Transcript
John Gesimondo: Welcome again, everyone, to my talk called Achieving Sustainable Mental Peace at Work Using GenAI. My name is John Gesimondo. I’m a senior software engineer at Netflix. I have ADHD. I might have autism. The topic of sustainable mental peace has been my outside of work but also affects work passion for at least 10 years now. In the last six months, since figuring out that I’m going to be doing this talk, I’ve been using GenAI for everything in my life so that you don’t have to deal with that. I’m going to give you the best stuff from all that exploration.
First, I need to define some terms. Sustainable mental peace, of course. You can approximate this as a steady, reliable presence of calm. You can undefine this as mental chaos, distraction, overwhelm, or really any state that’s unsustainable, like burnout or even mania. I will also talk about neurodivergence as a theme in this presentation. Let’s approximate that to be ADHD, autism, dyslexia. It depends who you ask what else is included there. The opposite of this would be called neurotypical, which most people are. Unless you’re neurodivergent, you are neurotypical. Or as you might think of yourself, normal. That also depends who you ask.
Sustainable Mental Peace Framework
First, I have to give everyone a framework because people like frameworks. This is my sustainable mental peace diagram. This diagram is one way from all of my experimentation that I feel that you can use generative AI to increase your sustainable mental peace at work. If you can relate to this diagram, you’re in the right place. If you’re a high achiever, or anxious, or neurodivergent, or any of those types of things, you’re also definitely in the right place. Even if you think you’re in the wrong place, I promise it’s the right place.
These five areas that would be recovering quickly from emotions, overcoming points of being stuck, enhancing planning and communication, adding time in a flow state, and divergent thinking are what I’m proposing, as I said, as a reasonable formula for how to use GenAI to increase your mental peace. What we’re going to do in this talk for structure is go through each component of the diagram and explain more about how this area will impact peace. Including how that affects different types of brains, and then see how GenAI can help in that area. For takeaways, you can go back to work with some new perspective on mental health, some recipes of how to use GenAI, one plan to achieve mental peace, some prompt examples and principles, and if nothing else, a better understanding of some of your co-workers’ brains.
1. Emotional Recovery
First, we have emotional recovery. This is not to say that all emotions are bad. I’m talking about the really intense ones, such as you getting triggered at work. You want to recover from those as fast as possible. Let me give you some background about how emotional regulation works for most people, and then how that might be different for neurodivergent folks. This is going to repeat in a second. What this is is kind of a battery meter, and you can see when something goes wrong, the difference. A neurotypical person might fall to 50% or so and recover quickly.
From the same issue, a neurodivergent person might tank their energy level and then take longer to respond. This is called emotional regulation, and it is a marker of, for instance, ADHD. If that doesn’t make enough sense, here’s some emoji. Something happens, you deal with it, you learn some tools, and with the extra time, you just smile. For neurodivergent folks, you’ve got a double wave, a tornado, and maybe a hurricane, and then you blow your own mind, and then you repeat and repeat. Then, that was a lot. Here’s my thesis. Emotional crisis state is basically the opposite of mental peace. This one’s very simple. Less time in crisis, more time in peace. There you go.
Let’s talk about prompts, bring this back to GenAI. The version 0.5 prompt, the beginner prompt, let me give you a scenario for the example. Let’s say I was at work and I was trying to explain something that in my head is so simple. The error messages in our application are going to confuse our users, and it’s going to cause us a bunch of support issues. I’ve been trying to say this for weeks, and I’ve also tried to say this four times in this same conversation, and nobody understands me. I just feel like shut down, because I just feel like nobody ever understands me. I know, logically, sure, sometimes they do, but I just, “Ugh”. That’s the prompt. I feel like no one understands me. I know they do, but help me get re-grounded, and then help me figure out why I always feel this way using CBT.
Let me highlight you some important parts here. Help me get re-grounded. That’s what I wanted to do first. Help me figure out why, so I can do better in the future. Using CBT is the laziest way of me suggesting an approach for the AI to follow. CBT is Cognitive Behavioral Therapy, which is pretty common if you’ve ever done therapy. What you get when you give that kind of prompt is a giant lecture, which is not what you want when you’re having a trigger. Here’s an example of the giant lecture that I got when I did this. Here’s some information about grounding in abstract. Here’s some more abstract about how CBT works. Like, ok, but no. Don’t worry if that wasn’t enough. Let me lecture you about self-compassion.
