I would guess that a significant amount of software is written for one person and we should celebrate it more.

Most software engineers tend to write for other people. They hope to create a popular open source library. They’re working on their startup idea. Maybe it’s a matter of industry norms i.e. pretending that the next engineer reading your code is an axe-wielding murderer.

Why not write one of one software? Written just for you, to do exactly what you want, without consideration for someone else. What if we didn’t place such a high premium on popularity and scale?

I imagine there are many other kinds of software that would be created that aren’t today. It would be more fulsome and go deeper into the problem instead of a shallower Pareto optimal of building for other people. It would be limited by the author’s time and ability so the shape of it might look very different—a greater emphasis on combining things that already exist and smaller in scope.

I imagine the ceiling for one-of-one software is significantly higher. For one, there is no faster feedback loop than the one you have with yourself and so you could rapidly iterate. With sufficient skill coding becomes the most convenient option and software becomes a malleable material for making what you want.

What are some examples of one-of-one software you’ve written?

See also:

• A simple example is my Emacs config which is a thousand lines of code that will only ever be executed on my CPU
• I built org-ai and an emacs integration just for chatting with my notes

Writing code is also writing for other people. We use many conventions in written communication to help with readability. Ever sit down to read someone else’s code and it’s giant wall of procedural statements for 100 lines straight? Imagine reading a paragraph that goes on for 100 lines.

I like to format code like prose. Statements in a long function that complete an idea should be grouped together visually—think of it as a paragraph and add some newlines for the reader to take a breath. Write linearly so it’s easy to read from beginning to end (you don’t read a book by jumping around back and forth between pages). Organize hierarchically from most general to specific (e.g. the last line is generally expected to be the return value so use early returns for edge cases rather than the main case).

This question comes up a lot from founders starting a company and raising a seed round—how much should you pay yourself?

On the one hand, you don’t want to be distracted worrying about your livelihood so you can focus on the business. On the other hand, it would be very odd to pay yourself a high salary while also having the largest upside if things go well via ownership.

Benchmarks of what other companies are doing might help. A survey of founder salaries by Pilot found that startups that have raised between $1MM and $3MM have an average salary of $109K. However, the average is not particularly useful and there are plenty of founders who will pay themselves $0 or the lowest amount legally required (minimum wage). Kruze consulting found that average startup CEO salary was $150k in 2023 (median $140k).

As a general rule of thumb, if you raise a seed round between $2MM to $3MM and you live in a high-cost-of-living area (SF, NYC) a startup CEO’s salary should be around $150k—any more should raise questions. As another rule, if an early-stage startup CEO has the highest salary in the company something is probably off.

See also:

• This might change when running a startup during a recession, but from the benchmarks listed earlier that has not affected CEO salaries in 2022
• The number one job of a startup CEO is finding product market fit

New York City’s largest union, District Council 37, is in the midst of negotiating a five-year deal that includes, among other things, protections for remote work. This is the first time I’ve read about large scale union negotiations that talk about remote work and, for NYC with it’s high vacancy rates, remote work could drastically improve staffing.

Read New York City’s Deal with Largest Union Would Include Remote-Work Plan

See also:

• Manhattan workers are spending $12.4 billion less than last year
• NYC office real estate value declined due to remote work
• Why hasn’t remote work become a political issue?

It always seems to take awhile to get resolve an issue with a state agency. They might send you a letter that you owe them taxes. You might respond to said letter and pay the taxes. You might then receive another letter…for the same taxes you already paid.

Why is that?

The way to think about state agencies and—by proxy—your tax accounts, is that they are eventually consistent. Changes can happen to the balance of your account, but what you observe (e.g. a letter) might not accurately reflect the current balance of your account. It’s not just balances—pretty much any information that changes will take time to settle into the right places.

Part of this is run-of-the-mill bureaucracy, but these are also large scale organizations and with distributed systems. (If you think about it, even moving papers around a building or two is a network which comes with network faults). Without eventual consistency, they would be unable to serve as many users and handle the complexity in which they serve them.

