After AI tools had been woven deep into the workflow for a while, the total workload stayed where it was. Drafting speed went up. The volume of checking and judgement stayed the same. This entry records what that actually feels like, the side effects nobody plans for, the tolerance for imperfection that quietly got raised, and a correction to the whole "human vs AI" comparison model.
# What the Workflow Actually Changed: What AI Saved, What It Did Not
Comparing the workflow before and after, one core fact covers most of it: what AI saves is the mechanical act of typing. Every stage downstream that requires judgement —— logic checks, verifying citations and data, tuning tone and rhythm —— carries exactly the same workload as before.
AI made "zero to something" fast. The distance from "something" to "good" still gets closed by hand. The first draft moved from manual to generated, and that saved time. The checking, the verification, the adjustment afterwards all stayed manual. That is the most direct sensation of working with AI: drafting got faster, checking stayed the same.
# The Core Friction: Unpredictable Output Is the Part That Grates
The most grating part of working with AI sits here: the output is hard to predict. Errors can be corrected. Unpredictability can only be absorbed.
Same prompt, same context, same model version, and two runs can land worlds apart. Sometimes the reply is precise, sharp, better than expected, and the tool feels extraordinary. Sometimes it drifts clean off topic, breaks its own logic, invents things outright, and leaves you wondering what you did wrong.
That uncertainty gives AI a strongly polarised face: optimists focus on the upside surprises; pessimists focus on the risk it carries. Both views grow out of the same single fact —— unpredictability.
For anyone who needs stable output, this is a structural problem. The only workable stance is to treat AI as an assistive system that helps most of the time and always requires verification. That is a property of the structure, and it means keeping an eye on it the whole way through.
# Side Effect One: Capabilities Stop Being Forced, Then Fade
AI lets certain capabilities stop being exercised. Over a long enough stretch, their water level drops.
Taking proper inventory, these dimensions shifted the most: overall logic and the links between sections remain under control; short fragments and clipped phrasing work fine; but producing continuous, precisely worded long sentences on the spot has become noticeably heavy. Reading and organising long foreign-language texts now tends to go straight to AI translation rather than being handled personally. The capacity for waiting and delayed gratification —— patience —— has dropped. Appetite for in-person social contact dropped with it.
The common thread: every one of those daily situations moved from "has to be handled actively" to "can be done in one click." The brain naturally prefers the low-resistance path. That is structural, and willpower is the wrong tool for it.
# Side Effect Two: Tolerance for Imperfection Drops
AI accelerates cleanup and organisation most of the time. That acceleration turns around and pressures the user into backfilling every past imperfection.
Personal bookkeeping is the clearest case. Before AI, the mindset was "log the rough shape, glance at where the money went at month end, a few missing entries are fine." After AI, the internal expectation became "categories must be consistent, past data should be backfilled, charts would be good, a complete database would be better." The tool upgraded the task into a systems engineering project while the pace of daily life stayed where it was. The pain lives in the data-completeness standard that AI quietly raised, rather than in the bookkeeping itself.
# A Triage Frame: Want / Need / Must
More tools → more options → higher cost of choosing. Three layers bring the decision load down:
- Must: skipping it breaks something, breaches an agreement, or leaves someone hanging → handle now
- Need: improves the thing currently in hand, and skipping it breaks nothing → schedule it
- Want: interesting, worth trying, no impact if skipped → whatever time is left
Without this frame, every option that merely looks decent gets filed as mandatory by default.
# One Step Back: The Comparison Baseline Is Wrong
Work with AI long enough and a particular anxiety surfaces: however much a human improves, matching AI on completeness, speed or volume of data stays out of reach.
That is a misplaced baseline — measuring humans on the metrics AI was built for. On speed, recall and data volume, humans lose. The live competitive axis today runs between people who use AI well and people who do not. That gap is the real shape of the competition.
Resistance to replacement concentrates in three things:
- Defining the problem: knowing what actually needs solving
- Deciding on incomplete information: picking a direction while the data is thin, and carrying the outcome
- Building trust: getting real people to want to work with you
# Conclusion: The Ability to Choose Stays at the Centre
AI left one basic fact untouched: the ability to choose remains the core of individual work. What it did was move attention off execution and onto definition and judgement.
A tool becoming ubiquitous leaves the core capability exactly as necessary as before. Smartphones replaced feature phones, and judging whether information is true, managing attention, dealing with people, holding focus long enough to finish something all became more important. AI follows the same logic.
Back to the opening question: did AI make this easier? No. It redistributed which parts are easy and which parts are not. The parts that stayed hard —— judgement, verification, carrying the risk of unpredictable output —— are precisely the core that cannot be outsourced yet.