In this article I want to explore different aspects of scaling a company that need to be taken into account if you're trying to grow exponentially. Generally when discussing about growth we consider revenue growth to be the metric we're measuring. One thing that is important, and that is considered throughout the article, is to be a profitable company. There's no point in growing exponentially if it is unsustainable or your expenses are always going to be superior to your profits.

In the best scenario, your business grows exponentially without requiring any additional human power. The second best case is that it grows sublinearly, meaning that as demand grows, the amount of people required to support the company grows slower than the demand itself. Past this point, growth either matches with how many people need to be hired, or worse, growth is slower as you hire more people. Here you may have exponential human power growth to get linear (or sublinear) revenue growth.

When hiring people, you need to be able to onboard them while having as little impact as possible on the existing employees. If existing employees need to spend time onboarding new employees, then nobody is working on product development or support during that period. Here again, you want the number of people responsible to onboard and train new employees to be smaller than the number of people being onboarded. In other words, you want the ratio of onboarded to onboarders to be higher than 1 to 1.

When a company grows exponentially, every time the company effectively double due to new hires, the culture is now at risk of instability. If for instance 100% of the existing employees were following the existing culture, and 100% of the new hires have a different culture, then there will be a culture clash with no specific half of the company being in control of culture. Unless culture continues to be managed through the process of onboarding, the initial group of people and their culture will rapidly become diluted.

We're not made to interact with dozen or hundreds of peers. As such, it is close to impossible for an individual to work with more than let say 10 direct peers. This means that teams or groups need to be created. This also means that interaction between teams needs to be managed somehow. Having multiple people from a team interact with another team leads to the implicit creation of a larger 20 people team, which does not work very effectively. On the other hand, having single points of contact between teams is a surefire way to have single points of failure. In this specific instance it is definitely necessary to have at least 1 or 2 redundancies.

It is also important to keep the people that have been at the company the longest. As the doubling period effectively divides by two the average duration an employee was with the company, losing those senior employees will speed up the drop in employee average tenure. This means that a lot of institutional knowledge will be lost along the way. This would also mean that you should possibly prefer to keep a senior employee at the cost of hiring a new employee, unless you believe that the amount of time necessary to onboard and be effective is lower than the average employee tenure.

The process of onboarding should be optimized as much as possible.

Anything that requires human power to scale linearly or superlinearly with itself needs to be optimized so that it grows sublinearly.

02 Jun 2021

One on one

History / Edit / PDF / EPUB / BIB / 1 min read (~126 words)
processes
  • Define and agree on the purpose of the one on one
  • Define the topics to be discussed prior to meeting
  • Track and update metrics that you care about every time you meet
    • e.g., Predictability, Ownership, Purpose, Progress, Belonging
  • Discuss the status of those metrics if necessary
  • Write in a shared document (between the two of you) what is discussed
  • If any action items are defined during those meetings, use a task tracking system to keep each other accountable
    • Always define a deadline on tasks to indicate when the task should be revisited
  • Avoid using one on one for status updates

01 May 2021

Answering coding interview questions

History / Edit / PDF / EPUB / BIB / 1 min read (~123 words)
Processes
  • Read the problem statement
  • Look at any provided unit tests
  • Add any unit tests you think might be relevant and not currently covered
  • Order the unit tests from easiest to hardest
  • Ask questions to confirm your understanding of the problem and verify edge cases
  • Think of a possible approach to solve the problem
  • Share your approach with the interviewers and get their buy-in
    • You may do this by writing pseudo-code as you are explaining your thought process
  • Implement your solution
    • If you encounter any difficulties, state them and explain what you are thinking
    • Make use of preconditions/early exit if possible to handle empty cases
  • Run your code frequently, even if it is not fully working
  • Discuss any follow-up questions with the interviewers
21 Apr 2021

Managing your reading

History / Edit / PDF / EPUB / BIB / 3 min read (~529 words)
General Processes

The process described below attempts to optimize reading quality books and enjoying the reading experience. As such, it promotes book exploration (discovery of new books) and reading books which have a high rating according to your own taste. Books which receive lower ratings (compared to other books) are moved down the reading priority list and will not be read until books that have higher priority (i.e., rating) either are finished reading or their rating decreases such that other books are now high priority.

  • Pick highly read books (use a site like goodreads to identify those books).
  • When reading a book, record the page you start and stop reading on, the time you start and stop reading and emit a rating for what you've read.
    • You can decide to optimize whether you want to optimize per page rating or per duration rating, that is, get the most value per page or by time spent reading.
  • Add new books to your reading list regularly. Those books are considered as having the highest priority and are then added to the prioritized list of books according to its rating.
  • When not reading a new book, read the books in order of priority and by interest at the time of reading.
  • From time to time you may look at your list of prioritized books and decide whether the books with the lowest priority should ever be finished. In some cases it is reasonable to decide that certain books will never be read completely.
  • As an alternative approach, one can use multi-armed bandits algorithms to decide which book to read next. Given that we can convert multi-armed bandits problem into the problem of selecting which book to read next given a sequence of readings and associated rating ("rewards"), the various algorithms (such as Epsilon-greedy or UCB1) will provide us with the next book we should read.
    • Interestingly enough, an algorithm like UCB1 will promote reading books we've never read first over reading books we've already started reading.

  1. Read partially any book for which you haven't given any rating
  2. Rate what you have read on a 1 to 5 scale, 1 being very bad and 5 being very good (see book rating)
  3. Compute the weighted rating of the book (the sum of rating times # of page associated to the reading divided by the total # of pages read so far for the book)
  4. Sort books by weighted rating (descending), then average estimated amount of time left to complete (ascending)
  5. If you have books that you haven't read yet, go back to the first step. If not, then pick the book at the top of the list computed in the previous step, then continue from step 2

  • It's better to read a good book than to finish a bad book

I make extensive use of Laravel Debugbar to track performance of parts of my application. I sprinkle calls to Debugbar::startMeasure and Debugbar::stopMeasure to track the duration of certain segments of my code. However, when this code goes into production, this dependency isn't present. This cause the code to break since it cannot find Debugbar anymore.

To solve this issue, I thought I would create a dummy Debugbar class and have it added as an alias, so that any code depending on Debugbar would still work, but end up as a "no operation". I found the article Dynamic class aliases in package which introduced the necessary piece of information to accomplish this.

<?php

use Illuminate\Foundation\AliasLoader;
use My\SuperPackage\FooBar;

class ServiceProvider extends \Illuminate\Support\ServiceProvider
{
    public function register()
    {
        $this->app->booting(function() {
            $loader = AliasLoader::getInstance();
            $loader->alias('FooBar', FooBar::class);
        });
    }
}

In my desired use case, I simply implemented the following changes:

In app/Providers/DebugbarServiceProvider.php (a new file)

<?php

namespace App\Providers;

use Illuminate\Foundation\AliasLoader;
use Illuminate\Support\ServiceProvider;

class DebugbarServiceProvider extends ServiceProvider
{
    public function register()
    {
        if (!class_exists('Debugbar')) {
            $loader = AliasLoader::getInstance();
            $loader->alias('Debugbar', NullDebugbar::class);
        }
    }
}

class NullDebugbar
{
    public static function __callStatic(string $name, array $arguments)
    {
        // Do nothing
    }
}

In app/config/app.php

    // under the 'providers' key, add
    'providers' => [
        [...]
        // This will take care of loading the service provider defined above
        App\Providers\DebugbarServiceProvider::class,
    ],

With those two changes, it is now possible to make use of Debugbar in most places and have it work even without the Laravel Debugbar dependency installed.