What gets you into a creative field is having really good taste. When starting out, you’ll often be disappointed because what you create doesn’t match your taste. The gap is why people give up on creative endeavors.

The way Ira Glass describes the taste gap perfectly matches the feeling I get when starting something new or trying to acquire new skills. I spend considerable time refining my taste and judgment which pushes the goal posts out a bit further. The gap is what makes me grow, but I have to be willing to be bad for awhile.

See also:

• Taste is the refined sense of judgement and finding balance that produces a pleasing and integrated whole
• Good taste must exist because good artists exist

One of the many challenges of staying compliant is that the feedback loop can be incredibly slow. By the time you find out there is a problem, it might be months after it became a problem. For example, if there is a payroll report that was not filed in Q3, you might not receive a letter about the failure to file (and penalty) until Q1 of next year.

One of the reasons for the slowness is the reliance on snail mail. It can take weeks to post the mail then more time to deliver it and yet more time for someone to check it.

Sometimes the only way to know there is a problem is by receiving a letter in the mail from a state agency. This makes mail essential to staying compliant—you can’t fix the issue if you don’t know about the issue.

Unfortunately, mail is difficult to manage for multi-state businesses. They need to consistently use a mailing address everywhere they register. They need to make sure they actually check the mail. They need to make sure the right mail gets to the right person to take action.

A recent study found that workers in Manhattan are spending $12.4B less than last year on meals, shopping, and entertainment near their offices. The author says this is due to remote work as workers are spending 30% less days in the office. This impacts local businesses that cater to office workers but also city tax revenue.

Based on consumer spending, it’s unlikely that this money is being saved, just spent elsewhere.

Read Remote Work Is Costing Manhattan More Than $12 Billion a Year from Bloomberg.

See also:

• NYC office real estate value declined for similar reasons
• On the other hand, not having to commute for five hours is equivalent to a 10 percent raise
• Remote work will become a political issue

In There Is No Software Maintenance by Henrik Warne, the author argues that software maintenance is just product development. Since software is never “done”, all of the things we call maintenance such as bug fixes and improvements are feature enhancements and iterating on the original feature.

I find myself agreeing that there is no definable “maintenance” period that happens between shipping and improving (or fixing) the it, but what developers actually fear is “maintenance mode.” This is end-of-life in disguise and one can’t help but feel like a bag-holder when putting effort into a thing that is going to be replaced or un-shipped—especially if this the majority of one’s job.

See also:

• Certain strategies can help reduce the feeling of maintenance—the default strategy for dependencies should be to not add more, choose boring technology
• Compatibility is leverage so using tools/frameworks that focus on compatibility reduced maintenance burden of upgrades
• The npm package ecosystem is a Chevette

In a paper CEO Stress, Aging, and Death, the authors studied the effects of managerial stress on lifespan and visible signs of aging. They found that CEOs' lifespan decreased by 1.5 years in response to an industry wide downturn. They also look older than there age by one year over the decade following a ‘distress shock’ like the Great Recession—James Donald, CEO of Starbucks, looked 3-5 years older than his biological age base on pictures from 2004 to 2010.

See also:

• Stress and anxiety are cumulative
• Running a startup during a recession

A study that looked into the effects of remote learning on grades found that attractive students, both male and female, received better grades during in-person education than remote due to COVID-19. It also found that the effect of better grades for attractive male students persisted while the premium on grades for attractive female students dissipated.

Read Student beauty and grades under in-person and remote teaching

Many people (myself include) bemoan paying taxes. You might not agree with the lawmakers and policies for spending tax dollars but it does help people. The fundamental disagreement centers around who it helps and how much it helps, not that it helps at all.

Maybe we’re thinking about it all wrong.

Paying taxes is an index fund where you pay in one place and it is distributed to thousands of state sponsored causes. The mix of these causes changes over time but, if you think about it, it’s a highly efficient mechanism for allocating money that would be impossible to do at such a large scale yourself.

