SAM I am

My first exposure to computing came in 1961 as a high-school intern at IBM’s Watson Scientific Computing Laboratory in New York City. On weekends, they let me use an IBM 650—a room-sized monster that required a 20-ton air-conditioning system and contained thousands of vacuum tubes, each apparently competing to see which one could fail first and thereby prove the value of preventive replacement.

A few years later, as an undergraduate at Dartmouth College, I co-wrote the “Dartmouth Time Sharing System,” one of the pioneering operating systems that helped make interactive computing practical. Later, I wrote software and managed software projects for a Fortune 500 company, founded a company that built computers based on Z80 processors and 8-inch floppy disks, and founded a software company with clients like GE, Phillips, and Visa. Over the course of my career, I actively participated in the computer revolution, the PC revolution, the internet revolution, and the smartphone revolution. Today, it’s difficult to imagine how we survived without any of those things.

Yet none of them has excited me quite as much as what I’m seeing today with artificial intelligence (AI). That’s a bold statement, and I don’t make it lightly.

Most of the airplanes we fly today were built decades before the internet existed. Many were built before personal computers existed. Yet I’ve spent much of my aviation career arguing that while our airplanes may be old, there’s no reason our maintenance practices should be. We’ve embraced engine monitors, digital borescopes, electronic maintenance records, and data-driven maintenance. Artificial intelligence may simply be the next step in that progression. Or perhaps a giant leap.

Meet SAM

This story begins with my colleague Eric Svelmoe, an A&P/IA.

Before joining Savvy and eventually becoming part of our management team, Svelmoe spent decades wrenching on airplanes, building engines, and running a large Part 145 repair station. Later, as one of Savvy’s most experienced account managers, he spent years helping aircraft owners navigate complex maintenance decisions.

Recently, I asked Svelmoe to document everything an account manager actually does. I expected a modest list. What I got instead was a lengthy document enumerating dozens of different tasks that account managers perform. As I read through Svelmoe’s write-up, a realization began to dawn on me. Certainly, some account manager responsibilities required deep technical expertise and decades of experience. But many functions on Svelmoe’s list involved monitoring, research, workflow management, prioritization, quality control, and follow-up. It seemed to me that such functions might be candidates for augmentation by AI. That’s when the idea for SAM was born.

SAM stands for Shadow Account Manager. I imagined an AI assistant quietly looking over the shoulder of every human account manager (HAM) on Savvy’s team, watching every active maintenance ticket and constantly asking itself a simple question: “How can I help?” Not replace the HAM. Help the HAM. Turbocharge his productivity. The goal was never to create an artificial A&P, but rather to create an extraordinarily capable mechanic’s assistant that could help our human experts become more effective.

Second set of eyes

One of the first capabilities I envisioned for SAM involved something surprisingly mundane: making sure the HAM doesn’t drop a ball.

Anyone who handles large volumes of email or support tickets has experienced it. A client sends a lengthy message containing a bunch of questions. Your response answers most of them but inadvertently overlooks one. It happens to all of us. The busier we become, the more likely we occasionally miss something.

So, the first skill I thought we should teach SAM was what we have come to call Completeness Auditing. Whenever a HAM posts a response to a client’s or shop’s query, SAM reviews the recent conversation and determines whether any significant question appears to have gone unanswered. If everything was addressed, SAM remains silent. If something appears to have slipped through the cracks, SAM points it out to the HAM and tries to suggest a possible answer. SAM is essentially a checklist with a brain.

A little TLC

One of my favorite SAM skills isn’t technical at all. It’s emotional.

It turns out that modern AI systems can be remarkably good at detecting sentiment in written communication. So, another of SAM’s skills is what we call Sentiment Analysis. If SAM notices signs of frustration, confusion, anxiety, or anger in client communications, it doesn’t say anything to the client, but it privately nudges the HAM: “This client appears frustrated.” “Additional explanation may be helpful.” “Some extra TLC may be appropriate.”

Most people think of AI primarily as a research tool. Yet one of its most valuable contributions may be helping us communicate more effectively with other human beings. I find that fascinating.

Research assistant

Other SAM skills are more obvious. Suppose a client asks whether a particular airworthiness directive applies to his aircraft. SAM can research the issue in a blink of an eye and draft a suggested response, saving the HAM valuable time.

Suppose a shop asks where to locate a hard-to-find replacement part. SAM can search tens of thousands of archived Savvy tickets and scour the internet, then suggest a list of possible sources before the HAM even reads the shop’s request.

