RUGGETS / ISSUE 61Explore this issue ↓

READ. THINK. DISCOVER.

Seven worth reading.

Product, engineering, AI, philosophy, leadership, and life — selected from Digna Legi.

Start reading ↓
RUGGETS by Digna Legi: a terracotta ribbon breaking free from a rigid black grid.
THE SELECTION

From tools and systems to attention and agency.

7 reads

There is more worth reading than anyone has time to find. Each week or so, we search widely thanks to AI, read closely, take notes and choose seven pieces that repay your attention. Every selection comes with a concise editorial brief, the key idea worth carrying forward, and a direct link to the original.

01

Systems & signals

What we build. What we pay attention to.

01 / Product

And You Will Know Us by the Company We Keep

Eugene Wei — Remains of the Day

≈ 24 min · Timeless

93/100

Eugene Wei locates a major social-media failure in graph design rather than in generic complaints about feeds, features, or human nature. Western social apps, in his account, made the feed depend on the social graph: who a user knows, follows, or feels obliged to keep. That architecture helps solve cold start by pushing people to connect quickly, but it later turns those early ties into stubborn feed-shaping machinery.

The unresolved problem is that social graphs and interest graphs answer different questions. A relationship does not reliably predict what someone wants to consume, and people rarely prune connections once they harden into social facts. Feed quality, context collapse, and awkward behavior become consequences of the same bet.

02 / Engineering

Harnesses are Situated Agents

dbreunig.com

≈ 3 min · Current

76/100

The author argues that a coding harness is not just a nicer interface around an agent or a bundle of tools handed to it. A harness situates the agent: the core loop of prompts, plans, files, and subagents remains, but the harness governs the world around that loop. Sessions, environments, repositories, memory, skills, teams, organizations, and model choice all become part of what the agent can perceive and do.

That distinction matters because agent capability is not only inside the model. What enters context, what persists between attempts, where code runs, and how organizational knowledge is exposed can change the work itself. The question becomes less what the agent is and more what world it has been placed inside.

02

Decisions & perspective

The rules we use to make sense of things.

03 / AI

Turn And Face The Strange

Fly.io Blog

≈ 10 min · Current

83/100

Fly.io’s argument begins with a shift in what cloud infrastructure may be for. Its older logic centered on fast global hosting and developer-friendly deployment for public applications. Agent-built software, the piece suggests, points elsewhere: toward personal, adaptive, lower-audience programs created and revised by autonomous systems rather than carefully packaged for broad release.

If that change holds, the valuable unit is not always the polished app server. It may be the cheap, durable, short-lived computer where an agent can work, pause, resume, and throw results away. Sprites are presented as Fly.io’s answer to that pressure, not as ordinary sandboxes but as computers built around agent labor.

04 / Leadership

Advice for time management as a manager

Ben Kuhn

≈ 6 min · Timeless

82/100

Ben Kuhn treats new-manager time management as a job change, not a calendar problem. The individual-contributor habit of choosing the most important task and grinding through it breaks down when the work is externally scheduled, interruption-heavy, and often blocked on other people. The manager’s output is no longer mainly their own code; it is the team’s ability to make progress.

That shift makes ordinary productivity advice too small. Kuhn’s answer is calibrated expectations, ruthless prioritization, delegation that increases future capacity, and protected blocks for deep work. The tension is practical: a manager still has tasks, but treating personal throughput as the score can quietly damage the system they are now responsible for improving.

03

Attention & agency

Patterns, agency, and the obligations we owe ourselves.

05 / Philosophy

How We Render Reality: Attention as an Instrument of Love

themarginalian.org

≈ 4 min · Current

88/100

Maria Popova presents Iain McGilchrist’s view of attention as more than a mental spotlight turned toward an already fixed world. Reality as experienced is neither wholly outside the mind nor merely made up by it. Attention is the relation through which something real becomes available, and different forms of attention alter what can be noticed, valued, and taken seriously.

The moral weight sits in that mechanism. Narrow, detached, empathic, sustained, or fragmented attention does not simply record different amounts of the same reality; it trains future perception and changes the world one is able to meet. The open question is how a mind’s manner of attending becomes a way of rendering reality.

06 / Life

On agency

henrikkarlsson.xyz

≈ 20 min · Timeless

86/100

Henrik Karlsson defines agency as autonomy joined to efficacy: knowing what genuinely matters to you, then studying reality closely enough to act on it. Either half can fail. Autonomy without problem-solving drifts into private preference without traction, while efficacy without self-authored aims can become efficient obedience to inherited goals, status markers, or default routes.

The sharper claim is that many obstacles look impossible because the desired function has been fused to a conventional form. People inherit routes, confuse symbols with ends, or avoid experiments that risk social humiliation. Agency begins in the separation: what is actually wanted, what reality permits, and which assumed constraint is only a costume for conformity.

07 / Engineering

Your Agent Isn't Dumb. Your Tools Are.

thetshaped.dev

≈ 7 min · Discovery

91/100

The central claim is that many agent failures blamed on weak prompting are really failures in tool-contract design. The model does not see the source code behind a tool; it sees names, descriptions, schemas, examples, and returned errors. That surface becomes the operating reality, so ambiguity at the boundary forces the model to guess before ordinary code blindly executes the guess.

Reliability, in this account, improves before execution. Narrow tools, typed inputs, bounded outputs, and errors that tell the model what to do next reduce the space for accidental interpretation. The useful tension is that better agents may still fail if their tools ask for judgment where a contract should have supplied structure.