Let’s upgrade this, because that was not helpful. Let me give you a 1.0, and we’ll build a prompt together. The beginning is similar. Act as a therapist trained in CBT. I’m going to introduce something new, which is adding a process to your prompt. I don’t see people do this very often. It works really well. I’m adding this one step at a time clause. Confirm that I’m ready. This is a human problem. I don’t want you to move on and give me a dictionary. I want you to take me one step at a time, and when I’m ready, we go to the next part. I use that a lot. Let me propose a process. Very simple.
First, help me get grounded. Then, three loops of identify a cognitive distortion. This is how CBT works. I’ll show you in a second. That might be relevant. Then help me talk through it. Then you can send me off with some optimism for my day. Again, I give the situation. I feel like no one understands me at work. I know they do. Here’s the response. This is much better. First, a grounding exercise. We sit or stand in a comfortable position. We take a deep breath, which since we’re at a talk about mental peace, let’s all take a deep breath. Trust me, the rest of this works really well. Yes, I actually felt better. Let’s move on to step two, a cognitive distortion. Note how this is actually so helpful. Is it true that no one understands you, or are there specific moments you can recall or a certain context where you commonly feel misunderstood? Yes. Most of the time people actually do understand me. In this case, they don’t. This actually helped a lot, so we moved on to the next one, when I said I was ready. Another cognitive distortion, but I will skip past.
We got through the CBT. This is actually very interesting compared to regular therapy. You never have to worry that you’re offending your therapist. I don’t know if anyone’s ever had that experience before? It’s like, yes, you’re doing magnification. It’s like, am I? I don’t know. I don’t think this is useful. You can just say that. Then it asked again, as I said, it was three rounds. When I said no to the third, it asked if I wanted another one. I said, “Let’s just move on. I actually feel better”. Here’s my takeaway optimism. The fact that I like difficult conversations is a feature, not a bug. Expect to be misunderstood. It means they’re really difficult, which is what you wanted. How are you feeling now? Of course, I had to take the ADHD twist. Let’s move on. I feel great.
Why is it effective to use an AI therapist? For one, access. Therapy needs insurance, needs you to find a therapist, needs you to go to the therapist on time for the appointments. You have to learn techniques at therapy, but then use them on your own without your therapist there. That whole thing, very difficult. In this situation, you don’t have to learn anything. You just open up your phone. You use the voice mode. You start from scratch. If you want to use the 0.5 prompt because it’s easier, do it. If you want to use the complex one and get a better result because you feel grounded enough to do that, also awesome. Do it. You can even just open a previous conversation and just start from there.
I think also another really important part for neurodivergent folks is, because GenAI is trained on such a large dataset, it knows what it means to have autism or to have ADHD. The type of advice you get from neurotypical people when you have ADHD is so unhelpful sometimes. I’m just going to have to leave it like that. The approach is totally customizable. I gave that process, but maybe next time I’ll try a new one. How’s it going? Before I used to have one or two days of distraction and now I feel calm in less time than it takes to do a one-on-one. I always have to deal with this. That’s part one, recovering quickly from emotional turmoil.
2. Get Unstuck
The next stage is overcoming stuck points, because life is too short to be stuck. This is a famous quote from none other than me. Here’s the difference between what happens when a neurotypical person does a task. It’s pretty straightforward. You list out the requirements. You get the tools. You do the task. You relax. With the extra time, you take time to repeat on a new task. Here’s what task paralysis looks like. Get stuck, spiral, wonder why you’re stuck, repeat, repeat, yikes. Procrastinate, obviously. Blow your mind, get sad. The last one, it depends. Sometimes you miss the deadline. Sometimes this decision’s made for you. Other times you finish, but maybe you’re not so satisfied, because why did it take so long? For everyone, the thesis is the same. Work doesn’t stop.
If you stop, work will pile up. Piles of work lead to overwhelm, and overwhelm leads to not peace. The bonus thesis for people with ADHD is that this task paralysis can last a lot longer and comes with a bunch of mental chatter and guilt and low motivation. The state itself in this case is the anti-peace. The less time you spend here, the more likely you will find mental peace. Not good. Do not recommend.
