The reading room · Digna Legi
Impact, agency, and taste
/100
80–100: high value. 70–79: worth the time. Below 70: below the usual publication threshold.
Evidence-reviewed score based on available publisher text. Evidence is substantive, though examples are centered on one high-performing AI company context.
Scores reflect one reader’s profile, not an objective quality rating. Best is a separate personal selection.
How scoring works →This brief · about 3 min with detail
Why read this
Kuhn separates high-leverage technical work into agency that drives outcomes and taste that chooses bets worth pursuing.
AI brief · Checked against source text
The main idea
Kuhn argues that once baseline technical ability is already high, the scarce differentiator is leverage: choosing and executing work with unusually high impact per hour. He divides this into agency, the drive to make outcomes happen, and taste, the judgment to pick problems and approaches that will work. The two fail differently: agency without taste chases the wrong target, while taste without agency leaves good bets unrealized.
Go a little deeper
Leverage often hides in unglamorous work
The essay pushes against the idea that only glamorous research bets produce major leverage. Tooling, documentation, developer loops, system design, and data inspection can multiply other people’s output or redirect priorities. Kuhn especially emphasizes that rote-looking data review is not junior work when the goal is anomaly detection: senior people may spot severe issues because they understand what matters and notice deviations others would normalize.
Agency means owning the root outcome
Kuhn’s version of agency is not mere initiative. It requires reasoning backward from the real goal, noticing when the assigned task is a bad route to that goal, and taking responsibility for the whole path to success. The distinction matters because a person can complete the stated assignment while making the underlying system worse, or can preserve value by changing, shrinking, or cancelling the task.
Non-obvious bets need a portfolio
High-leverage projects are often not obvious in advance; otherwise they would probably already have been done. That creates a predictable social problem: useful bets may meet skepticism because their value is uncertain or execution looks tedious. Kuhn’s practical answer is not blind defiance but calibrated autonomy: state intended action, leave room for objections, and balance speculative bets with lower-risk work until trust in judgment is earned.
Taste improves through explicit prediction
Taste is presented as a trainable predictive faculty, not a mystical aesthetic trait. Kuhn recommends making forecasts explicit before choosing: what will happen if this experiment, tool, document, or design goes this way? The mechanism is feedback. By later comparing predictions with reality, a person updates the models and search heuristics that determine which problems and solutions they even consider promising.
A case from the article
The scheduler that should not be built
During a work trial, Kuhn was assigned to build a custom scheduler for a compute cluster. After observing the team and thinking through the design, he concluded the project would create a maintenance burden for an already overloaded cluster team. The useful move was not better execution of the assignment, but recognizing that completing it could be net-negative for the root goal of keeping the clusters working well.
How the case is made
The case is made from workplace observation at Anthropic, supported by concrete project anecdotes and a practical taxonomy of behaviors.
Where the idea has limits
The argument is explicitly framed around a highly selected research and engineering environment, so its priority order may not transfer to settings where basic technical competence is still the main bottleneck.
A question to take away · from Digna Legi
Where do you repeatedly notice others missing something obvious, and is that frustration actually evidence of unusually good taste?
What the original adds
The source includes more tactical advice on permission, project scheduling, prediction, retrospectives, and data inspection than this brief can preserve without reproducing the whole essay.
About this brief
AI-written, then separately checked for source support, useful detail and clarity. The author’s claims and our editorial question are kept separate. The original remains the author’s work. How we select and summarise →
How was this brief?
Rate this summary, separately from the author’s article.
Optional. Saved in this browser; shared only if you allow analytics.
How was the original article?
Rate the author’s original after reading it.
Optional. Saved in this browser; shared only if you allow analytics.
Digna legi. Worth reading.