The slides are probably a bit too terse to be useful so here’s some notes. Feel free to get in touch if you’ve got questions or feedback:
I’ve been dealing with software & systems for about 20 years now, and this presentation goes for 90 minutes, so obviously we’re not going to cover everything in detail.
This talk isn’t really about how to write software, it’s about how not to burn through all your money on software development without achieving anything.
I’ve split it into three parts:
Computers (these days) are very fast:
So there are many problems you can easily solve in software.
All attendees but one were indeed interested in doing something software centric …
But computers are also very dumb. They’re still quite lousy at:
As a startup, you can’t really afford to be researching interesting challenging problems in computer science, so if your idea needs any of these things to be useful you’re probably better off leaving it to Google.
I mentioned this unfortunate kerfuffle re: facial recognition.
These days, infrastructure is so easy to rent or buy that you can built anything you want, from a single CPU to hundreds of CPUs, from $5/month up until your money runs out. But as a startup, you don’t want to spend more than you need to.
How do you know what to actually spend? Look at your business plan. You should have a milestone your aiming for … maybe say 10,000 users. Work out what you need for that milestone. 10,000 users who use your service once a day is maybe 1 request per second. You could probably get away with running on a free AWS micro instance.
Likewise, if your application doesn’t need high reliability, don’t bother building it. It’s a lot of fun reading up on how Netflix do it but for startups its often not necessary and you probably can’t afford it.
SOURCE CODE IS NOT AN ASSET.
Kurt talked about how your intellectual property is an asset and how even your losses are an asset (to a tax accountant, anyway …) but it is important to realize that your software itself is not an asset.
Measuring programming progress by lines of code is like measuring aircraft building progress by weight.
– Bill Gates
Software is something you have to build to make your business work, but it isn’t the point of your business, just a tool. The less of it you have to write the better. Use existing components whenever you can.
If your planned exit is by selling your business, your buyers will be interested in your customer base, your brand and maybe your algorithms. They’ll rewrite your actual software to work their way anyway.
(Does anyone think Facebook couldn’t have replicated the software of WhatsApp for a whole lot less than they paid for it?)
The key to using existing components is to divide and conquer. Split the problem up into parts and choose a good tool for each part.
The main division here is about trust. You can never trust a device you don’t control, so keep your business rules in the back-end and just use the front-end for presentation.
This is actually a very old architectural model called Tiered Architecture which splits into Presentation, Logic and Data. These correspond to a front-end, and back-end and a database. The presentation layer is all about showing the data to the user, the logic layer about enforcing business rules and the data layer about enforcing consistency and not forgetting stuff.
There’s also interesting things like Message Queues to consider, which have actually been around a while but are suddenly trendy again. They’re a way of getting around people’s expectations that everything will be instantaneous on the web. You respond to requests immediately but queue the actual work up to be done more slowly in the background.
There’s also external services such as email and credit-card gateways to consider. Some of these are very good and some are rather painful to deal with.
we mentioned some details here but that can wait for another blog post I think …
There’s a range of options:
As you go down the list, the results get more sophisticated, more capable and more flexible, but also more expensive to develop. There’s an upgrade path there where you can start out with something very simple and improve it as you go along.
I’ve tried using jQueryMobile and such to make HTML5 apps which look like native, but people’s expectations of native apps are so high that they always thing the HTML5 imitations look sluggish compared to a native app. Take the same app and make it look like a website, it seems really quick compared to a website.
Some of my favourite Web programming / HTML5 resources:
There’s also native apps for various platforms:
The nice thing about apps is they get sold in an app store, so you don’t have to set up your own payment infrastructure. On the other hand, the app store takes a pretty big percentage.
I once had a conversation with a client which went something like this:
Client: “Can you write us an iPhone app?”
Me: “Sure, what do you want it to do?”
Client: “I want it to get our logo on our customer’s home screen”.
Me: “But it’s got to do something, right?
Client: “Of course, or they wouldn’t put it on their home screen”.
The app wasn’t there to solve a technical requirement, but a marketing one. Users can install your HTML5 site on their home screen but it is still a different path that the app store one and users may find it confusing.
You could also consider mixed approaches where a relatively simple native app does hardware interfacing, and the passes off to a more visually exciting HTML5 site for display back to the user.
The word “agile” has been subverted to the point where it is effectively meaningless.
– Dave Thomas, co-author of the Agile Manifesto
Everyone seemed to have heard of Agile, which was nice. Mostly I talked about breaking the work up into Sprints and Phases and thus not getting stuck in a Death March or for that matter a series of Death Sprints.
Mostly I want to emphasize here that this is mostly a matter of organizational culture, not a set of rules you can follow.
I also mentioned:
I’ve just finished a project which used Trello to coordinate tasks between five team members in five different locations, three in Australia and two in the UK! I’m totally blown away by just how well it worked.
Nobody particularly likes writing documentation and it seems like the kind of thing a startup can’t afford to do. But actually if done right it is a great way to clarify your ideas before development starts.
If you sit down and write:
Q: How do I log in? A: You can log in with a username and password, or via Facebook.
… well, you just wrote a testable requirement without even thinking about it. Likewise, if you write:
Q: How do I pay if I don’t have a credit card? A: At the moment we can only accept automatic payments by credit card.
… you just documented a limitation that you can come back to at the end of your first development phase.
There are also other kinds of documents to consider
… these don’t have to be a formal process, but work out some way to keep them up to date and track changes, as they provide a valuable record of how your ideas are progressing.
There’s a logic riddle which goes something like this:
Q: Why do cars have brakes? A: So you can drive fast without crashing.
Why write tests? So you can iterate fast without breaking your production servers!
We kind of rushed through this bit because time was running out and the group didn’t seem to be at this phase just yet. I’ll have to have a think about how to write a better talk and/or blog post about these decisions at some point.
You can do anything in any language. You can also eat soup with a knife.
– Ville Vainio.
There are many different languages out there, and they each have good points and bad. But there’s a reason this question is way down here at the end of the slides.
If you were starting a delivery business, you wouldn’t decide, on day 1, what brand of cars to buy. You’d have a look around at several brands, get some idea of which brands supplied which kinds of vehicles at the prices you needed, then shelve that information untl you were ready to actually buy.
Likewise, the decision about what platforms, languages and frameworks to go with can be put off until after you’ve nailed a lot of other specifications down. As a startup a large input may be who is actually available to hire.