Another SAM skill is timeliness monitoring. If a ticket appears to be languishing, SAM can bring it to someone’s attention before the client begins wondering whether anybody is listening. SAM can recognize the urgency and send a text to the HAM’s phone reminding him to follow up ASAP.

One particularly interesting capability allows a HAM to explicitly ask SAM for help by making a private ticket post beginning with the words “Hey SAM” followed by a plain-English request. SAM then responds to the prompt. This permits the HAM to ask SAM to perform research, summarize information, review a situation, answer a question, or proofread a proposed response.

Knowledge workers often spend more effort finding information than applying it. If SAM can reduce that burden, our human experts can devote more attention to the things that require human judgment.

Awesome

Once I had the basic concept for SAM, I described my vision, combined it with the HAM job description Svelmoe had prepared, and asked an AI system to generate a formal requirements document to pass to Savvy’s software engineering team. A few minutes later, it produced a detailed document describing 17 different SAM skills divided into three implementation phases.

I sent the requirements to Thanos Diacakis, Savvy’s longtime chief technology officer. Normally, when I propose a new software project to Diacakis, I expect a response along the lines of: “Which existing project would you like me to delay so we can work on this one?” This time, his response consisted of a single word: “Awesome!”

Watershed moment

Diacakis and other members of Savvy’s leadership team reviewed the document and suggested a bunch of changes.

At that point I expected the project to disappear into the software engineering pipeline for a few months before anything useful emerged. I was wrong. About a week later, Diacakis scheduled a Zoom call and demonstrated an early working version of SAM performing Completeness Auditing on actual live maintenance tickets. I was astonished. I told him so. Then Diacakis said something that stopped me cold.

“With the help of AI coding assistants, we’ve actually reached the point that we’re able to implement software faster than you can generate requirements.”

I immediately recognized that statement as a watershed moment. For most of my life, software development was limited by the speed at which programmers could write code. Today, the bottleneck is understanding what we want the software to do.

Computers speaking English

One of the most surprising things about SAM is how we teach it. When I started programming computers, we communicated using punched cards and programming languages. Today, many of the refinements we’re making to SAM involve nothing more exotic than editing English prose. Each SAM skill is governed by a detailed English-language rule set describing the behavior we want. If we don’t like some aspect of SAM’s behavior, we revise the rules. In English. On a web page.

SAM’s alpha test

As I write this, SAM is in alpha testing. We’re adding skills one by one. At present, SAM’s ticket posts are visible only to a small internal leadership group. We’re evaluating, critiquing, and refining SAM’s behavior before exposing it to HAMs. We don’t want to unleash any SAM skill on our HAMs until we’re convinced it’s ready for prime time.

The plan is to deploy skills one at a time as they mature. Some may prove useful right away. Others may require considerable tuning. It’s too early to tell exactly how useful SAM will become. That’s OK. Experimentation is part of the process.

Old airplanes, new tools

The shortage of experienced aviation maintenance professionals is real. Aircraft owners encounter it everyday. Maybe AI can help.

AI doesn’t magically create expertise. It doesn’t replace decades of experience. It doesn’t substitute for human judgment or wisdom. But what if it helps one experienced professional accomplish significantly more? What if it prevents important details from being overlooked? What if it reduces administrative workload and allows experts to focus on the work that only experts can do? That’s what excites me.

Why I’m excited

There’s another aspect of AI that I find particularly intriguing. Throughout my business career, one of the persistent challenges has been “loss of corporate memory.” An organization spends decades accumulating knowledge, experience, judgment, and wisdom through hard-earned lessons. Then key people retire, move on, or pass away. Priceless institutional knowledge disappears with them.

Aviation maintenance is especially vulnerable to this problem because so much expertise resides in the minds of experienced mechanics, inspectors, and maintenance professionals. For the first time, AI offers the possibility of preserving at least some of that corporate memory. I’d like to think that someday, long after I’m gone, a mechanic or aircraft owner might be able to ask an AI system, “What would Mike do?” and receive an answer that reflects not just what I knew, but how I thought about the problem. We’re working on just such an AI model now. But that’s another story.

I began this journey feeding programs on punched cards to an IBM 650. Today we’re building systems that can read maintenance records, understand conversations, detect client emotions, conduct research, and help experts make better decisions.

I’ve participated in almost every major technology revolution of the past seven decades. Yet none of them has captured my imagination quite as much as this one. I’m excited, and I just can’t hide it. Welcome to the age of AI.

[email protected]

savvyaviation.com