Let’s go through and get unstuck. First of all, when you’re stuck, start with self-care, because sometimes you’re stuck for a reason that has absolutely nothing to do with what’s going on in your head or the task at hand. If you take medication, make sure you took it. If you drink coffee every day and today you skipped, that’s not good. It hurts. It hurts your head. Don’t do that. Eat food. Drink water. Move. Breathe. Still stuck? We need a new tool, Momentum Bot. Yes. The v0.5 of Momentum Bot is even simpler. Open up voice mode and just say, “I’m stuck, help”. Let me give you the 0.8 prompt. “Help, I’m stuck in a loop. I’m trying to get my slides ready for this conference talk, but I just can’t get myself started. I keep doing other filler work instead of what I’m supposed to be doing. Please help me get unstuck”. You’ll see parallels to before. Once again, you’re getting this giant framework that would be super helpful if you weren’t stuck already. Remember what I said about neurotypical advice? Giving you a whole framework at once, it’s a lot.
Decent, but I’m going to keep moving to the 1.0. In the 1.0, I actually came up with this off the top of my head. If it looks like a long process that I came up with over a long period, it’s not. You can make this up too, and reuse it. We mentioned the self-care, so check if I’m hungry. I know that I never remember that I can reward myself for doing things that I don’t like to do. It’s important that sometimes it will tell me to do that. Anything else you can do that would help, that’s a good catch-all. Throw a wild card in there. Reminding me that I have ADHD is always a good one. Sometimes that’s the answer. The process is simple here, similar to before. Choose whichever one of these you think is most likely to help the situation and talk to me about it.
After each loop, just ask me if I’m unstuck. If I’m not unstuck, we go again until I’m unstuck. Then I give the situation again. Here’s what the output looked like when I did this. It started with strategy six, which is to break down the problem. What do I need to do first? If that’s overwhelming, we can go to number eight and do conscious complaining. I love conscious complaining. I just want to give a shout-out to this. You can Google it. Conscious complaining is ranting out loud until you find it insane. Here’s the example.
There’s just way too much content. How am I supposed to choose? It’s going to take forever. I have way too many ideas. Why did I do this to myself? I know it’s going to be good, but couldn’t I have just chosen something else? It’s like, wait, I have too many ideas? Isn’t the problem usually that I don’t have any ideas? Too many ideas is great, isn’t it? I just have to get them organized? No, this isn’t so bad. Now, did the AI do anything? No, it’s just a rubber duck in this case. I love conscious complaining, so take that home with you. I’m going to skip through here and show you that it did a bunch of strategies. This reframe was really nice.
The topic may feel hard, but that’s because you’re sharing something that really matters to you. You’re the perfect person to do this talk, ADHD and all. The stress means you’re about to create something valuable. Let me tell you, when I was on Muni on my way here, I was totally remembering this and using this as a mantra in my head to relax before we started talking. It’s really helpful. I think that is giving a good demonstration of why AI can help in this area of mental peace. It definitely brought me some today. Again, pushing limits of technology. I think I covered enough of these. Sometimes the thing you’re stuck on is something that the AI can just start for you. Let’s say you have writer’s block. You’re supposed to write a memo. You don’t know where to start. Generate me a memo. It’s beautiful. That’s two.
3. Planning and Communication
Number three is to deal with planning and communication. Mega struggle for myself, at least. Do you remember earlier when we got stuck? Was it because of planning and communication? You might have heard this advice, do it the way that works for your brain. I love this advice. It’s so empowering. Then this happens. Some projects, you have to interact with other people that have a different brain than you, so we have to do something else. Structure Bot, your neurotypical shim for seamless planning and communication. Here’s my theses for this section. If you properly break down a project, then you get to enjoy this sense of progress as you go through it. When you have a sense of progress, you have a feeling of peace. There’s also this loop. When you plan properly and you communicate well, you have to repeat yourself less often. Less repeating of information means less interruption from people that don’t understand the information, which also leads to more peace.
Neurodivergent and neurotypical people can agree on this one. It just might affect neurodivergent people a bit more. This is why. I wish I could say that this was exaggerated, but truly, the right side of this screen really is what I feel like when I plan a longer than one quarter project. What is going on? Nothing has any structure. It’s all a bunch of tasks. I can see the big picture pretty well, but there’s just chaos when it gets any lower than that. Compared to the left side where we have a chart with some relationships and clearly subdivided problem and maybe even like the Kanban board, this is great.