Could it be better? Absolutely, but like an index fund, it’s a blend of a thousand things and there is baby/bath-water problem if you think we should just never have to pay taxes.

See also:

• Civilization is composed of fast layers and slow layers, governance is one of the slower layers
• Governments are comfortable thinking long term when it comes to infrastructure
• Taxes might also be the only way

A recent study from Northwestern Mutual found that Americans expect to need $1.25 MM to retire comfortably, up 20% from last year. This is a staggering amount more than most people have considering the average savings of Americans age 55-64 is $57,800 and the median is $6,400.

In the early days of a startup, hiring the early team is one of the biggest challenges. It can be tempting to hire contractors and part-time workers to get some help in the short term.

This is not a good idea because they won’t have skin in the game. Part-time employees won’t have enough time to take over any important responsibilities leaving you to do them. You won’t control the schedule of independent contractors so it’s more difficult to plan and execute. Part-time workers and contractors can also leave suddenly which is disruptive and generates more work to replace them.

Full time employees care more. They’ve hitched their career prospects to your startup. By taking added risk, they have more to lose if things don’t go well. This is an important advantage of early-stage startups—the most effective people care a lot.

These days colima is a viable alternative to Docker Desktop on M1 Apple silicon. The main issue I have migrating is the speed of file system syncing which become apparent when working on frontend React with hot reloading.

First install colima:

brew install colima

Then create a profile:

colima start myprofile --edit

Run the virtual machine:

colima start

Run docker as usual:

docker compose up

In the modern economy, managerial capitalism is the prevailing way companies are run. While startups tend to resemble bourgeois capitalism at the beginning (founders with full ownership and control), they eventually transition to a managerial model to scale (managers with low ownership but full control). Ironically, this transition is why startups are possible in the first place—it creates an opportunity for new startups to innovate. We know this is true because the big established companies don’t continue innovating and building all the new things themselves.

Dwarkesh Patel’s interview with Marc Andreessen

The classic model of capitalism (bourgeois capitalism), in which the proprietor of the business fully owns and controls the business, was replaced by managerial capitalism where managers, despite owning small amounts of the company, exhibits full control over the business.

As a result of the massive scale of modern enterprises, it was inevitable that a managerial class are needed. The upshot is that highly skilled managers can operate increasingly large and complex businesses. The downside is that managerial capitalism takes less risks and, as a consequence, new things don’t get built.

See also:

• Baumol’s cost disease shows how management salaries and productivity are also decoupled
• At worst, managerial capitalism is a negative art—significant gains come from avoiding risk rather than taking it for higher possible gains

There is a noticeable difference between the speed of the most productive software engineers I’ve worked with and the slowest. What do they do that others don’t? Is it possible to learn how to be faster?

Speed is easy to dismiss as unimportant for knowledge work (thinking is often the limiting factor), but that we can’t go faster or trying to wouldn’t be valuable.

Reduce friction to thinking

Software engineering is often a ‘shots on goal’ endeavor—the more attempts you can make in the same amount of time, the more progress you will make.

It should be easy to write some code and run it. How much friction exists in your setup that makes it difficult to run code? Tiny pieces of automation and muscle memory play a factor (e.g. checkout branch, open a file, execute just the code you are working on, stubbing out data, etc.).

Reps not years of experience

Solving problems quickly benefits from having seen and solved many problems, but don’t conflate ‘seen many things’ as ‘number of years of experience’. You’ve probably worked with people that have decades of experience that are below average in productivity.

The way to get experience is to build many things. I don’t know of any shortcuts to learning by doing other than to have side projects and really try to build them out.

Be decisive

When faced with a new problem, you can cut down on time by doing a quick breadth-first search (what solutions are available) then being decisive about a path to go down. So much time is lost by not being decisive.

Execute quickly and leverage your tools

Here’s where being in the top percentile of proficiency in your local development environment and programming language pays off. Code is very easy to change and you don’t need to be perfect until you hit production. Being really fast at expressing the solution will help you iterate through each sub problem quickly. The ability to think through code looks effortless from the outside because the distance between what’s in your head and what gets implemented is so small.

