Unmatched Productivity, A Case For AI at Work

“I strive to be lazy. No really, I love being lazy! If I find any tasks or processes that are repetitive and that can be automated, I’ll find a way to automate them.” These were the exact words I used when applying to my former role at Meta. While my phrasing might not have been ideal (and my interviewer made sure to tell me, after chuckling, that I should phrase that differently), it perfectly captured my intentions for that role. I've always been focused on getting as much done in as little time as possible while ensuring the quality of my work is still the best of what I can do.

It wasn’t really laziness either, but it was about removing even the smallest tasks that annoyed me so I could focus on the things I really enjoyed. Luckily, not long after joining Meta, ChatGPT was released, and it led to a new level of efficiency I couldn’t have dreamed of in my previous roles. Driven by my strong desire to be “lazy” I dove deep into AI, searching for ways to approach streamlined process development. Together with my team (some of the smartest and kindest people I’ve ever met), we used it to transform everything, from generating data transformation scripts, to solving problems that normally took hours to debug, and even quickly generating ideas for responding to difficult or upset clients.

My journey took another leap forward once open source projects like Llama, Qwen, and Stable Diffusion became available. These tools allowed me to automate processes, offer improved consultation services, and optimize business workflows while still being able to work on the things I love. I’ve been incredibly fortunate to explore the intersection of AI and productivity first hand, and my goal here is to share that knowledge, helping others stay competitive in today’s hyper-efficiency focused world.

A New Standard to Efficiency (A data driven approach)

Now, I want to avoid providing any information without really diving into proof of just how transformative AI can be. Inversely, I also want to avoid being mind-numbingly boring with data on how important AI is to your success and efficiency. So I will aim to provide a mix of both as we go through this, taking both anecdotal personal experience and mixing it with studies that are publicly available.

At Meta I was in charge of a support engineering team that handled everything from easy to answer questions, to diving deep into mobile SDKs, REST APIs, or just making things work in highly customizable asynchronous environments where concurrent tasks could overlap each other (race conditions). When AI was introduced to us, there was a bit of apprehension toward adopting it in our day to day work as there was a general fear of it replacing what we do, however, after exploring it deeper we found it best viewed as a strong tool for experts, and an even better tool for onboarding and filling in knowledge gaps. We were able to drastically improve our response times, handle times, and even our deflection rate, giving us more room to further improve the processes we had and focus on skill growth.

This directly relates to a study from the National Bureau of Economic Research where they found the average agent would see a 14% increase in productivity, and novice/entry-level workers saw a 34% improvement (Brynjolfsson et al., 2023). This improvement here was measured in how many issues per hour an agent was able to navigate. What might be most interesting in this study is that it found expert workers did not see much of an improvement. I strongly believe this can be summed up to how AI is utilized and then how it is measured in terms of productivity. For our team, I did find that this was indeed true in terms of total number of completed items, but the difficulty of the tasks my team was able to take on was the major kicker. Rather than getting stuck on deeper SDK or coding issues that we would normally send to our engineering team, we were instead able to tackle more of these difficult issues. This meant our team was directly impacting in a positive way, what our software engineering team focused on, and we were able to start seeing improvement company wide. The use of AI for our team quickly transitioned from being a nice idea, to a necessity without requiring much more than use of a pre-built tool. Unfortunately, it isn’t always that easy for everyone.

The Theory of Overcoming Hurdles Faster with AI

While I can’t speak to the metrics our team might have benefited the development team, I can certainly speak to how our collaboration changed. It was fast paced! We were able to collaborate quickly on issues and talk to a certain level of expertise our engineering team was not generally used to. This came down to a new automation we had in place that we didn’t even consider being an automation.

Before I dive more into that, I want to mention how there’s generally a direct relationship between fast results and the quality of work. If you increase how fast you handle something, it generally means quality suffers, and vice versa. This is really where AI shines!

While AI might not change the problems you encounter, it will change how fast you can find solutions to those problems. Maintaining accuracy while increasing speed is the “holy grail” of any productivity driven workflow. I can say from experience, most of your time spent on a problem is either spent working through trial and error workflows, or spending hours digging on Google, Reddit, Stackoverflow, or any other community driven forum. This is the process that needs the most refinement and up until now, it isn’t something I would have considered to “automate”.

While our team mostly consisted of developers despite being a support team, there was certainly a knowledge gap between our platform use experts, and our platform code experts. The efficiency we gained in that communication and collaboration led to happier developers, faster resolution of internal tasks, and an overall sense of comradery that we didn’t have before. While AI didn’t eliminate our difficult problems, it was an invaluable tool in the process of finding, implementing, and communicating solutions quickly. Now outside my anecdotal and maybe overly positive outlook on AI’s use, it can also be directly seen in a study from MIT where programmers using CoPilot saw an improvement of being 55.8% faster (this translates to a whopping 126% increase in throughput), without a change in the ability to complete the assignment.

This sounds like some fairly amazing metrics, but it isn’t as easy as just providing users with tools. Just like any automation tool put in place, there has to be a process to not only ensure adoption, but to measure its benefit.

Introducing the Double Diamond Design Process with AI

I was first introduced to the Double Diamond Design Process in an AI design course, and I quickly realized it can be applied to far more than just design or programming. It is a powerful approach to solving a wide range of challenges in both personal and business workflows. This process consists of two segments (diamonds), discovering and defining the problem, followed by developing and delivering the solution. While it sounds like it might be complex, I assure you it creates a simplified approach that may come across as just being common sense.

Discover: Understand the problem in detail

In this first phase, you take time to discover all aspects of a problem. This might mean gathering data, looking into existing workflows, or speaking to stakeholders to uncover inefficiencies or bottlenecks. If it is more individually focused, then take the time to really break down some of your small annoyances you have with regularly completed tasks.

