There should only be one primary action the user can take on the screen at a time. When there is more than one action to take it becomes confusing. Which should I do first (ordering)? Which am I supposed to do (decisions)?

As a simple analogy, think of the difference between a plane’s cockpit and a soap dispenser. The former is filled with knobs and switches for all sorts of functions, the latter has only one thing you can do—get soap.

If you want to make UI feel intuitive, requiring little to no explanation, distilling the interface into a series of single actions (choices) is your best bet.

That doesn’t mean there isn’t a place for more complicated interfaces, but if you see this happening, you’ll know it increases the level of understanding needed by new users.

Tags are used to create heterarchical systems that lets you iterate on retrieval. The tag names are abstractions like categories, projects, or ideas that are encoded on top of a set of items (like notes) to make it easy to find later.

A built-in assumption with tag-based systems is that you will use tags consistently—a tag vocabulary. That’s why tags are much easier to develop individually than as a group. It’s easier to map your way of retrieving information than having others agree on how to collectively retrieve information. Twitter hashtags are the exception, but I would argue these are a form of Schelling points.

See also:

• org-mode uses projects (hierarchy) and tags (heterarchy) to manage large numbers of tasks
• My tag vocabulary is used for driving functionality rather than organizing topics

Coordination in the absence of communication. For example, two players in a game can’t talk to each other but they want to team up for a quest. They show up at a time that seems most likely (noon) on a day that is most likely (New Years) in a place that’s most likely (an inn). This works because people can use the shared context of their world to superimpose coordination.

Does remote work make the principal-agent problem worse? If so, why?

A few possibilities come to mind:

• It’s actually not worse but empiricism makes us think it is
• Remote workers necessarily have more agency (either through technology or job role) and therefore greater conflict
• Remote teams need to be more deliberate about everything and principals are making it worse by not adapting to the demands of remote work

The following list of things are my choices for boring technology.

Backend programming languages:

• Python
  • Frameworks:
    • Django
    • FastAPI
  • Package manager
    • pip

Frontend programming languages:

• Javascript (TypeScript is nearly there)
  • Frameworks:
    • React
    • React router
  • Package manager
    • npm
    • Use create-react-app for as along as humanly possible

Databases:

• Postgres
• MySQL
• Redis

Deployment:

• Ansible

Hosting:

• AWS

Notes:

• I’m hoping terraform to become boring. It looks like more engineers have experience with it, but few have set it up from scratch.
• There are no boring package managers or build tools for the JavaScript ecosystem.
• Redis used to be novel, but now it’s boring which is good. Try not to use it in novel ways though.

The likelihood of pivoting seems to increase as the number of founders increases. Solo founders are the least likely to pivot compared to co-founders of 2, 3, and 4. The most likely to pivot is when there are 4 founders.

Pivoting is often needed to find product market fit although having lack of conviction to fully explore an idea can also cause failure.

Solo founders are less likely to pivot because they don’t necessarily have a support system to bounce ideas around making it easier to fool themselves.

Four or more founders pivoting most frequently might be consensus seeking and deference to each other. Maybe there’s just too much resistance which makes efficient decision making difficult to stick with (easier to decide to change than to move forward and be wrong).

Read Co-Founding Considered Harmful by Florent Crivello.

See also:

• Companies started by solo founders survive longer and generate more revenue
• Fundraising as a solo founder

The American Community Survey from 2021 by the US Census Bureau showed that the largest share of working from home was in major cities and metropolitan areas. For example, San Francisco, District of Columbia, Brooklyn, and Chicago are all in the top 10 in remote work.

This might feel counter intuitive—if housing prices are rising (blamed on remote work) and city office real estate value is declining (also blamed on remote work) shouldn’t suburban and rural areas see the largest share of working from home?

The most obvious reason that comes to mind is that the highest density of workers who can work from home already live in major cities. Young people of working age live in urban areas generally (“Millenials love cities” was a common theme from the late 2010s). The largest portion of people that would move out of urban areas to work remotely probably already have.

Finally, people have preferences that are better suited to cities (like me). Political views might also have some bearing on people’s decisions on where to live or just having things to do.