A good heuristic for when you’ve reached this kind of proficiency is when others comment about feeling nauseous watching you code.

See also:

• Speed is undervalued
• Typing fast is thinking with your hands
• The Godzilla speech applies here, treating code as destructible helps you move faster

You don’t need full-time G&A roles as your first 10 hires in an early stage startup. There is not enough work at that stage and you are better served outsourcing these roles or doing it yourself. For example, you can hire an external bookkeeper but run payroll yourself using easy self-serve tools like Gusto or Rippling, you can hire a lawyer to write contracts but handle payments yourself using Stripe.

It is very tempting to hire for G&A roles because it feels like “real businesses” have them. While true, the early days of a startup have more important priorities like building the product and finding the first few customers, every dollar and each hire is crucial. That’s why founders generally prefer software tools to give them additional leverage to inexpensively cover more ground until it makes sense to build out the function.

Because the human brain is a universal machine, it contains multiple strange loops that are coarse-grained copies of other strange loops housed in other brains at varying degrees of fidelity. Those you resonate most with in life (family member, a spouse, etc.) you know so well you can almost feel their feelings, recall their memories, and experience the world they would. These can not be explained by mere rote memory (perception is not reception and awareness) but something more closely resembling our own self-hood.

When people pass away they live on, in the sense of consciousness, through these collective copies. The fact that they are imperfect copies does not detract from this simple idea. After all, we wouldn’t say that a book translated from another language is somehow a different book.

From I Am a Strange Loop.

Anyone can follow instructions—performing a well-understood task step-by-step until it is completed. However, most challenges a startup faces are not well-understood.

It’s best to build systems that lead to making the right choices most of the time and optimize for hiring people that work from directives (like goals, policies, roadmap). This enables people to solve problems they haven’t encountered before without every detail needed to be figured out.

See also:

• Principal-agent problem
• A team charter is a mission plus metrics
• Themes conceptualize a roadmap

The easiest way to know that a new hire isn’t working out (and not fool yourself in the process) is to have a plan before they are hired. A job description and 30/60/90 plan are important tools to clarify what you expect this person to do and the key milestones to get there. Without them, you will confuse yourself and it’s unfair to the people you hire—feedback will feel arbitrary and unfair.

When the new hire onboards, use the 30/60/90 to anchor all feedback. Assess their progress by sharing written feedback and make adjustments you both agree to. If they are not hitting their goals or it’s obvious they will be unable to take over a key responsibility of their role, it’s time to part ways.

How do you evaluate a job role you’ve never managed before?

Often times at a startup, you will be hiring people that you do not have any background in or prior experience managing. Rather than try to derive the job from first principles talk to 3-5 experts in the area about what you think you need and what they believe a great person in that role would look like. You can stop when you start hearing the same thing from multiple people.

Use this information to inform recruiting (job description, what recruiters should screen for, what profiles to target), interviewing (key areas to evaluate in interviews), and onboarding (30/60/90 plan).

A country with a frontier is shaped by it. It favors people with rugged individualism because common services are not readily available without an existing economy in place. Positive-sum interactions in settled areas are required because people always have the option to leave. Finally, people seeking high variance opportunities will follow the frontier in search of outsized gains.

Eventually, the frontier dissipates as it becomes settled and formalized. At some point you don’t need Lewis and Clark, you need a town to trade goods and a courthouse to adjudicate laws.

The frontier thesis is particularly appealing for explaining American culture and industry. We often refer to new areas of business as a frontier and it’s not wrong—it’s unregulated with high variance outcomes that appeals to a different kind of person (with crooks and bandits along the way). Eventually the frontier becomes settled and other kinds of people are needed to make it prosper.

See also:

• The internet was (and maybe still is) a frontier which has American values encoded into it
• There is a downside, these frontier-driven values are incompatible with communal threats
• While land-based frontiers are finite (there is only so much land on Earth), there are infinite digital frontiers because computers are universal objects