I’m jumping straight into the 1.0 on this one because there’s so much to show you. It’s really cool. Let me give you the overview, which is that my team is making an orchestration service, this is real, that creates consumer accounts for special use or testing purposes. The reason, in this case, why I’m giving you a real example is because it’s going to actually help me do the work, not just talk me through it. I’m now in charge of the system’s reliability, and I just need you to get me started. Let’s talk through the problem, exhaust all the topics. T
hen we can break it into milestones, work lanes, and success criteria, and all of this. This is an aside. A theme I like to do with AI or a tactic, is to break things into stages so that you can confirm that you and the AI are on the same page at some point in the process. If there is a case like this where you want to enumerate the whole project before writing the memo, you should do that as a separate step. Because as you get further into the writing process, it’s harder to change the content. I always make sure to do that, and recommend doing the same. I give it the opportunity to circle back. We’re going to start with the outline format. That’s a way to confirm that we have the same content in mind. What we will see.
First, it’s doing the thinking. I used a thinking-capable model for this. This is a o1-preview from OpenAI. This is the thinking it does in the background. You can see that it comes out with this outline as I requested. Problem statement and reliability challenges is great executive summary content. Availability, scalability, fault tolerance, disaster recovery. Look at this stuff? This is so good. I didn’t put any of this in. This is exactly what I needed in my outline. We have proposed milestones, which are pretty good. I’m going to skip ahead.
Editing is a thing that sometimes I feel like people don’t use GenAI because of this. It’s very easy. We don’t manage data, so you can take that stuff out. We also have a security team that manages security. Continuous integration and deployment are already covered by another team. It was able to remove those sections, no problem. We move on. I ask for time estimates. Then I ask for a memo format. This is one of the most useful things, is what I like to call shape-shifting. If you’re not the best communicator or project planner, shape-shifting is the best. We can take this outline, and we can turn this thing into a memo.
Then, here’s the memo. It actually looks great. Similar to before, but a little bit buttoned up. We’ve got these timelines and a table, as I requested. I asked for two pages, because at my company, that’s the standard for these types of things. If it’s not close enough to the standard that you like, then just give it an example. It’ll come out just like the standard that you like. Scroll, scroll, scroll. Ok, great. Shape-shifting, now I have to give a presentation to my director to ask for funding for this project. I want you to prepare some talking points and prepare a Q&A.
Here’s my opening statement. It’s critical. Here’s problem information, solution information, value prop. Look at these questions. They’re so on point. Why do we need to do this now? Does it align with our goals? What’s your success metric? What’s the ROI? I actually did this. I just felt so calm that I was able to answer these questions in one sentence, because I had prepared already. It was amazing. For a special bonus, generate me a slide deck about the same information using reveal.js, which is what I’m using for this presentation. It’s just easier than generating Google Slides, because it’s in HTML, and language models are good at coding. That’s the shape-shifting, which is super useful. I’m going to, again, tell you to go look at the slides for the pro tip content. We did cover some of these here, about thinking models, using bullet points or an outline until content is locked.
4. More Flow
At this point, we have a tone change. This track is all about socio-technical resilience, and without saying the word resilience, I’ve been talking about things that are resilience topics. Recovering from emotional strife, getting unstuck, and dealing with planning when that’s not your strong suite. Let’s change the tone here a little bit and talk about things that are going to add peace, rather than preventing you from losing it. The next one is about adding more flow. Here’s the relationship, back to mental peace. GenAI can definitely save you time on routine tasks. That means you have more time to change your blend of work to have more meaningful and engaging work. Engagement leads to flow. Flow leads to peace. Better results also leads to a sense of satisfaction, and satisfaction leads to peace.
If you’re asking why I’m saying you’re going to get better results just by changing your mix of work, it depends on your brain chemistry. I showed this before about how neurotypical people tend to do tasks. Here’s a positive about ADHD, and autism for that matter. Here’s what’s going on when someone with ADHD disappears into the corner and comes back with the best thing you’ve ever seen. It’s a bunch of this. Why are they doing this instead of the previous section when they just got stuck? It’s interest. There’s a lot of research on this. You can look up the interest-based nervous system. Interest is everything for neurodivergent people. Because of that, as I said, when someone who’s neurodivergent is interested in what they’re doing, they will actually create better results. Better results, more satisfaction, more peace.
