I miss thinking hard

Impact of AI on “Thinking Hard”

• Many agree with the article’s core feeling: LLMs make it too easy to get a “70% solution”, reducing occasions where you sit with one hard problem for days.
• Others say the opposite: they now think more and at a higher level (architecture, requirements, trade‑offs) while offloading boilerplate, debugging drudgery, or API spelunking to AI.
• Several describe a shift from “scientist/mathematician” style deep focus on a single idea to “manager/orchestrator” thinking: more context switching, supervising agents, reviewing code.

Quality, Technical Debt, and the 70% Solution

• A recurring worry: AI encourages “good enough” solutions that hide technical debt and edge‑case ignorance; compounding “AI slop” could later explode.
• Seniors report junior engineers pasting in AI‑generated code they don’t understand, adding unnecessary dependencies and longer PRs that are harder to review.
• Some liken full-agent use to outsourcing low‑quality work: fast prototypes that then require 10× human effort to stabilize.

Cognitive Load, Flow, and Skill Atrophy

• Many report greater mental exhaustion: rapid context switching between reading, prompting, reviewing, and testing is tiring, but doesn’t feel like the same rewarding deep focus.
• Some say reviewing AI code is harder than writing it, and that they lose a strong mental model of the system when they don’t write the code themselves.
• There’s concern about “cognitive atrophy”: offloading search, recall, and design decisions may erode problem‑solving muscles and long‑term system understanding.

Tools, Abstractions, and Craft

• One camp treats AI as just another abstraction layer (like compilers, libraries, or frameworks) that frees humans to tackle more ambitious problems.
• Critics counter that LLMs are not stable abstractions: outputs are nondeterministic, leaky, and must be checked in as code, not as reusable high‑level specs.
• Strong “craft” sentiment appears: coding as hands‑on learning and discovery, analogous to pottery, carpentry, or darkroom photography; AI is seen as skipping the formative part of creation.

Workplace Pressure and Pragmatism

• Some commenters say “just don’t use AI”; others note employers track AI usage, push for AI‑first workflows, or hint at layoffs for “inefficient” non‑users.
• This creates tension between personal desire for deep thinking and organizational incentives for speed and volume.

Proposed Coping Strategies

• Use AI narrowly: boilerplate, search, refactors, test generation, or quick prototypes, while doing core design and key implementations by hand.
• Seek harder domains (systems, performance, crypto, embedded, math, philosophy) or non‑coding hobbies (woodworking, chess, physics, Project Euler) to keep the “Thinker” exercised.