Our Michigan workforce intelligence project just published its September edition — new labor data through August, graduation-horizon program planning, and cross-sector analysis of AI task impacts. I want to walk through what the numbers actually say, because the headline figure is the least interesting part.
Evidence cutoff: September 26, 2026 (America/Detroit). All figures below come from the September edition of the public research repo.
The rate is not the story
Michigan's August seasonally adjusted unemployment rate was 5.0%, against 4.1% nationally. On its own, that looks like a modest gap. But the report pairs it with the rest of the picture: participation at 58.8%, an employment-population ratio of 55.9% — and monthly employment and labor-force declines that both totaled 25,000.
That last detail is the one that matters. As the report puts it: a rate moving only slightly can conceal a material change in participation. When the labor force itself shrinks by 25,000 in a month, the unemployment rate stays deceptively calm. If you're deciding which training programs to expand or close, the rate alone will mislead you.
The broader U-6 measure — 9.1% for the July 2025–June 2026 window — captures more of the slack, but note the vintage: it is not an August reading. Recent graduates faced about 5.6% unemployment in Q2, with degree-based underemployment at 42%. These measure different populations and cannot be added together — a discipline the report enforces throughout.
Warning signals, not forecasts
Two yellow flags in the data: September's final national consumer sentiment index fell to 48.1 (from 51.7 in August), and Chicago Fed district respondents reported below-average hiring indexes even as manufacturing activity strengthened.
The report treats these exactly right — as warning signals to examine alongside payrolls, hires, hours, and employer orders, not as a forecast of Michigan job losses. That restraint is the whole point of evidence-centered analysis: say what the data supports, mark the boundary, stop there.
Five decisions for education and industry
The edition's core is five recommendations. They're worth quoting in compressed form because each one resists a lazy reading:
- Redesign narrow task training first. Standalone typing, data-entry, and routine transaction-processing curricula need immediate review — but preserve transferable business knowledge while adding data quality, controls, exception handling, and customer problem-solving.
- Retain foundations in AI-exposed professions. Software, accounting, legal support, engineering, media, and medical imaging face task change. Teach students to define requirements, evaluate outputs, and manage risk. AI exposure alone does not establish occupation decline.
- Prioritize Michigan demand and replacement needs. State projections show growth in industrial maintenance, cybersecurity, software, data science, nursing, and electrical work. And declining occupations can still have substantial annual openings — match capacity to verified regional demand, not national projections alone.
- Build skills bridges before layoffs. Assess incumbent skills, secure a real receiving role, pay for supervised learning, verify competence through work samples. Track wages, benefits, hours, and sustained employment — not just course completion.
- Plan by graduation cohort. Review short credentials for 2027, associate pathways for 2028, bachelor's programs for 2030, longer professional pathways for 2032 and beyond. These are decision horizons, not occupation expiry dates.
Number 2 deserves emphasis because it cuts against the loudest narrative in the room. The report's line is blunt: neither the unemployment rate nor an AI capability demonstration is sufficient to decide which programs to expand or close.
Radiology: adaptation, not obsolescence
The report uses radiology as its worked example of getting this right. The official projections do not establish radiologist obsolescence — Michigan projects growth for physician radiologists and radiologic technologists, and the ACR describes persistent workforce shortages. The prescription: preserve clinical preparation and add role-specific AI evaluation, imaging informatics, monitoring, and workflow assurance.
It's a template for how to think about every "AI will replace X" claim: check the projections, check the shortages, adapt the curriculum, keep the foundations.
The pipeline behind the numbers
One more finding, and it's about method rather than labor markets. On September 27 the project executed its first connector-triggered audited release: a content commit triggered the full GitHub workflow, and every published artifact was verified afterward — all 6 downloadable assets and 33 ZIP member hashes matched, plus manifest version, source commit, and PDF page counts.
No permission expansions, no copied tokens, no browser logins. The evidence assessment cleared its bar only after observed run data replaced prose assertions — and when a tooling incident occurred (two assessments colliding on a filename), it was documented in the open rather than smoothed over.
That's the standard we hold our own analysis to: every number traceable to a source, every release hash-verified, every limit stated. The September edition declares no program obsolete on a speculative date and commits no named institution to anything. In a space full of confident forecasts, that restraint is a feature.
The full reports, evidence ledger, and source registry are public at github.com/AXIOVEX/michigan-workforce-intelligence. The verified September release is here.