Framework Automation · Los Angeles
The sweater and the calendar.
The machine can save minutes. The factory's more interesting product is time: the option to wait before yarn becomes the wrong SKU.
Procedural illustration · see the actual sweater
This is not a story about an American machine making labor cheap. It is a story about whether proximity makes waiting valuable enough to cover a higher cost.
- Domestic content
- 11%
- closest Commerce category · 2023
- Possible time advantage
- ≥5 mo
- one USITC small-batch case
- Modeled option value
- 5.4 pts
- of retail · illustrative scenario
01 · The baseline
The machine arrived. The industry left anyway.
In the Commerce Department's closest category—apparel, leather, and allied products—imported finished goods accounted for 88 percent of US purchases in 2023. Domestic content accounted for 11 percent. The remaining point was imported content inside domestic production.
BLS supplies the longer view. From 1990 to 2025, real apparel-sector output fell to 10.0 on an index where 1990 equals 100. Hours fell to 9.9. Output per hour ended only 1.4 percent above its 1990 level. The old industry did not become a tiny, spectacularly productive automated sector. It mostly disappeared.
Sources: US Department of Commerce, Purchased in America 2023; BLS major-sector productivity series for NAICS 315. Indices are not physical garment counts.
02 · Minutes versus months
The machine is not the interesting clock.
Whole-garment knitting can eliminate panels and side seams. It is not new: SHIMA introduced commercial Wholegarment equipment in 1995. In one 2012 study—not Framework, and not this sweater—measured process time fell from 123 minutes to 113. Cutting-and-sewing time was 13 minutes lower; knitting time was three minutes higher.
Ten minutes is useful. It is not an explanation for reshoring an industry. The stronger wager is on the calendar. One USITC case, Boathouse Sports, reported under a month from domestic order to shipment and at least six months from China. Its average order was only 17 garments: a favorable small-batch case, not an industry average and not a Framework lead-time measurement.
Framework advertises shorter lead times and pricing competitive with Asia. Founder Denver Rayburn argues that the usual unit-cost comparison misses a “complexity tax”: an early forecast turns uncertainty across sizes and styles into finished inventory. No public matched quote or customer result shows how large that saving is.
Sources: Joel Peterson, 2012, 12-gauge case study; USITC staff paper, 2018; Framework Automation and Rayburn's “Complexity Tax” essay. The measured comparisons are not observations from Framework.
03 · Price the wait
Proximity is an option. Options have limits.
In a normalized inventory model, if demand six months out has a 30 percent coefficient of variation, moving commitment from six months to one supports a unit-cost premium of about 5.4 percentage points of retail. If the alternative source takes three months, the modeled premium falls to 2.7 points. At two months, it is 1.5.
These are scenarios, not Framework estimates. The important feature is the slope toward zero: a faster offshore or nearshore supplier captures much of the same timing value. The model also omits capacity constraints, minimum orders, setup costs, lead-time variance, and observed markdowns.
Timing-only newsvendor model. Retail = 100; landed offshore cost = 35; end-season recovery = 10; lognormal demand; the chart excludes financing.
04 · Follow the risk
The risk changed form. It did not disappear.
Framework advertises “zero inventory risk.” Its standard agreement generally requires customers to prepay the total yarn cost. On termination, a customer may also owe for unused Framework-supplied material and unshipped goods.
That can still be a better form of risk. One cone of navy yarn can serve several future sizes; a finished medium sweater cannot. But less specific inventory risk is not zero inventory risk.
The public case is narrow but plausible: automated domestic knitting can win where demand is uncertain, variation is high, and the next-best source is slow. Framework has not published the operating numbers needed to show that it does.
Source: Framework standard manufacturing agreement, accessed August 2026.
Style / size risk ↓commit the SKU later
Material capital remainssomebody still owns the yarn
The verdict
The sweater proves a factory exists. Reindustrialization still needs the spreadsheet.
Five measures would make the claim testable:
- Matched landed unit cost
- Median order-to-ship time
- First-pass yield
- Machine utilization
- Markdown and stockout performance
I'm keeping the sweater. I'd still like the spreadsheet.
Reporting notes
What the evidence can—and cannot—show
The archive date keeps this experiment below newer posts; reporting was updated September 1, 2026. Framework's product page describes an 8-gauge, one-piece cotton sweater made in Los Angeles. The garment establishes the object, not factory cost, scale, or customer results.
The domestic-content shares come from Commerce's Purchased in America 2023. The time series use BLS real apparel output, hours, and labor productivity, each rebased to 1990. These are economic indices, not garment counts. The local snapshot was retrieved September 1, 2026. Download the annual observations plotted above.
The 123-to-113-minute comparison is a 12-gauge case in Joel Peterson's 2012 thesis. The lead-time comparison is Boathouse Sports in a 2018 USITC staff paper. Neither is a Framework measurement.
The timing model assumes retail price 100, landed offshore cost 35, end-season recovery 10, lognormal demand, and forecast error declining with the square root of lead time. It does not model multi-SKU demand pooling. Denver Rayburn's complexity essay motivates the hypothesis; Framework's agreement supplies the yarn-payment terms.
Model table and reproducible files
| Six-month demand CV |
|---|
| Versus 6 months |
| Versus 3 months |
| Versus 2 months |