Read Working From Home Is Not an Urban Escape Hatch from the Washington Post.

The abundance of stuff—both online and offline—means that every conceivable niche is available to everyone. Search engines are niche machines that are increasingly sophisticated at directing attention to these fine grained interests.

Have these obscure interests always existed or does the search engine cause us to grind our attention finer? What or who are we ingratiating?

From a passing sentence in My Obsession by William Gibson on Wired.

See also:

• Maybe niche interests is a hunt for rarity like CryptoPunks are the Star Wars action figures of the future
• Identity is a powerful motivator for behaviors and “I’m the kind of person that wears rare vintage watches”

For as long as possible, in an early-stage startup, one of the founders should write the check for any expenses of the company. That includes payroll, vendors, SaaS subscriptions, professional services, and how much you’re paying accountants and bookkeepers (whom you may incorrectly think are looking out for the company’s cash).

The reason this is so important is that expenses tend to grow very rapidly as the company grows. Left unchecked, you can get surprised by large bills and overpay for a long time. Every dollar going out is a dollar less in runway.

It takes money to grow, but startup failure is nearly always due to running out of money. Shouldn’t you keep an eye on that?

See also:

• This is especially true when running a startup during a recession when raising more money is not always an answer to high burn rates

My personal site now gets 1,000 clicks per month from Google searches. It took two and half years of consistently adding content nearly every day to get to this point. There are 884 pages indexed by Google and 186 that are not (for mysterious reasons). It takes ~24k impressions to deliver 1,000 clicks. The majority of the content comes from my published notes rather than longer form content like blog posts or essays.

It’s kind of incredible how much content is needed to generate a comparatively small amount of traffic through SEO. This could be because the things I write about tend to be fairly niche and far ranging in topic.

See also:

• This is an update on the effects of note blogging on SEO

There’s now an official API for Grammarly and a language server. That means we can use eglot to get Grammarly in Emacs buffers. Here’s the setup which mostly works.

Update: Grammarly announced they are discontinuing the API so this won’t work after January 10th, 2024. There are no good replacements. If I find one, I’ll add it here.

You can get an Grammarly API client ID by creating an app here.

Build the Grammarly VSCode extension:

git clone git@github.com:znck/grammarly.git
cd grammarly
# Install pnpm if you don't have it installed already
npm install -g pnpm
# Build the extension and language server
pnpm build

Wrap the node based language server in a script:

#!/bin/sh
node /{PATH TO REPO}/grammarly/extension/dist/server/index.node.js --stdio

Add it to the path or copy it to a global bin:

cp /{PATH TO REPO}/grammarlylsp /usr/local/bin/grammarlylsp
# Make it executable
chmod +x grammarlylsp

Configure eglot to use it when in markdown mode:

(defclass eglot-grammarlylsp (eglot-lsp-server) ()
  :documentation "Grammarly Language Server.")

(cl-defmethod eglot-initialization-options ((server eglot-grammarlylsp))
  "Passes the initializationOptions required to run
   the server."
  `(:clientId "{GRAMMARLY API CLIENT ID HERE"))

(add-to-list 'eglot-server-programs
             `(markdown-mode . (eglot-grammarlylsp ,(executable-find "grammarlylsp"))))

A recent analysis of the trading volume of NFTs found that it fell 97% from $17 billion in January to $466 million in September.

Read NFT Trading Volumes Collapse 97% from January Peak from Bloomberg.

See also:

• Mania and speculation is a good example of investment strategies that only work in a bull market
• NFTs provide digital scarcity among digital abundance but the falling volume has more to do with speculation than the principle of scarcity
• NFTs joining cryptocurrencies as not being immune to market conditions
• Zero interest rate policy (ZIRP)

An effective way for announcing new features is to use the user/problem/answer tweet thread format. It makes it clear what the thing is and who should care. It also shows that you are thinking deeply about the right solution.

Another benefit to this format is that it helps you clarify the value. What’s worthy of going in the tweet thread? Maybe instead of writing the press release first we should be writing the tweet thread first.

