Loop engineering is new, exciting, and boosts agentic AI and conventional generative AI.
getty
In today’s column, I examine a newly emerging AI trend known as loop engineering. The idea is that you tell the AI to get something done by instructing the AI to perform a loop that iterates until the AI reaches a specified condition or final state. It is on your shoulders to suitably engineer the loop. If you do a lousy job of specifying the looping aspects, the AI might run afoul or inadvertently chew up pricey computing cycles.
This looping approach contrasts with the typical method of simply interacting with AI on a step-at-a-time basis. Usually, when you converse with AI, you ask the AI a question and get a single answer. You and the AI take turns. One by one, you get the AI to do something or get some action done for you. Loop engineering emphasizes that sometimes it would be better to instruct the AI to run a continuous loop and let you be somewhat out of the loop as it does so.
This rapidly evolving AI trend entailing looping is primarily focused on the use of AI agents. Nowadays, you are supposed to make use of loop engineering when invoking agentic AI. That being said, there is nothing wrong with applying loop engineering to conventional generative AI too. I will show you how. The skills of loop engineering ought to be utilized by anyone interested in doing suitable prompting in AI and being mindful of best practices for prompt engineering.
I have been analyzing and showcasing prompt engineering techniques for quite a while. If you’d like to see what has been covered so far, see my detailed description of over eighty useful prompt engineering techniques and methods at the link here. Seasoned prompt engineers realize that learning a wide array of researched and proven prompting techniques is the best way to get the most out of generative AI and large language models (LLMs).
A vital consideration in prompt engineering entails the wording of prompts.
Capable prompt engineers realize that you must word your prompts mindfully to ensure that the LLM gets the drift of what you are asking the AI to do. Sometimes, just an added word or two can radically change what the AI interprets your question or instruction to consist of. Generative AI can be hypersensitive to what you say in your prompts. It is often a touch-and-go proposition.
Casual users sometimes catch onto this prompt-writing consideration after a considerable amount of muddling around, involving exasperating trial and error. Many users don’t ever become especially proficient in writing prompts. They just enter whatever comes into their minds. That’s probably okay if you are a casual user and only infrequently use AI. Not so for serious prompt engineers.
Shifting gears, let’s consider what happens when using AI agents. I will first discuss what agentic AI consists of. After providing that foundation, I will launch into the new focus on loop engineering.
AI agents are the hottest new realm of AI. To comprehend what agentic AI is, let’s start by considering conventional AI.
Imagine that you are using conventional generative AI to plan a vacation trip. You would customarily log into your generative AI account, such as making use of ChatGPT, GPT-5, Claude, Gemini, Llama, Grok, CoPilot, etc. The planning of your trip would be easy due to the natural language fluency of generative AI. All you need to do is describe where you want to go, and then seamlessly engage in a focused dialogue about the pluses and minuses of places to stay and the transportation options available.
When it comes to booking your trip, the odds are that you would have to exit generative AI and start accessing the websites of the hotels, amusement parks, airlines, and other locales to buy your tickets. Relatively few of the major generative AIs available today will take that next step on your behalf. It is up to you to perform those nitty-gritty tasks.
This is where agents and agentic AI come into play.
In earlier days, you would undoubtedly phone a travel agent to make your bookings. Though there are still human travel agents, another avenue would be to use an AI-based agent that is based on generative AI. The AI has the interactivity that you expect with generative AI. It has also been preloaded with a series of routines or sets of tasks that underpin the efforts of a travel agent. Using everyday natural language, you interact with the agentic AI, which works with you on your planning and can proceed to deal with the booking of your travel plans.
Suppose you tell an agentic AI to book you a hotel for your upcoming vacation. The AI agent says it will do so, and immediately scans various online websites, trying to find a hotel that meets your needs. After having found one, the AI agent tells you what it found and then books the hotel stay. This is all perfectly fine. You have reserved a hotel stay without having lifted a finger to do so. Nice.
I’m sure that you’ve had circumstances in life where you kept watching the marketplace and later realized that a better deal on a hotel had become available. You would then cancel your first hotel reservation and book a new one. Perhaps the better deal gets you a great discounted price. Aha, it pays off to be diligent and always keep your eyes and ears open.
It sure would be handy if AI could do the same on your behalf.
Here’s what you could do. You tell the hotel-seeking AI agent to find you a hotel and book it, but that this isn’t the end of the story. You want the AI agent to keep looking. You might tell the AI agent that for the next few days, it is to scan the hotel websites and be on the lookout for a better deal. If the AI agent finds a better deal, it is to cancel the earlier reservation and book a new one.
