Eric J Ma's Website

This blog post is for anyone who is experiencing, or has witnessed people experiencing, AI-related burnout. I consider it a post-mortem of my own experience.

For the first two weeks of April, I was experiencing a severe bout of anxiety. To those who know me, this is very much foreign and out of whack from my usual self -- which I describe by my spirit animal, a capybara that can calmly sit atop a crocodile and be at peace with many other animals. In fact, my profile pic at work is a capybara. Anxiety is decidedly not part of the capybara persona.

And yet, during those two-ish weeks, I observed myself experiencing the following symptoms:

1. Heavier breathing
2. A tingling nervous feeling in my arms
3. A lack of desire to get out of bed
4. Racing heart rate when I arrived at work
5. General elevated heart rates throughout the week
6. No desire to use my personal laptop for hacking after hours
7. Decision paralysis at home, not knowing what to make for breakfast, for example
8. A desire to just binge watch youtube on the couch and in bed

Observation 6 was really telling: most evenings, I'm hacking on my laptop. I'm hacking with AI. I'm tinkering with new tools. I'm blogging in Obsidian, like I am now when I'm writing. But during those two weeks, I didn't want to do anything.

There were additional compounding factors at work. I was overbooked at work, frequently having triple or quadruple bookings on time slots. (These stress me out, I don't like saying no to people by way of declining an event.) There were issues with operations and personnel that were way less exciting than learning about quantum computing, designing and architecting ML systems for molecules, and guiding my teammates on AI evals.

Moreover, it's spring, my least favourite time of the year, because, swirling like an unrelenting dust hurricanes criss-crossing multiple states, plant gametocytes unleash terror on every cell and connective tissue on my eyes, nose, and lungs. 'Tis the season to be sneezing.

For the majority of those two weeks, it was a pretty shaky time for me. I simultaneously became snappier towards those who were asking me impromptu questions at work (whether on Teams or in the office) and more rant-y towards those who were asking me how I was doing.

Unsatisfyingly for me, I still cannot pin down the exact cause of the stress. But what helped me here was a change in perspective.

Firstly, the timing of a week out in London, UK was superbly fortuitous. My wife planned it out, and, instead of being a leader and taking initiative, I just followed along with the flow. This helped immensely! I was, after all, suffering from decision paralysis. Not having to weigh the consequences of decisions was a weight off my shoulders. I just quietly followed others' lead in our travel group. (We traveled with another family together.)

Secondly, while in the UK, we visited two museums whose exhibits helped me reset my perspective on time. We saw ancient relics from prehistoric to Biblical times at the British Museum, and cosmic time scales at the Science Museum in the outer space exhibits. Pondering the passage of time, it's hard to resist comparing those timescales to the compression of time I've experienced at Moderna with our multiple reorganizations, and to the first quarter of the year when I went all-in on experimenting with coding agents.

Yes, time at Moderna seems to be compressed, and with the restructuring of the company over the years (I finally understand why it's called restructuring, and I don't mean that cynically, but in the most precise sense of the word), some folks have had 7 managers over two years. I have been fortunate to have only had one managerial change. And the pace at which I was able to test ideas with coding agents was fascinating, thrilling, and exhilarating, but it also left me feeling unmoored and drifting, sometimes questioning the permanence of what I was building.

But the contrast of the relative permanence of ancient relics and the immense scale of cosmic time when compared to the three months of coding agents or four years of work at Moderna reminded me of so many interconnected ideas, of which this is just a sampling: the sediment of the fossil record being like the sediment of software, of the teachings of the Teacher in Ecclesiastes:

"Meaningless! Meaningless!" says the Teacher. "Utterly meaningless! Everything is meaningless."

Ecclesiastes 1:2 NIV https://bible.com/bible/111/ecc.1.2.NIV

Meaningless here comes from the Hebrew word Hevel (הבל), which has a deeper meaning than mere meaningless: it is a metaphor for the transitory nature of our work -- and even more so for AI assisted coding, in which we can, at will, reshape entire codebases to fit the image in our heads, or, even more, fit the vibe that we seek.

Pondering the natural evolution of human systems and innovation, and contrasting that against cosmic time and long arcs of history, made me realize how things that last are rarely only due to thoughtful input from humans -- they persist in a path-dependent fashion, reliant on both (a) timeless design and engineering, and (b) path dependent reliance on circumstances, demand, and opportunity seized by their creators.

Evidently, this grander and broader sense of time gave me much to ponder on.

Thirdly, I saw a different pace of life. Brits, like the Swiss folk I saw in Basel, go out for drinks at 4 pm. It's part of the social fabric, the "ways of working". This is much unlike my frenzied pace at work where I am still fielding calls at that time, or catching up on the dozens of unread Teams chat messages that have piled up while my attention was diverted somewhere else in a 1:1 or other meeting.

