"Go West, young man, go West. There is health in the country, and room away from our crowds of idlers and imbeciles."
A $12 billion bet, and you’re not at the table
The new tools are incredible. 4 billion dollars of capital expenditure in 2025, aimed straight for the ability to generate correct code. $12B planned for 2026. “an AI model that can do everything a human can do”, but the thing they do best? Still coding. Incredibly and broadly capable, but like a fish out of water outside of the command line interface.
Computerization is driving most of new tooling being developed at the interface between engineering and the sciences. Tooling drives science, but there are not many new, physical tools anymore per unit of time1. Digital substrates simply iterate faster.
And so I have to understand TCP networking, SCADA, and the distinction between microcontrollers and microprocessors, to implement control algorithms; and RL, because the deep learning revolution manifested to me as the revolution of my friends in ML talking about everything in terms of deep reinforcement learning. It bugs me to use these terms so much without understanding what they mean, and the context they carry.
because the table is open to anyone
Literature suggests the way to be the best at something is to practice it, in as realistic a situation as possible. Compared to other STEM subjects, computers are one of the tightest versions of practice-reward complex ever invented. Can you train physically for 14 hours a day? You can use computers for that long. Do you want movies for free? Figure out torrenting, then figure out why it works. Do you want to be good at video games? Look at these statistics, and understand the algorithms that produce them. Do you want to make music? Learn how audio is represented and played back. Set up multiplayer game servers, and mod them. Compete on Kaggle or Hack The Box. Back up your course wiki page and host it. Crack your school’s RFID access control system. Pay for nootropics with cryptocurrencies; decompile the relevant part of claude code to fix one of the many bugs their 612+ incremental updates introduce; set up a personal site, make it pretty, and make sure your server doesn’t go down.
Chemical engineering has too high a capex barrier of entry for an individual, or even small teams, to iterate on. Everything from the documentation to data is internal, regulated and kept down on lock. In contrast, working with computers is like a breath of fresh air. There is freedom to work on what you’d like, choose between tools, discard or build alternatives to the worst ones, and push through entire projects in a month. And there are more basic luxuries, like how the source code is (almost) never mismatched with the compiled code, like a process and instrumentation diagram can be mismatched with the chemical plant.
The level of accessible documentation in most commonly used software is incredible. And on computers, and with electronics, because of the significantly lower capital cost, there is freedom to experiment and to innovate.
And the odds are in your favor
I’ve been following startups in my area of the sciences for a while. Of my darlings there are two types. Scifi Foods, Northvolt, and Believer Meats are the ones that haven’t made any money and died. Cornish Lithium, Ginkgo Bioworks and Solugen are the ones that are not making money, but are not dead yet. In 2025, a wave of cultivated meat startups died. Between 2021 and 2023, the same thing happened to biomanufacturing startups. In the 2010s it was crop waste ethanol and biofuels from algae. It’s kind of a pattern.
Compare this against the (relatively aesthetically bankrupt) Eleven Labs and Magic Pony - both Imperial College graduate founded software startups - which nonetheless are successful. I go to Boston and sit on a peer’s presentation, and they talks about their experience founding their own software company, bringing an optimization technique from quantitative trading to computational pharmacology. “since the acquisition my bank balance has been looking better,” she jokes. What is it with the ability of software to make money, and the inability of hardware?
As a friend writes, with computer algorithms, ”Consistency begets Universality”. Even if hardware and software startups take the same2 amount of capital to start, universality begets startups that work. Innovations in the physical sciences simply don’t operate on the same timescales as those in software, or the same physical uncertainties, or the same capital costs. And so, “Universality begets Scale”. I went into chemical engineering for how hard it could scale - $1 million throughput of product per hour, two engineers in charge of monitoring it all. It turns out, software has the same.
with options you can’t get anywhere else
It’s nice to be able to work from a computer. You can travel, customize your home setup, not go into the office. You can work at any hour of the day. The pay is significantly better than physical engineering, from graduate roles to ceiling pay - $200k is an achievable number3 early/mid-career. In practice, this means so much more money to donate to effective causes, slack to take career breaks, and buffer I can extend to the people I care for. And, in an understated but real way, pride in making more than my parents rather than less.
In the absence of any external force, I spend 12 hours in front of a screen each day, and enjoy this. Working with computers means that side projects can count as portfolio-worthy work. And it’s not just that fun happens to be accreditable, the work is fun. AI research is fun4, and the compensation is insane, considering how easy it is to get the reps you need to qualify and contribute.
Software is the easy way out, but I love the easy way out. Doing anything else “because it’s difficult”, is not individually rational for maximizing expected returns.
surrounded by your people
When I look at the people I would like to be like, they are all in software.5 People I want to be around - are good at things, have high trait openness, and do interesting work - cluster in tech. Maybe that’s because there are 5x more electronics engineers than there are chemical, and 10x more software engineers, and so it’s easier to find them from the bigger pool. Maybe it’s because many of them are naturally my kind of people at the personality level.
Software engineering is dignified work. It’s not dirty, dangerous, or generally dull. It doesn’t ask you to placate the general public as a service worker is. It is stable. All this means that people can build stable lives around this breadbasket - have spare time and energy to pursue things outside of work, on some days to just hang out.
San Francisco is the cultural centre of my world; not Wilson, NC (Believer Meat’s plant) or Marshall, MN (the Solugen Bioforge). It gets lonely outside of SF. I would like to get closer to the center of the action.
and the door is still open
Computers are new. If I want to, the knowledge I learn in computers can and will be applied to engineering problems. CAD kernels; PCB layout software, like KiCAD and Atopile. Why do our native process simulators seem so outdated? If I learn how KiCAD works, what can we bring across to our native process simulators?
What did young, smart, ambitious people find what they want in in each century and decade? In the past, it was taking the civil service exam; travelling to East India; being on Cold War projects. In the 21st century West, looking at the disproportionate success of big tech, computerized trading, and now big labs, it’s close to uniformly computers. Unlike many, I don’t think this trend is about to die. People using computers can do, on an absolute scale, more and more each year, and Jevons then applies.
…data on productivity growth by sector since 2000 from America’s Bureau of Labour Statistics. Between 2019 and 2024 the “information” industry—which covers areas from software and telecoms to publishing and film-making—came top with an annual rate of around 6%. The Economist
Chemical engineering, in contrast, is static. New, more efficient unit operation have been developed in the last 40 years but rollout is glacial - mostly because of the 50 year operating lives of new commissioned plants. Computers as an industry is dynamic - we’re in a boom right now. Old platform booms, like telegraphs, electricity, oil and railways, tend to last 10-15 years. To this one we’re about 3.5 years in. I fully intend to catch the next 6, and position myself to work on general purpose robotics - probably also driven by cloud models! - which may be just around the corner.
If you’re not doing computers in 2025, what are you doing? Computers promise everything, and are delivering on that promise. The future for it is so bright it’s almost blinding.
This is specifically the case for chemical engineering and construction, and less for pure biology, of which cryo-EM, nanopore sequencing, expansion microscopy and more are maturing - though nanopore sequencing and cryo-electron microscopy both relied on algorithmic improvements to work.
Hardware on average raises only 8% more funding, or $6.6M, up to Series C. - Grant Gregory, The State of Adventure Capital (Cantos, 2025) Slide 184, data from Carta via Peter Walker.
At select Bay Area interpretability shops
n=4 people have told me this
Make ten friends out of financially secure transgender women off the internet and 9 are employed as software engineers.
But it’s not just trans women. 3/5 of my cousins who are doing well, australian and chinese, are ALSO in software engineering!
Your best post yet