AI LLM Software Engineering Management Communication

How AI Changed Barely Anything At All

In The Cost of Working Without AI, I wrote that losing access to AI would probably cost me about 70% of my current output. I stand by that. These tools have changed how much I can get done enormously.

I've also felt for a long time that most advice about working with AI is familiar advice about working with people. I made much the same argument in All Your Coworkers Are Probabilistic Too. This hasn't suddenly occurred to me because a model got better at writing code.

AI is turning more of us into managers. And I suspect a substantial part of the excitement about learning to work with it is people discovering delegation for the first time.

Congratulations, You're Managing

Spend some time on LinkedIn and you can watch the discoveries arrive. Give it context! Define what a good result looks like! Break the work into smaller steps! Review what comes back! Write down instructions you keep repeating!

For this article, I asked an AI agent to do a quick web search for examples. These are a few posts it found, picked because they illustrate the point. I haven't studied the authors' working habits or management experience. The enthusiasm is easy enough to find, though.

A LinkedIn screenshot enthusiastically announcing a breakthrough with AI coworkers.
From a post about delegating work to AI coworkers.

The practical discoveries are rather familiar too. Explain the result you want and the constraints that matter:

A LinkedIn screenshot describing how to specify the result Claude should deliver.
From a post about improving an AI workflow.

Document work you keep repeating:

A LinkedIn screenshot describing how documenting recurring work makes it reusable.
From a post about reusable processes.

Useful advice. Also advice I would have happily given a new manager ten years ago. Clear instructions didn't become a good idea when we started putting them into a chat window.

Many of us never had much opportunity to manage. The overwhelming need was to get things implemented. You could have good ideas, understand the problem, and know how to organize the work, but still spend most of your day carrying out every step yourself. You didn't necessarily have a team, or the chance to find out whether you were any good at directing one.

Now more of us do. That includes people who found the interpersonal complications of delegation frustrating. With agents, I can give feedback and see another attempt while the thought is still fresh. In my experience, there's often less misunderstanding. I can discard work without worrying about its author's feelings. The pace is different, and much of the human element is gone.

I was already used to managing people. My job has been becoming more of that for a while, including work I used to implement myself. I can see why somebody encountering this for the first time would be excited. I am excited too. I just find the advice rather less revolutionary.

We Always Needed the Documentation

People bring context that an agent doesn't have. A colleague knows what your team means by the usual release process. They remember the previous project, the conversation that never made it into the ticket, and the reason for an apparently arbitrary constraint. They can fill in quite a lot when you're imprecise.

AI can do that for plenty of general knowledge too. But the accumulated tribal knowledge of your organization hasn't necessarily been written down anywhere it can find. People have spent years compensating for those gaps. An agent exposes how much the organization relied on them doing so.

I spent four years at a company where onboarding was painful: little was documented, departments had competing ideas, and the productivity and codebase showed it. My own onboarding, like that of many others, took weeks and we all still had a surprising amount of tribal knowledge to absorb after months. After a year, I knew enough to write comprehensive onboarding and development guidelines. A developer I'd hired arrived while I was on vacation and contributed code on their second day. And he didn't have any questions about the process when I came back from vacation. That made me very happy.

So we document the process, explain the constraints, and record why we made a decision. We make specifications clearer and put checks around the work. Onboarding, communication, documentation, development guardrails, verification. We already knew these things mattered. We were just remarkably willing to get by without them.

Now there is an immediate cost to leaving that context in somebody's head: the work we want to delegate goes badly. Writing it down and making it reusable can apparently save a lot of time:

A LinkedIn screenshot claiming substantial time savings from an established AI workflow.
From a post about reusable context, workflows, and scheduled work.

That knowledge was worth making accessible when it was a new employee who needed it too.

Now the Bottleneck Is Us

As more implementation becomes cheap to delegate, understanding the problem, developing ideas, making decisions, and communicating intent account for more of what determines the result. Those were always important. They will increasingly determine how much useful work we can get done.

I can have several things implemented while I think about something else. I still have to decide whether they were worth implementing, understand what came back, and judge whether it's good enough. Having a bad idea implemented faster doesn't improve the idea. It does let me discover the problem sooner, provided I actually look.

This is familiar territory for anybody who has managed work through other people. I'm hardly the only one having that reaction:

A LinkedIn screenshot recognizing familiar management practices in advice about working with AI.
From a post making the same management comparison.

What has changed enormously is our capacity to put the old advice into practice, and how many of us now find ourselves directing work. Implementation consumed so much of our time that thinking, documenting, checking, and refining often had to fit around it. Now I can delegate more of the doing and spend more attention on those things.

We have vastly more capacity, and more of us are managers. Apparently, we still need to understand what we want, explain ourselves, and check the work.

Good thing we've been practicing.

Comments

Comment on this article
Names are not protected without a secret.
Identifies your name and lets you delete comments.
or drop anywhere in this form JPEG, PNG or static WebP, up to 1 MB.
Loading security question...
No comments yet. Be the first to comment!