How do you skip over or shortcut routine tasks using GenAI? I’m going to say here that knowing when to use GenAI to shortcut things is actually more important in my experience than knowing exactly how. Let me just give you some heuristics about when. These are what I identify as the most successful or most likely to get done on the first try type of GenAI tasks. I want to also call out, there are many talks here at this conference about GenAI this year. I’m sure there’s a lot more deep takes on this topic. I’m giving ones that are more Copilot depth of difficulty, because this is about shortcutting.
If it’s too hard and you have to do a whole engineering build, it is not a shortcut. I wish it was, though. I love doing those. Things that are most likely to be successful, writing scripts, especially file processing scripts, writing a utility class, adding comments to all the functions in a file, rewriting from one language to another, writing unit tests. These are LLMs, large language models. They’re very good at language tasks. If your task is a language task, such as rewriting from Java to Kotlin, it’s going to do really well. If you can translate from English to French, you can translate from Java to Kotlin if you have enough training data. For all these things, utilities, language tasks, translation, and when things are all in one file, it usually works better than you would expect. It’s getting better all the time.
Then we have these, what I like to call the plan and then delegate use cases. We touched on this a little bit earlier when I was talking about build the outline together and then have the AI turn it into a memo. It’s a similar thing here. It’s, let’s design the data architecture together, and then when we’re done, you go ahead and turn this into a giant file of SQL statements. Once you did the first part right, the second part becomes a translation from pseudocode to real code or from description of tables to actual table statements. Expect success if you break the problem like this. Same for testing. If you’re especially trying to do edge case testing, you can walk back and forth to identify what the edge cases are.
Then, once you’ve agreed on what those important edge cases are, then ask GenAI to write all the tests for you. Then, these are things I would probably skip. I would skip anything that involves a whole bunch of files. Anything where you need the complete picture, it’s very unlikely you’re going to get it. RAG is very complicated. Don’t do logic. Just don’t. Document the entire application. Again, RAG is complicated. If you work in an obscure language and you’re like, all those things you told me, they didn’t work. Yes, that’s also a problem. Luckily, I work with Java every day and it’s great. My favorite, try to fix my CSS positioning. No. First of all, that’s impossible. Very hard. CSS is a visual task primarily. Stick with language tasks, I would recommend. The key thing here, though, is you just need to practice.
If you practice trying new things that you haven’t tried yet using GenAI, or regular things, but new way of trying to do them, you will often find some delight and some disappointment. That delight will teach you very quickly what will work and what won’t in the future. Most importantly here, do not shortcut the things that you’re interested in. If you do that, you have blown up the mental peace thesis. The goal here was to outsource the stuff that you don’t enjoy doing and do more of the stuff that you do.
5. Divergent Thinking
That lands us at divergent thinking. Let’s review first. For recovering from emotions, we’ve got the AI therapist model. For overcoming stuck points, we have the Momentum Bot. For the enhancing planning and communication, we have the Structure Bot. Using GenAI for routine tasks can help add time and flow. So far, we’re looking pretty good. We’ve calmed the seas a bit. We’ve managed to get some wins on the resilience front. We’re starting to turn to add more enjoyment at work. What’s going on with number five? What’s up with divergent thinking here? This is a heart-to-heart, so we’re going to see what comes to mind here. I hope from the framework I’ve laid out that when you all go back to work, you take something from this. I hope it helps you get closer to a source of sustainable mental peace. I hope it helps you understand your co-workers. I hope it just gives you some perspective.
The thing is, when we get there, when we get closer to this sustainable mental peace, or we spend more time in a state of peace, at least we can say that, where does that leave us? Life isn’t all about reducing suffering. We need to do something here. This is where divergent thinking comes in. Sometimes what you need to use GenAI for to find mental peace is to try something way out there. To find a new hobby that’s maybe similar or totally different from the ones you’ve done before. Maybe you need to visit a new country. Maybe you need to learn a new language. Maybe you need to give back or mentor. Or maybe the work you’ve been doing lately, it just needs a new flair. Or maybe there’s an opportunity that you can brainstorm about ways you can increase your scope in areas that you’re passionate about at work.