Examples:

• Announcing Mosey Mailroom
• Stripe Atlas improvements

Cost basis tends to get the most attention when it comes to pricing. I remember business classes in college spent endless hours talking about cost plus pricing, margins, and so on. However, perceived value—the difference between the price and what people think it’s worth—gets overlooked. If the perceived value is high relative to the price, it’s a no-brainer to buy.

The set of features and the price establishes the perceived value. What you get for the price and how you describe it to prospective customers does much more to increase perceived value than lowering the price. Another way to increase perceived value is with add-ons that are priced independently (e.g. phone support $x/month) helps assign a value relative to the bundled price.

Read Three Ideas on Pricing.

See also:

• Versioning provides a way to sell information without cannibalizing high margins
• Willingness to pay should be at the core of product design
• Is your pricing too high?
• Monetizing Innovation

Workers moving from tech hubs to red states for remote work could be inhibited by conservative legislation like anti-abortion laws. Tech companies in particular tend to be in liberal states and have more progressive leanings. It seems likely that the states and cities that will benefit the most from remote work might be in blue states.

Managers are essential to scaling a successful organization but have a large blast radius if things go wrong.

The best organizations don’t suffer fools. The center of gravity (culture and decision making) is with individual contributors. This works because ICs are better evaluators of a manager than other managers or leaders, making it more obvious when it’s not working out.

I remember when I first started working at Stripe I felt like the dumbest person in the room. I was amazed at how smart everyone seemed and the writing…gosh, the writing! If I wanted to be like that too, something needed to change.

At first I thought I needed to improve my critical thinking. Looking for a solution, I poured over mental models and frameworks.

It didn’t really help though. I can hang my thoughts off of a neat model but can still get away with sloppy thinking—just packaged in a smarter looking format.

Turns out, what I’m really trying to do is ‘think better thoughts’.

(As an aside, it’s wild that humans can even have this kind of metacognition!)

I don’t want to be smarter, I want to be more creative, reason about problems clearly, and draw upon useful knowledge. Better thinking is a skill I can take steps to develop, being smart is merely a judgment (and a lousy one at that).

How do you think better thoughts?

One of my favorite Paul Graham-isms has to do with finding startup ideas. The best way to find amazing startup ideas is to become the kind of person who has amazing ideas.

As annoying as it sounds, the only surefire way to have better thoughts is to be the kind of person who has better thoughts. Afterall, identity is a powerful motivator and improving thinking is a collection of habits and behaviors.

To me, that’s reading better books and writing things down.

Read better books

(I’m going to focus on non-fiction books here due to their practicality in thinking better thoughts, but of course there fiction books which apply.)

It’s not like I read one book and suddenly think better, but reading increases the cardinality of thoughts I can have. Reading helps me think “wider” and combine more ideas—an important element of creativity and discovery. Thinking wider leads to compounding benefits as more ideas become available like an ever expanding pallet.

Reading books about thinking was surprisingly helpful. The Clock of the Long Now helped me think about long term thinking and deep time. The Beginning of Infinity helped me explore epistemology and how to not fool myself. The Minto Pyramid Principle helped me think about writing and how to explain things to others effectively.

Reading books opens up new pathways follow your curiosity. Those new pathways lead to other books to read and more ideas to think about. For example, The Beginning of Infinity led to thinking about consciousness which led to reading I Am a Strange Loop and physics.

Not just any book will do. I feel that “better books” are ones that are relevant to your interests but slightly off the beaten path. A simple heuristic is to look for canonical books in the field if inquiry that gets mentioned often by people working in the field but are seldom actually read.

Write things down

There are many benefits to writing and it gets even better when I write about what I’m reading. I take notes to clarify my understanding of the ideas I find interesting. I think about what I’ve read more if I’ve written anything about it. It sharpens the ability to express ideas to others.

This last point proves to be another important part of thinking better thoughts—writing grows knowledge and enables others to add to it. Human knowledge is meme replication so ideas need to be effectively replicated. This is critical for making progress as a group whether you are working on a team or building a company.

Now that I’ve written this down, the through-line here is that cultivating curiosity is a useful strategy for thinking better thoughts. Reading books and writing naturally leads to more curiosity.