Witnessing the slower pace helped me reset my own expectations on my time. It also relates to Mario Zechner's talk on the pi coding agent he created, in which he appeals to the AI Engineer audience to slow down. It also helped that we mostly walked everywhere, rather than zipping around the city in a car (unless we were genuinely tired) or an ebike -- the latter being how I usually make my way around the greater Cambridge area. Again, it was an intentional change in pace.

Fourthly, having a week out where I did not check my work phone or laptop -- I left them at home -- and could focus on my family was insanely refreshing. By the time March rolled by, I found I had not paid enough attention to them at home. At first, my kids would ask me to play with them every night, but over time, as I turned them down over and over, they got acclimatized to playing on their own and stopped asking me. I'm glad I noticed this earlier than later! They grow faster than we realize, and being present is, I now realize, very important for me, especially given the circumstances I experienced when I was younger.

Finally, as I had a chance to catch up with my younger brother Evan and his wife May, I observed a different set of life priorities. They really know how to enjoy life! And they deploy their finances in a way that reflects this set of priorities. They've been all over Europe making memories, tasting food, and more. And yet for myself, even before I got married and had kids, I was always hesitant to spend my own money to enjoy life. (A bit of this might come from seeing first-hand financial decimation due to debt one generation above, and from being a student for another 10 years after graduating from junior college in Singapore.) Instead I would redirect my time and energies to working and learning... but sometimes at the expense of my health and enjoyment of moments around me. I would travel only for work, and that severely limited what I saw.

There are clearly other things that helped me reset.

Averaging about 13K steps a day definitely helped to the point that I picked up the habit of morning walks upon returning to Boston.

An afternoon playing retro computer games in the Science Museum, in which I picked up the game Kingdom Hearts for the first time and played Pac-Man with my older kid, helped give me a good few hours of mindless enjoyment.

Not talking AI all day long helped remind me that there is a world of stuff happening outside of wafers of sand pulsing with electric currents.

I used my phone to connect with Scripture again, and it gave me many indirect theological reminders that the technological tools I have at my disposal need to be in the service of others, not myself.

And of course, as someone who enjoys food much, it was amazing to try out new dishes in London's food scene.

If these all sound mundane to you, the magic may simply lie in the mundaneness that humans need. We need the stillness of meditation and the connection with other people. By contrast, AI is both incredibly stimulating and isolating -- it keeps the mind buzzed on solo work, taking away the old frictions that necessitated the negotiation between real human beings, conditioning us to accept sycophantic answers... Quite significantly, it is the antithesis of what allows for humans to thrive, if we develop a reliance and attachment to it.

If you've made it this far, then here are some of my recommendations, if you're also feeling the same kind of burnout I was experiencing. These are also the changes that I will be making going forward.

1. Plan the work you want to accomplish for the day with AI and then stop when you've accomplished it. It should occupy the four best hours of your day. Allow for space to think and strategize, and to handle administrivia.
2. Touch grass daily, literally, please.
3. Go for a long meandering walk each day. Do it first thing in the morning if you can get up. It'll be beneficial for your daily brain and your body. Do not bring your phone. Let your mind wander, enjoy nature, the sound of birds chirping and leaves rustling.
4. If you are building something for someone else using AI, make sure you have constant contact with that person or group of people. It is very rewarding to hear feedback on how things can be made better, and it is superbly grounding too.
5. Learn how LLMs work and are trained. They will help you appreciate why AI is not a sentient being to become attached to, and should instead be treated as a (very high power) tool to be wielded in creative ways.
6. Go play some board games with real people. At my level of sophistication, tic-tac-toe and Zingo with my kids works perfectly well. Find your crowd.

Point 5, I believe, is the most important thing to remember. AI is a tool, and a really high powered one at that. It is tempting to squeeze out every last ounce of productivity -- but to the detriment of yourself. As I wrote in an earlier blog post, there is a sweet spot of amplified capacity that you can hit, but shouldn't go any further. It is tempting to count the opportunity cost of not pushing yourself just a bit more, but as Jeremy Utley (who quotes Clayton Christensen) puts it bluntly, it's not worth the while. (How will you measure your life? by Clayton Christensen is also worth a read.) We are human beings, not human doings. Learning how LLMs work and are trained will help you appreciate that LLMs and coding agents are just stochastic parrots, albeit nonetheless a useful class of such parrots when steered properly.

I hope reading this blog post helps you. Above all, if you're genuinely feeling anxious because of AI, know that the hype will die down, and the genuinely useful patterns and parts that help us live better lives will remain. We will need to redesign our economies, societal structures and sense of where we derive our worth, no doubt. Humans can have tasks automated, but what makes us human, and thus valuable, is not replicable in a machine. Ultimately, people want to be in relationships and fellowships with other people. And that makes us indispensable.