Of course, generative AI can do this very well because you can just ask for more divergent thinking. “That idea’s not crazy enough. That travel plan isn’t crazy enough. I need to use more PTO. I have unlimited PTO”. Think bigger. I think the lesson here is that since we all have GenAI at our disposal, maybe we can all be a little bit more neurodivergent. By doing that, maybe we can all increase our mental peace together.
ADHD and Autism (Diagnosis)
Do I have ADHD? Do I have autism? My friends have always told me I have ADHD, but do I, though? Yes, I super strongly recommend this website called idrlabs.com. This thing has like a million personality tests, and they have citations for all of the research that they base these tests on, so helpful. Please don’t diagnose yourself. If you get a very high score, go talk to an actual psychiatrist. If you get a really low score, you don’t, I promise.
Questions and Answers
Participant 1: Do you think this is just something that is really purely individual or is this something that progressive engineering leaders can start to maybe support somehow in the workplace? Anything you’ve thought about?
John Gesimondo: Definitely, I feel like everything I’ve talked about here, it affects different people to varying degrees, of course. As I said at the beginning, I think everyone can relate to that overall diagram. In terms of the individual aspect, the main thing that I think will have the most practical impact is if you suspect that someone that you’re managing has ADHD, or autism, or something else, figure out what they’re interested in because they might not tell you. As I said before, if they’re working on something that they’re interested in, it’s night and day different of how your experience will be working with them.
Participant 2: Are you not afraid that by using this technology you increase your mental health, but you also increase your productivity, I think? Are you not afraid that now it is an advantage for you, and at some point, everybody will use it and then productivity rises to a new level and then you will become blocked again trying to write those prompts? What do you think about that?
John Gesimondo: I had so much content that I wanted to put into this deck that had nothing to do with GenAI, but that’s where I decided to draw the line of what’s in and what’s out. It’s a very important question. How do you not just scope creep your way right back to now you’re busy and you’re already using GenAI?
Participant 2: Yes, indeed. Let’s say in one or two years from now, all your peers are doing it as well and what’s expected of you is to reach this level.
John Gesimondo: I don’t have the answers. I don’t think any of us have the answers, but let’s just hope.
Participant 2: We are going to find out, I think.
John Gesimondo: Yes, we are definitely going to find out. If nothing else, enjoy this period. You all made the investment to come here to this conference and to this talk. As such, you’re a little bit earlier on the adoption curve. Enjoy your peace for the next, let’s hope it’s like two years?
Participant 3: One of the questions that I’ve had when I try to use GenAI, because you have to get from prompt 0.5 to 0.8 and then prompt 1. There’s a bit of a curve there. How do you really get to the right point and not exhaust yourself trying to get the right answer from GenAI? I feel like in that time frame there might be a little bit of, “I’m just going to give up on this. They’re not going to give me the right answer”.
John Gesimondo: Are you asking this in the engineering sense or in more of the humanity prompts that I had at the beginning?
Participant 3: I think it’s a little bit of both.
John Gesimondo: For the human-ish ones, if you can’t figure out how to tune the prompt, just go back and forth. Just have a conversation. I think that tends to work out well, and by the end of it you’ll be closer to coming up with a better prompt for next time. On the engineering side, the best rule of thumb to use is, if I had to break up this question so that I could work on it alongside a junior engineer, how would I do that?
As I said before, some problems have to be planned and then executed. Let’s design the data and then let’s make the SQL statements. If you were working with a junior engineer, you might do the same thing. Take a first stab at this and then I’ll review your work. Then once we’ve agreed that that’s the right work, then we’ll go ahead and do the coding. I use this a lot. It also really helps identify a reasonable process to get there. That’s usually, at least for me, the first thing that comes to my mind is like, if I want to delegate or collaborate, how do I break up the problem so that that’s possible? That really helps.
Then I think over time you’ll get better at some of the techniques I used. Sometimes you have to say, think out loud because language models have to do language in order to think. Think out loud or organize or follow this process. Repeat this many times. There’s just a little bit of nuance of how to do these things but you get used to it. If nothing else, think of it as a junior engineer you’re working with, or just have a conversation and we’ll let it tune over time.
See more presentations with transcripts
