Wendell Berry’s Nine Tests for Any New Tool

← Back · ACT III · Chapter 16: MAVEN: Controlled Evaluation

Wendell Berry died on August 31, 2026, at 92, at his home in Port Royal, Kentucky. He farmed with horses and wrote with a pencil, and in 1987 he published an essay, later reprinted in Harper’s, explaining why he was not going to buy a computer. A number of people had told him one would greatly improve things. He answered with nine standards a new tool had to clear before he would bring it into his work.

The nine are usually remembered as a list of eccentric specifications. Taken as an evaluation, they hold a surprise. Only the third of the nine asks whether the new tool does the work better. The other eight ask what the tool would cost the arrangement it was entering.

The Nine, and What Eight of Them Measure

Cheaper than the tool it replaces. At least as small in scale. Clearly and demonstrably better at the work. Using less energy. Using solar or bodily energy where possible. Repairable by a person of ordinary intelligence with the necessary tools. Purchasable and repairable close to home. Sourced from a small shop that will take it back and service it. And last, not a replacement or disruption of anything good that already exists, family and community relationships included.

Set the third one aside and look at what remains. Money, scale, energy, independence, proximity to the people who can fix it, and the working relationships that already functioned before the tool arrived. What Berry evaluated was what adopting the machine would cost the arrangement he already had, which in his case was his wife Tanya at a Royal standard typewriter bought new in 1956, reading his drafts as she typed them.

The Test No Evaluation Checklist Contains

Compare that to how organizations actually assess a capability today. The modern vendor evaluation is a long and serious document, and it scores the vendor: financial stability, security posture, support model, references, roadmap, total cost, contract terms. Some of it is very good. None of it asks what already does this work, how well it does it, or whether the new capability understands what it is replacing before it replaces it.

That is the displacement test, and it is missing almost everywhere. An organization adopting a capability counts what arrives and rarely counts what stops. The handoff that used to work because two people had done it together for six years, the judgment that lived in someone’s head and never made it into a process document, the check that happened informally and caught things: all of that is inside the boundary the new capability is drawn around, and none of it is counted.

When critics wrote to Harper’s arguing that a computer was simply cheaper labor than a wife, they were answering an essay that had already named the stake. If the computer took over his wife’s typing, Berry wrote, “what would be superseded would be not only something, but somebody.” That is why his ninth standard exists, and it is the one a capability standard cannot do without. A tool can clear every performance test and still take more out of the working arrangement than it puts in, and nothing in a conventional evaluation would show it.

Why His Test Could Say No

There is a second thing worth taking from the essay, and it is about provenance rather than content. Berry wrote his nine standards before anyone was trying to sell him a computer, and he published them, so that anyone offering him one afterward had to answer a test they had no hand in writing.

Most organizations reverse that order. The business case is built with the vendor in the room and often built by the vendor outright, and the standard gets written after the tool has been chosen, to justify the choice rather than to test it. Criteria written before the purchase can still say no, which is seldom true of criteria assembled after a demonstration has already persuaded the room.

Writing your criteria before you engage a vendor is ordinary procurement advice and most buying teams will tell you they already do it. The harder question is what goes in them, which is where I wrote in The US Government’s AI Contract Clause Tests Everything but Value. about a federal clause that gets the authorship exactly right and then attaches its only real consequence to the supplier’s conduct rather than to the buyer’s result. Owning the test is not the same as pointing it at the right thing. Berry’s nine are worth reading precisely because eight of them point somewhere a modern test does not look.

What This Means for a Buyer

None of this is an argument against buying things, and reading it as one repeats the mistake Berry complained about in his own critics, who argued with the essay he had not written. The durable part is narrower than the technology it was aimed at.

If you are adopting a capability, write down before the first demonstration what currently does this work, how well it does it, and who would notice if it stopped. Keep that alongside the performance criteria and give it the same weight. Then, at the review you scheduled before signature, ask both questions rather than one: did the new capability deliver what was projected, and what did it displace that nobody costed.

A standard that counts only what arrives will tell you a capability worked. It cannot tell you whether value was created or simply moved from somewhere that was not being measured.

References: Wendell Berry, “Why I Am Not Going to Buy a Computer,” New England Review and Bread Loaf Quarterly, 1987, reprinted in Harper’s Magazine with reader letters and Berry’s reply. Death and biographical detail from Kentucky Lantern, August 31, 2026, and NPR, September 1, 2026.

Keep reading: ACT III · Chapter 16: MAVEN: Controlled Evaluation