Hopefully, by the end of this stage, it should feel like you’ve had a good venting session, and you’ve identified issues you might not have realized were there in the first place. A great goal checklist might include:

• Data collection: workflows, processes, and areas where most of your time is being spent.

• Feedback: If you’re working with teams to do this, then engage with your team to identify pain points they encounter daily, especially repetitive tasks.

• Performance: Get a snapshot of current metrics, this is extremely important since you need to identify what you are starting with to measure success.

The goal from this phase is to identify almost a wishlist of things you would like to be (or might already be) lazy about.

Define: Refining the Problem into Actionable Objectives

Once you have finished your discovery, you need to refine these areas into clear, actionable problems:

• Map out specific pain points such as slow response times, or regularly receiving bad data.

• Set measurable goals such as reducing response times by 20%, or cutting data entry timelines in half.

• Determine the types of problems you encounter. Is there a theme in repetitive tasks, do you have poor data analysis, or do you have low customer satisfaction?

Here you should clearly know what your problems are and hopefully start to see ideas on what can be done to address them.

Develop: Creating AI-Driven Solutions

Now we get to the fun part, which is creating solutions. Telling people to just use AI will not work. I’ve seen this time and time again where organizations purchase great tools, provide a login, and then expect teams to figure it out with zero direction. You must develop full processes around how these tools can be implemented:

• Prototype solutions and make sure the tool in use matches the intended use case. Whether it’s a chatbot to reduce response times, or some great AI analysis tool. There is a vast selection of tools to use, and it will take time to find out the best tool for the job.

• Ensure there is training to use the tools effectively. It often takes some research to know how to “work” with AI tools and prompt them effectively for the solutions you’re looking for.

Deliver: Implement and Iterate

This is where you get to see all your hard work pay off. Rolling out your AI tool isn’t just a one-time thing though, it’s a never-ending process:

• Start with small group testing, especially if costs are involved. Select a small team, or if you are doing this solo, apply it to a subsection of your work.

• Gather feedback and track metrics. Compare it to the metrics you defined before and see if there is measurable improvement.

• Then finally, improve or change direction. Based on the performance of your solution, there might be a better tool to try.

Enter the AI Powered Autodidact

Now that we’ve explored the Double Diamond framework, it hopefully hints that the success of AI in any workflow doesn’t just depend on the tool we use but how well we implement it and the knowledge behind that effort. This brings me to my favorite concept: the AI-powered autodidact, an individual who leverages AI not just for productivity, but for self-learning.

The only way to improve these tools and processes further, is to constantly and consistently be in a growth centered mindset. Freeing up time with automation provides an opportunity to focus on the next challenge, or further growth. This naturally requires breaking down knowledge barriers. This quickly enables anyone not only to build better automations, but also sets us up in such a way to consistently learn on the job!

A great personal example I have is since we had more time to take on tougher issues, my team starting encountering more issues with RegEx (Regular expressions) patterns. Without getting into too much detail, RegEx is something you can use to quickly match data. Even for experienced developers, this can be challenging if you do not regularly work with them. So in an effort to help my team in learning, I created a specialized GPT. This was built on top of ChatGPT and it acted as a grumpy professor that gave you 20 randomized questions on RegEx patterns. If you got an answer wrong, it would chastise you (humorously), and then explain why it was wrong and provide reading on how to improve. While it was fun, it ended up being engaging and a great learning tool for my team.

All of this is to say, AI, when utilized effectively, can be a great tutor and resource for learning skills you might otherwise think are out of reach. It is a tutor that is always available, can be requested to talk in preferred style, will not be annoyed to repeat itself, and is available 24/7 with access to an unprecedented amount of data. If anything through this chapter, I hope you can consider the empowering nature of AI as a learning companion not only for your current role, but also in areas you might not have thought about.

Thriving in the AI Era

To be clear, AI isn’t just a magic button you press and voila, all your problems are solved, but it is setting a new precedent of productivity that almost necessitates you to embrace the change. That does come with a bit of caution or discretion, because although AI is great for automation and productivity, it absolutely needs oversight and refinement. If you just toss a question to ChatGPT or Claude.ai, and start copy/pasting that as responses to clients, you (and your client) are likely to be in for a shock.

I also want to call out that while most of my discussion has been specific to the tech industry, it isn’t exclusive. A study on the productivity in manufacturing from 2010 to 2021 from Beijing Jiaotong University found that every 1% increase in the involvement of AI in business directly improved business resource efficiency by over 14%. So whether you're in tech, manufacturing, marketing, or even fitness, AI is transforming workflows across the board.

These tools are also only going to become more advanced, likely leaving a large gap between those that use AI tools and those that don’t. While I don’t suspect that AI will replace us, it will make us better at what we do, and you have to stay in the loop, continuously focus on learning, self-development, and sharpening your AI skills.

My best recommendation going into the AI era is to scan AI news and trends weekly, but please don’t let it overwhelm you. Become familiar with the tools that make sense for you, or, even better, find areas that are exciting, and are something you would enjoy to learn more about! Try out AI tools, even if it’s just for fun, to get a feel for what they can do. There are plenty of open source tools out there and even free lessons to learn how to build your own free version of popular tools. Dive into AI communities online and see how others are using it. You may find cool solutions you can build into your own workflows.

Lastly, I want to stress that AI isn’t just about solving problems faster, it is a tool to empower you to focus more on what you love and excel at. So take some time to build great solutions and then enjoy some of the hard earned opportunities to be “lazy” and explore the opportunities for growth you’ve put in place.https://arxiv.org/abs/2302.06590