Voila, you have defined a loop for the AI agent. The AI agent will be working behind-the-scenes and looping to find you a better hotel package. No need for you to constantly interact with the AI agent. By setting up a loop, it is one of those fire-and-forget launches of a missile. You can rest easy knowing that the AI is tirelessly searching to get you a better deal.
Welcome to the wonderful world of loop engineering.
Have you ever seen a dog that incessantly chased its own tail and didn’t seem to be getting anywhere in doing so?
A big issue with AI and loop engineering is that the AI can go awry if you’ve done a middling job of specifying the looping aspects. I’ll use the example of the AI agent that’s looking to find you a better hotel stay. If you didn’t say how often it should loop, the AI might do so every nanosecond and keep doing this for days upon days. The likely charges for consuming the vast amount of computing server time are going to readily exceed any discount you might get on a hotel stay. Oops, you didn’t do a proper job of telling AI what to do.
Another difficulty with your loop is that you didn’t tell the AI to check with you before canceling your existing reservation and booking the new one. This could be calamitous. Imagine that the AI agent informs you that after looping for a few days, it luckily found you a seemingly better-priced hotel. Your old one is canceled, and the new one is booked. Success is at hand. But when you look at the specifics of the hotel, you see that though it is better priced, it is a hotel with a seedy reputation. Had you been consulted by the AI, you would have told the AI not to book the hotel. Meanwhile, your prior hotel is no longer available since AI canceled your reservation. Sad face.
All told, loop engineering is referred to as a form of engineering because you aren’t supposed to just willy-nilly establish a loop. You are to carefully consider the particulars for the loop and appropriately engineer them. AI is going to pretty much abide by what you have specified (I’ll say more about this momentarily). Any loopholes or gotchas in your stipulations are going to be a mess. The onus is on you to set up a loop that makes sense, covers all the bases, and reduces the likelihood of the AI going afoul.
Loop engineering is a newly emerging aspect. There isn’t yet any standardized way of doing loop engineering. Indeed, if you talk with five different AI specialists, they probably each have a different definition of what loop engineering even consists of.
I’ve come up with my own strawman definition:
Anyone doing loop engineering should consider instilling these five precepts:
The aim is to plan your loop, compose it mindfully, ensure that it will eventually stop, and decide whether you want to be in-the-loop or remain outside the loop. The other important facet is that you should make sure to test the loop before you permit AI to go hog-wild. You can tell the AI to try out the loop for a few quick iterations. Inspect what transpires. Adjust the loop as needed.
Before I show you an example of a loop, let’s see what a non-loop looks like. We will use the circumstance of booking a hotel stay.
Here we go.
You can see that I am undergoing a typical turn-by-turn effort with AI. I asked to find hotels in Boston for this weekend. The AI replied that there are a lot of choices and perhaps I might want to narrow down based on price and location.
I will continue the dialogue.
I went ahead and told the AI to book the hotel, but I might want to later cancel the reservation. The next day, a buddy told me that there was a different hotel that is having a discounted special for this weekend. I logged back into AI.
Everything worked out. Of course, I had to log back into the AI and tell it that a different hotel had a better deal. That seems like a pain in the neck. The AI should be working on my behalf to find a better hotel.
Suppose I was starting over on my effort to book a hotel. I will give AI a loop to be undertaken. The loop should include the various parameters and properties that were mentioned in the five precepts.
I gave the AI various details about the loop.
One aspect is that I have deliberately taken myself out of the decision portion of the loop. I have told the AI it can proceed to book and cancel, and that I only need to be told about this. If you were worried that the AI might go astray, it would be easy to tell it to first seek permission before taking any such action.
The AI went through some test loops with me. Other refinements came up. It might take you a few iterations to get things straightened out. Doing so is worthwhile because you don’t want the AI to go ape and cause trouble.
During loop engineering, your mindset is usually focused on workflow. What are the series of tasks that the AI is to perform when looping? The traditional use of generative AI entails being prompt-centric rather than workflow-centric.
Please realize that loops are beneficial but also can be dicey. What if the AI hallucinates during a loop? Suppose the AI doesn’t adhere to your loop instructions? Maybe you were vague, and the loop stipulations were interpreted differently than you intended. Loops are powerful. Good loops require attention to detail and aptitude.
Loops are harder to adopt in conventional generative AI than they are in agentic AI. Well, at least right now. The odds are that this will change as conventional AI becomes more advanced. The ease of writing one-shot prompts is that you normally can’t go off the cliff with just one bad prompt. With a loop, you are potentially starting a flywheel that might cycle endlessly and spin out of orbit.
The famous American naturalist Thornton Burgess made this pointed remark: “Who runs in circles never gets far.” With AI, the reality is that a circular loop can be extremely useful, such as my example of booking a hotel stay, but it won’t get very far without the right kind of setup. Make sure the AI is fruitfully running in a circle, and you’ll be okay.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。