What is an API?

I was having a beer with a couple friends last night and was asked the question I almost always get asked when I tell people I work on an API: what the hell is an API? So I thought I’d try to explain it here, in the simplest possible terms. I’m also writing something about why APIs matter and what they can teach us about ways of working, so this will be a useful reference point.

What is an API?

An API is a set of rules that dictate how two pieces of software should communicate with each other. Those rules are written in code and documented in text so that they can be easily understood. API stands for Application Programming Interface, but you don’t really need to know that because everyone just calls it an API.

APIs standardise software interactions so that we don’t need to come up with new rules every time we link two software programs together. A good parallel would be the English language. Two people who have never met but who follow the rules of English grammar and syntax are able to communicate without needing to agree the rules of communication every time they meet someone new.

Give me an example.

My favourite example is flight booking. Imagine your go-to flight booking site (Google Flights 👊). Every airline has an API that allows booking sites to retrieve flight information from it. When I use Google Flights, I give it my origin, destination and dates. Google Flights takes that information and sends a request to all of the airline APIs (British Airways, EasyJet, Wizz, etc) that it has access to. The airline APIs send a response to Google Flights that contains the flight times, stopovers, price, etc. And Google Flights then displays that information to me.

What’s happening behind the scenes?

Without getting too far into the weeds, an API will offer a bunch of actions that Google or other travel providers can take by using the API. These actions are exposed as ‘endpoints’. An endpoint is basically a URL, which looks like this:

https://api.ba.com/selling-distribution/AirShopping/17.2/V1

Each endpoint allows me to take a specific action, which is often called a ‘resource’. If I look at the British Airways API, some of the actions I can take (resources I am offered) are flight availability, seat availability, pre-purchased luggage offers, and flight booking. AirShopping, in the URL above, is the flight availability endpoint.

APIs work on a request/response basis, like any good conversation. At a very high level, I can make a request to an API to retrieve data (as in the example above), to create new data (if you went on to book the flight), or to update or delete existing data (if you changed or cancelled your flight). The API will respond either by providing the data I have requested, by confirming that I have successfully created/updated/deleted data, or by telling me I’ve done something wrong.

If I wanted to use the AirShopping endpoint above, I would pass a bunch of data to that endpoint (origin, destination, date, etc), and I would receive a response with a bunch of data in return (available flight times, cost, stopovers, etc).

Why do APIs matter?

APIs matter because they make it easier for teams and companies to work together. When Uber launched in 2009, it didn’t have to build its own maps or spend months working with Google to integrate Maps into the Uber app. Uber’s engineers simply looked at the Google Maps API documentation and figured out how best to use Google Maps given the rules that the documentation provided them with.

APIs allow companies like Uber to launch and scale quickly as they can consume services that they don’t need to build themselves (maps, payment, SMS, etc). At the same time APIs also allow the companies who provide those services (Google, Stripe, Twilio) to offer them to what is effectively the entire world.

APIs are also a fundamental building block of the internet, and have played a major part in allowing the web to scale to the degree that it has. When you enter a URL into your browser, your browser is making an API call on your behalf. It requests the data that sits at that address (the URL) on a server somewhere, receives the data in a response from the appropriate server, and then follows the instructions in the data that describe how the page should be constructed. APIs are behind everything.

That’s it?

That’s it. There are some definite best practices in building and using APIs, and there is not a single API standard that is universally agreed on. But for the purposes of ‘what the hell is an API?’, that’s all you really need to know!

Systems and 10x engineers

The 10x engineer debate has surfaced again, to the delight of the internet.

10x engineers are fun to talk about (and do exist, for what it's worth!). But there is a lot of nuance missing here, particularly concerning where that engineer's product is in its lifecycle.

A product that is very early in its lifecycle has the characteristics of a multiplicative system. Multiplicative systems have a special property, which is that if any part of the system is a zero, the result of the entire system is a zero (eg: 12 * 5 * 0 * 10 = 0). For an early stage product, this means that if your architecture or product-market fit or distribution (etc.) aren't right, your product is sunk.

A product that is in a more mature stage of it's lifecycle starts to take on some of the characteristics of an additive system. Additive systems can have zeros in them without summing to zero (eg: 12 + 5 + 0 + 10 = 27). Yes, you could still have a catastrophic event that takes the product to zero, so it's not a purely additive system, but the failure of single features or strategies don't immediately sink the ship.

This matters because the type of system impacts the calculation of an engineer's impact. In a mature product you have 10x engineers that are just way more productive and creative than other engineers, and help the product to make step function changes. But in a new product, you have a totally different dynamic, which is that engineers can make the difference between a product living or dying. I have worked with engineers that again and again can turn a 0 in to a 1 (or even a 0.1) and keep the product viable. That impact actually can't be calculated:

(1 - 0) / 0 = 🤯

This is a pretty simplistic model for thinking about engineering impact, but suffice to say the 10x meme is selling some engineers short.

Crypto + remittances

I get asked pretty frequently about when the inevitable crypto domination of remittances will happen. I don't see it being inevitable, at all, so in case I'm right I thought I'd share it here so I can be right in perpetuity :). And if I'm wrong, it will hopefully be useful to look back at why.

