This Week's Feed — sree.build

Currently Reading

Thinking About

Alignment as the New Implementation

When code is cheap and agents are fast, the scarcest resource becomes agreement on what to build. Maggie Appleton's "Zero Alignment" framing posits that every team is shipping faster into more misalignment. The tools we coordinate with (PRs, Slack, Jira) weren't designed for this speed. What does the planning layer actually need to look like?

Department Chair Behavior in the AI Transition

The Daycare vs. Department Chair framing maps almost perfectly onto AI replacement risk. The behaviors that make someone a "Department Chair" — proactive communication, owning a lane, bringing options not just problems — are the same behaviors that make someone useful alongside agents, not replaceable by them. Worth thinking through what this means for how you orient and adjust your career focus over the next few years.

The Virtue of Constraints

Bryan Cantrill's point about LLMs lacking "laziness" (the programmer's drive to simplify) stuck with me. Humans optimize because time is finite — that constraint is what forces elegant abstractions. AI has no such constraint, so it grows systems instead of simplifying them. Is the lesson that we need to artificially re-impose constraints on AI-generated code, or that the human needs to stay in the loop at the abstraction layer specifically?

Daycare or Department Chair: Which Are You?

Codie Sanchez's bluntest piece in a while: two archetypes of employee — ones who require constant hand-holding (Daycare) vs. ones who own their lane and bring problems with plans attached (Department Chair). The AI angle lands hard: daycare work — checking boxes, waiting for direction — is exactly what AI handles first. The future is a small number of high-ownership humans plus a swarm of agents.

ct.contrarianthinking.co

Security · AI

Cybersecurity Looks Like Proof of Work Now

Anthropic's Mythos model completed a full 32-step corporate network attack simulation in 3 of 10 attempts; AISI confirmed it's a step-change over prior models. dbreunig extracts the uncomfortable takeaway: hardening a system now means spending more tokens finding exploits than your attacker does. Security is becoming a token budget war, not a cleverness war.

AI · Coding Theory

The Peril of Laziness Lost

Bryan Cantrill makes the case that Larry Wall's "laziness" — the programmer virtue of optimizing to never repeat yourself — is what LLMs fundamentally lack. LLMs don't feel the cost of future complexity; they'll happily grow a layercake of junk. The human constraint of finite time is what forces elegant abstractions, and outsourcing too much removes exactly that pressure.

AI · Productivity

How AI Made Me a Builder

A designer writes about AI finally closing the gap between imagination and execution — not by replacing craft, but by making the feedback loop immediate. Building in code forced her to learn in context, which made her sharper. The loop: hit a wall → read + ask AI → test → come back with better prompts. Underrated framing of how AI actually teaches.

Also This Week

AI · Business