The difference between “go read these books about philosophy, it’s good for you” and “why does Godel’s incompleteness theorem keep coming up in these other books, what did he find” are huge. One feels like homework, the other feels like discovery.

Enterprise Value (EV) to Next Twelve Month (NTM) revenue multiples in public markets for SaaS businesses rose sharply from 2019 to 2021 up to as high as 24x. Now, those multiples are falling below pre-pandemic levels to 10x. This has major implications for retail investors, VCs, and running a startup during a recession.

For example, startup fundraising rounds will come down (potentially for a long time). Growth is less valuable than revenue. Monetizing late stage valuations probably won’t work right now.

Read What Does the Post Crash VC Market Look Like? from Mark Suster.

One of the most profound findings from Godel’s incompleteness theorem is that meaning can be mapped onto a system that was specifically designed to prevent it. Principia Mathematica, the system Godel was poking at, was designed—amongst other reasons—to solve the paradoxes of set theory and logic with a constrained set of rules. Godel abstracted over the symbolic logic (converting proofs into numbers) to show a self-referencing statement could not only be represented in PM but was also undecidable.

This proof shows that new meaning can be mapped onto existing things through analogy and abstractions are real.

From I Am a Strange Loop.

When selling B2B software, self-serve and product-led growth is appealing business model (no sales people, account managers, etc.). However, it’s much more common that you would still need a sales team and account managers for enterprise customers. They expect to be able to talk to someone about their challenges and guide them through implementation.

See also:

• Very few B2B companies successfully implement PLG because not every product segment will have passionate users that would spread the word

Remote work has not yet made the jump to universality. We’re still in the phase of “specialized work objects” which are a combination technology to perform the work, jobs that can be performed remotely, and businesses that can operate anywhere.

What needs to be true to achieve remote work universality?

Technology to perform the work We already have this with networked computers and the internet. Access is widely available (though not ubiquitous) and software allows millions of people to do their job.

Jobs that can be performed remotely Some jobs are not possible to be done remotely, though I would argue this is more of an extension of the technology problem. Over time more jobs will be possible to be done remotely through automation, robotics, or applied AI.

Businesses that operate anywhere There are myriad rules and regulations that are different in every location. Compliance has no universal object that enables them to operate anywhere (though that’s what we are working on at Mosey).

To run a one-off ECS task from the aws CLI in Fargate you need to specify the subnet, set assignPublicIp to enabled (otherwise it will fail to pull the container image), and optionally include overrides to run the command you want.

Example:

aws ecs run-task\
--cluster foo\
--task-definition bar\
--count 1\
--launch-type FARGATE\
--network-configuration '{"awsvpcConfiguration": {"subnets": ["subnet-123456"], "assignPublicIp": "ENABLED"}}'\
--overrides '{"containerOverrides": [{"name" : "MyTask", "command" : ["python", "example.py", "arg1_here"]}]}'

Building tools to help users that are not experts in the thing they are using the tool to do presents unique design challenges. You can’t assume prior knowledge. You must provide a sense of progress towards mastery even though the subject matter might be unfamiliar. You need to simplify without being overly reductive. You need to identify the unknown-unknowns without being overwhelming.

A tool that enables non-experts to become experts is incredibly valuable especially when the expertise is difficult or expensive to acquire on one’s own.

See also:

• No-code tools built for a citizen developer
• Mosey helps anyone manage multi-state employment and tax compliance without needing to be a lawyer or accountant

Some of the biggest criticism of the remote work movement is that people aren’t actually doing work. There are several flavors of this: working in your pajamas, not doing real work, or, the most sinister, that the company culture will suffer. This fails closer inspection immediately.

All of these are a euphamisms for “I can’t see you working so work isn’t happening”. Unfortunately, this is a common fallacy, particularly for managers that need a sense of control. This position reveals a major failure of management—lack of trust.

Tell me you micro manage your employees without telling me you micro manage your employees.

All working arrangements require trust regardless of whether it is in-person or not. Remote work exacerbates this problem and forces managers to confront their parochial errors about how work actually happens and how to measure progress. Remote teams need to be more deliberate about everything, not the least of which because managers can’t rely on surveillance.