At its core, a remittance is the transfer of ownership of a defined amount of money from one person to another. There are a lot of ways to do this: I can hand the money to a person that gets on a plane or a bus and delivers it to you; I can go online and tell some intermediary that I want to send money to you and have them do it on my behalf; I can simply think that I would like you to have $10 and it is incepted into your pocket (still working on this, not much progress since the Willy Wonka and Mike Teavee incident).

If we think of crypto (or some crypto) as money, then crypto is pretty good at this core function of transfer of ownership. Assuming I have some crypto, and you have a crypto wallet, I can pretty quickly and cheaply transfer ownership of that crypto from me to you. Plus, Elon might even solve my inception problem.

However! Transferring ownership is not where most of the cost of a remittance comes from, nor most of the value. This is a key misunderstanding of the "crypto is coming for remittances" line of thought.

The cost of remittances are high because:

1. It's expensive to take payment from the person who is sending money, especially by debit/credit card

2. It's expensive to meet local know your customer/anti-money laundering regulations

3. It's expensive to exchange one currency for another, particularly in parts of Africa, Asia and Central/South America

4. It's expensive to deliver money to the preferred receiving method of the recipient (methods of receiving are not generally fungible, I'm talking here about bank accounts, mobile money, cash)

On a remittance of £300, these costs in aggregate might be anywhere from £3-£10.

There are some clear ways that crypto could address each of these:

1. Create an inexpensive way to pay for crypto (circumvent the card companies), or, eventually, salary payments, disbursements, etc are paid in crypto and no payment from fiat is required

2. Centralize/digitize/tokenize identity (Tim Berners-Lee is working on this)

3. We all agree to use the same cryptocurrency, which becomes completely fungible for any purpose, or, short of that, a crypto FX market is created that trades at effectively zero spread

4. The crypto ecosystem evolves to the point where the crypto that you receive can be used for any purpose: buying things, paying other people, making investments, saving, etc.

Some of these are more likely to be solved for than others. Someone could significantly undercut the card providers to provide a cheaper way of paying for crypto, though this is getting less likely in at least some markets where interchange is capped (eg. the EU). The card companies also have a great moat. Centralized, digitized identity is likely to happen in some way (or at least I hope). Right now it feels extremely unlikely that a very fragmented market for crypto will centralize around one token (it probably shouldn't), and I don't see where a zero-spread crypto FX market will come from. I do think it's highly likely we'll see crypto become a better means of exchange, unit of account and store of value going forward.

If I could pick one reason that crypto won't come to dominate remittances it would be that we have a bunch of "good enough" solutions right now that keep the market extremely fragmented (I see a corollary here in why Facebook wasn't able to create a WeChat-style super app; we already had enough "good enough" solutions from prior technology, whereas China did not). By "good enough" I mean that they get the job done, not that the cost or other elements are necessarily acceptable.

I do not expect that we'll see the cost of remittances continue to stagnate. There is too much competition and too many parties with very large, software enabled networks that are interested. But my default is that it won't be crypto that tips it, at least until some of these challenges are addressed.

Aggression

I’m just finishing up the book Hard Drive, which is about Bill Gates and the early days of building the Microsoft empire. Gates is now one of our most celebrated business figures, and has a folksy sort of aura around him, thanks to his foundation and other endeavors like Gates Notes.

The problem with that picture is that Gates was ridiculously, almost comically aggressive as CEO at Microsoft. Microsoft used its position as the OS of choice to eviscerate applications from other companies (at one point the internal tagline for MS-DOS was apparently "DOS isn’t done until Lotus won’t run"). It dangled partnerships to get a look at competitors' tech, and then dropped the offer and replicated the tech. It used its huge amount of leverage from owning both the most popular OS and the most popular languages to force other companies (including Apple) to bend to its will.

This is all very relevant to the university admissions scandal currently rumbling Wall Street and Hollywood. In those worlds, as in the rest of the business universe, aggression is expected and rewarded. McGlashan et al undoubtedly knew what they were doing was wrong. But in the context of "everyone else is doing it" and the college “legacy” system that rewards alumni fealty, they probably felt that an aggressive move would again be rewarded, as it had been for their entire careers. Once turned on, aggression isn’t so easy to turn off. And this is not new:

The story is the same in tech. We always hear about how employees at Facebook are ‘well intentioned people’ but went a bit too far, were a bit too zealous. But aggression gets rewarded, so what do you expect? When the target is MAUs or DAUs, that’s what you get. I don’t know anyone who would argue that Facebook’s product has improved on the individual level, or that they’re more engaged with it than they were five years ago (let’s exclude Instagram here). But in aggregate terms it looked like more DAUs which looked like more ad dollars which is an Unassailable Good.

The difficulty with aggression is that it needs to be buttressed against a moral/ethical standard that puts aggression in its place. And that standard can't be static, it has to move in pace with the times in which it exists.

My first job out of university was in the commodities trading markets. Aggression there looked like ‘acquiring flow’, aka providing extremely tight spreads to your customers so that people want to trade with you. When a lot of people are trading with you, you have a pretty good sense of what's happening in the market. And when you know what’s happening in the market you can trade on it (or ahead of it). In the 80’s, 90’s and early 00’s that type of aggression was rewarded extremely well. Then the LIBOR scandal broke, and that aggression was rewarded with DOJ investigations and, for the worst offenders, jail time.

Aggression will continue to be rewarded as it always has. But applied without limit, the rewards to aggression will be harsh.