation ones, not the wrenching.

And if a program near you is part of a coordinated pathway — employers, community college, and case management wired together the way Upskill Waco does it in McLennan County — that coordination is doing real work. Coronado's own framing was that fragmentation, not the absence of training, is the binding constraint.

Where these numbers stop

The occupation exposure and automation risk figures above are model estimates, not observed adoption counts. They tell you the shape of the risk, not a headcount of jobs already gone.

The Dallas Fed's $5,000 figure has a limit its authors named directly: it's a cross-sectional average from Census data with controls, not a randomized causal estimate. People who enroll in credential programs may differ from people who don't in ways the controls don't capture. Coronado said plainly that increased earnings are not guaranteed, that industries and employers value credentials differently, and that the value shifts over time with business needs. That caveat belongs on the sentence, not in a footnote.

Sources

• Roberto A. Coronado, "Developing a strong workforce requires collaboration," Federal Reserve Bank of Dallas, August 12, 2026 — https://www.dallasfed.org/news/speeches/speeches-leaders/2026/260812rc
• "How Houston finds workers in an AI boom," Federal Reserve Bank of Dallas, July 29, 2026 — https://www.dallasfed.org/fed/leadership/logan/listeningin360/2026/2604

Update History

• 2026-08-15: Initial publication. Analysis of Dallas Fed research on AI exposure in non-degree occupations and short-term credential returns, cross-referenced against our 2026 occupation exposure estimates.

---

AI-assisted analysis. Source documents were read in full; occupation exposure and automation risk figures are our own model estimates and are labeled as such. Wage and employment projections are from the U.S. Bureau of Labor Statistics.