Is This the Best You Can Do?

“Don’t keep a dog and bark yourself” is pretty good advice. I first read it in a leadership book a few years ago, but I can’t remember which one.

It is, of course, about delegation. Once you hire people to do something, you, as a leader, should stop doing that thing yourself and let them take care of it. The idea is simple and makes perfect sense, but it is harder to put into practice than it sounds.

We all like control, and we all think that if we want something done right, we need to do it ourselves. Often, that assessment is correct. We tend to know our own strengths (Dunning-Kruger effect excepted), and other people may not do things exactly as carefully or as thoughtfully as we would. The problem is that this does not scale. There is only so much one person can do, so eventually you have to trust other people and help them succeed.

A big revelation for me came when I started asking people how we should do something instead of telling them. Even junior people sometimes came up with creative, innovative solutions that I had not considered. Sometimes I asked a very open question. Sometimes I added more constraints and explained the goal we were aiming for. In some cases, I had more background and context about the problem or the system. In others, I only knew the business outcome I wanted.

People are smart. They can come up with great solutions without me having to bark.

Now let me tell you a story, and I promise it will all come together at the end: an anecdote involving Henry Kissinger and his aide, Winston Lord, who later became the U.S. ambassador to China. The story, which Kissinger himself has recounted, illustrates his demanding leadership style and his emphasis on excellence.

As the story goes, Lord submitted a draft report, or in some versions a speech, to Kissinger. Kissinger returned it with a question: “Is this the best you can do?” Lord revised it and resubmitted it, only for Kissinger to ask the same question again. This happened several times — up to eight or nine drafts in some tellings — each time with Lord saying it was his best effort, only to revise it further after reflecting on it. Finally, on the last submission, when Kissinger asked once more, “Is this the best you can do?”, Lord firmly said yes and insisted he had put everything into it. Kissinger then said, “Good, now I’ll read it.” (troll face)

Alright. Plot twist. This post is not really about people and leadership. It is about coding and AI.

If you have read the excellent book “Vibe coding”, you probably know the technique of asking for small things, one at a time. Your AI assistant figures out the implementation based on your instructions.

An example prompt might be: “Add a new tab showing capacity usage per cluster.” This is great, and it works. I use it all the time. But here is the thing: the SOTA models, at the time of writing this post — Opus 4.6, GPT 5.4, MinMax 2.5, and Kim 2.5 — are really smart and can reason very well.

So recently, I have been experimenting with asking for ideas instead of telling Claude, my drug of choice, exactly what to do. I am finding that a prompt like “What is the best way to show capacity usage per cluster?” will, more often than not, produce solutions that are more creative, cleaner, and better than my original implementation idea.

And the same rules I use with people apply here, too. My prompt can be more detailed, for example, by adding constraints, describing the outcome I want, or explaining the audience I am targeting. For example: “What is the best way to show capacity usage per cluster? The main audience will be non-technical business people trying to decide where to allocate new customers, but about 20% of users will be engineers working on capacity planning.”

Now for the second part, which I find kind of funny. You let your AI come up with a solution and implementation, write the code, the tests, and everything else. Then, when it happily announces that it is done, do not even bother looking at the code yet. Just ask: “Is this the best you can do?” Not once have I seen it answer, “Yes, this is the best I can do.” It always improved something on the second pass.

What I’m learning is that the old leadership advice still applies, just to a very different kind of worker. If you hired the model, don’t bark the implementation yourself. Give it the goal, the constraints, and the context; ask for ideas before instructions; and when it says it is done, do your best Kissinger impression and ask, “Is this the best you can do?” The real leverage is not in using AI as a faster typist, but as a capable collaborator that often sees options you missed. And just like with people, the magic starts when you stop micromanaging and start demanding better thinking.