AI Shortened the Schedule—and Overloaded Me

On this page

1. A shortened schedule can be a trap
2. Decision bandwidth overload
3. Deciding quickly does not necessarily reduce the backlog
4. When AI is fast, I slow down first

Recently, doing projects with AI, I ran into a slightly awkward situation: proposals came out fast, but I no longer had the energy to keep making the calls.

I had several things going at once: the personal site needed its overall structure and visual style reworked; I was building a health and chronic-disease management system for myself; and I was organizing years of past-experience material into structured documents to build a personal engineering knowledge base.

AI genuinely took on a lot of the work. Hand over the requirements for the personal site and the health system, and frameworks and proposals came back quickly; years of accumulated historical material got an initial inventory as requested. Looking only at that progress, everything was moving along smoothly.

But when it reached me, every project had a pile of things to settle. Frameworks and proposals, and the many details inside them, were waiting for my judgment; and once the knowledge base’s basic information had been inventoried, how to extract and organize the content afterward raised many more cases to decide.

Add daily content creation and research, and I suddenly felt I could not carry it. The fatigue was obvious in both body and mind, and it was hard to concentrate enough to make careful judgments.

The effort AI saved was real, and the extra exhaustion was real too. Two things that seem contradictory both happened, exactly as they were.

A shortened schedule can be a trap

Looking back, the problem lay in the judgments that arrived all at once.

A project goes from idea to completion through many choices. Is the framework suitable, is the proposal worth refining, how do you confirm the details, can the result be accepted. Execution used to take time, so these judgments were spread across the whole schedule along with progress.

Now AI has sped up most of the execution steps, so a project quickly reaches the next point that needs human judgment. Things that could once wait a while now crowd together and arrive at once.

The total number of decisions a project requires may not have increased. But the same 50 judgments and decisions spread relatively evenly across a 30-day schedule, versus concentrated into 3 days or even 1 day of continuous handling, place completely different demands and pressure on a person. The former leaves you room to digest, compare, and think again; the latter can have you make one call and immediately move to the next.

So being able to run several projects at once in the past does not mean you can now, much less handle more. You may have coped before precisely because the decision points of a few projects happened to be staggered. Once AI speeds up execution, not only does a single project’s decision demand become concentrated, but parallel projects are more likely to pile up together.

Decision bandwidth overload

I call this “decision bandwidth overload.” The amount of judgment, trade-off, and review a person can effectively handle in a period of time is their decision bandwidth. When the demand arriving in a short span exceeds that capacity, judgment starts to distort.

It is different from judgment ability. Being able to think something through given enough time does not mean you can handle a batch of such things arriving in succession. Decision bandwidth is also hard to convert into how many calls you can make per day: deciding how to reword one sentence and deciding where a project should head require investment that is far apart.

What makes it worse is that when several projects press in together, the energy consumed can far exceed a simple sum.

Take the things I had in hand: the website, the health management system, and the personal engineering knowledge base could each be pushed forward by AI. But switching among them meant constantly switching lines of thought. The considerations for judging a website proposal cannot be used directly to judge how to organize personal experience. Every time I changed projects, I had to reconnect that project’s goals, background, and earlier trade-offs.

If the previous thing is not yet thought through and the next is already waiting, a person easily ends up hanging between two. Attention gets pulled back and forth, and the thing in hand becomes harder to think through. By this point what has increased is not just the number of to-dos; even the state in which you handle each one is affected.

In my view, decision pressure in this situation can grow exponentially. If three projects each need five units of energy, pressing in together in a short span may no longer be 5 + 5 + 5 = 15, but more like 5 × 5 × 5 = 125.

It is easy to turn on AI and advance several projects at once, but that does not produce several selves capable of careful thought at the same time. AI makes projects run faster, and it can also make the person in charge hit their own decision bandwidth limit sooner — decision bandwidth overload.

Deciding quickly does not necessarily reduce the backlog

Faced with all that waiting, the easiest idea is to just choose quickly and let AI keep going.

But making decisions continuously under high pressure can distort the act of deciding. Differences that were worth comparing get skipped as too much trouble; questions you would have pressed on get waved off as good enough. It looks like everything was handled, but in fact the purpose of deciding may have slid from thinking the matter through to just getting this thing off your desk.

Think about whether each of us has had a run of repeated mistakes from making decisions quickly and continuously under pressure. That is exactly why I did not want to keep making calls while already exhausted.

A decision not carefully considered can send AI down a direction that may not be right, and with its high execution and drive, quickly produce a large batch of results. By the time you look back and find the direction unsatisfactory, what needs adjusting is no longer just the original proposal. Of the things already made, which to keep, which to change, which to abandon — piled together, these add new items to fix and new decisions to make.

The more results there are, the more effort sorting and adjusting takes, and the harder it is to bring yourself to tear them down. The backlog you wanted to clear quickly may come back with more problems, consuming energy you already lacked. These sunk costs easily become interference in your next decision.

This is the chain reaction I worry about: judgment demand concentrates, the person is worn out, hasty decisions get executed quickly, and the correction that follows pushes the judgment burden even higher.

If I keep opening projects by past habit and keep making calls at the pace of AI’s output, I worry this decision bandwidth overload becomes the norm. Execution speeds up sharply while the quality of deciding may fall sharply.

Of course, AI can also help compare proposals, filter information, and check results, so the human share of review, judgment, and decision investment can be reduced as appropriate. When a task is mature enough and the requirements clear enough, many things can be handed to AI to judge and handle on its own.

But in working with AI there are always some key judgments you must hold yourself. Even with execution fully delegated, whether the result fits your goals and whether a deviation is worth redoing still come down to your own judgment and your own bearing of it.

When AI is fast, I slow down first

This time I slowed down first. I simply reordered things, cut down on parallelism, and concentrated energy on finishing one at a time.

The aim is clear: each time one is finished, one hanging to-do is gone, and so is some of the mental pressure that interferes with other judgments.

Cutting parallelism has a cost. Some projects move later, not every piece of work AI could do is handed over right away, and sometimes quota and gifted token balance go to waste. I am willing to accept these costs first, to leave myself room for careful judgment.

Even if AI returns a result quickly, that does not mean I must decide the next step immediately. When to push on and when to stop and think is a rhythm I set myself. Otherwise, the execution time saved easily turns into pressure to make the call right away.

This does not mean every project must line up single file from now on. For things that can be fully delegated and rarely need intervention, parallelism is clearly valuable. What I want to rein in is this kind of project, the ones that keep requiring my own trade-offs.

Once AI speeds up a project’s execution and delivery, whether the person becomes the new bottleneck has to be taken seriously.

How fast AI can build something is one account; whether I can keep making reliable judgments is another account that also has to be reckoned.