Using Agile Methodologies To Make Games?

simoniker writes "Using Agile methodologies for programming is a concept that's been around for a while now, but some firms are now applying the concept to video game development." There has been a lot of talk lately about what the 'next big thing' in development will be. Could this be it? Or is this just another step along the way? From the article: "Agile puts the emphasis on producing demonstrable iterations of a game almost immediately into production, creating prioritized vertical slices that iterate on the most critical elements and features. The method also puts great emphasis on the organization of teams and the relationships therein, as well as the cycles in which teams must plan and carry out their project objectives."

Comments

The Bane of My Existence

[eldavojohn]

When I was in college, I was 'taught' how to develop software.

And so started a cycle of hatred towards methodologies about how to be productive in application development.

To me, the wild wild web is still here. I still get my kicks from coding without set in stone documentation and I still hate schedules.

I've always had the view that every project was a wild animal. Worse yet, it's a wild animal in sheep's clothing. And you approach the sheep with a shepherd's hook. When the veil is lifted and you see man eating marsupial with alligator teeth and a scorpion's tail, you have no choice but to throw the shepherd's hook at them and give it all you've got.

And this is how I approach managing a development project. You remember prior projects and throw together a bag of tools that have worked before and then you set to taming the beast. If you tell yourself "Waterfall works every time" then you're just going to find yourself with a shepherd's hook facing a lion or a bear.

Instead, you learn to adapt to every situation and that's the important thing. The rules are few and loose. The customer has the power to destroy everything and you have to deal with it. The best development is done on the fly with just enough documentation to convey the big idea of what's going on and keep everyone on that page. Given real life schedules and timed deadlines, there is such a thing as too much documentation.

Agile development is better in that it allows you more play and doesn't inhibit spur of the moment innovation. I think some of my most demoralizing moments have been when I realized some great new possibility for a project only to have my manager tell me that so much documentation would have to change that "maybe we'll put that in next year's scope." I find this to be ridiculous.

What was happening was people were starting to assume that waterfall was the silver bullet for project management. "What kind of project is it?" "I don't care, we're using the waterfall." And the big problem is that the waterfall is only considered 'adaptable' if you're ok with reworking everything from step one. Is this really necessary for every project though?

Another new thing we have these days is a "framework" that fits a specific type of problem well. You can throw these together on the fly and have very little documentation because the framework provides a well known implementation strategy (see Spring's MVC).

It is my opinion that using an incremental deliverable approach with frequent customer meetings and executive power at any point in the project is the most successful strategy. The "rules" you have to adhere to are up to you and should be purely a case by case basis.

There has been a lot of talk lately about what the 'next big thing' in development will be.

Why do we constantly look for the "next big thing" when the "big thing" is simply experience?

Re:The Bane of My Existence

[mkw87]

when the "big thing" is simply experience

Tell that to my fiance!

Re:The Bane of My Existence

[Anonymous Brave Guy]

I, too, was taught about software engineering processes during my CompSci days. But in my case, the waterfall model was the textbook example of how not to do things...

Of course, you can take things too far that other way as well. Being too rough and ready -- something encouraged by an "agile" approach -- may get you best results locally, but what matters is getting optimal results globally across the lifetime of the project and all its features. Just as the old hands look at managers trying in vain to match reality to a waterfall model and smile, so they look at young enthusiasts diving in with little to no big picture planning and wait for things to fall apart.

Re:The Bane of My Existence

[cp.tar]

Why do we constantly look for the "next big thing" when the "big thing" is simply experience?

Because managers don't trust engineers.

They don't understand what's going on, and yet they have to manage it.
So they hear about a new methodology, drink it up like a common sucker drinks up Scientology, and turn it into a religion.

Everything you don't understand you fear, and then you turn it into a religion.
All too common behaviour, all throughout our history.

If they tried to understand it all instad, they probably wouldn't be managers; they'd be engineers.

Most people don't care about how things work; they only want them to work and to work always.

Magic-minded lot, all of them.

Re:The Bane of My Existence

[Jerf]

Why do we constantly look for the "next big thing" when the "big thing" is simply experience?

To know how important experience is, one of two things needs to be true: You either need to have it, or you need to be someone who is still interested in "wisdom" in this knowledge-centric era, and be willing to listen to those who have experience, and pick the correct people who say experience is important and believe them*.

Both are fairly rare.

The rest follows from "rarity" in the obvious manner.

(*: I believe wisdom is underrated nowadays, but I won't pretend that the problem of figuring out who to listen to is not itself hard.)

The other thing is that there are next big things, but I suspect we are slowly but surely running out of "next big things" that will be comprehensible to more than, say, 25% of practicing programmers. Switch to Ruby, learn how to really use it, and add a good automated testing philosophy in, and you've gone a long ways towards extracting all the additional productivity you can get out of your environment without a major upheaval. For instance, I think it's reasonably likely (although not certain) that functional programming is finally going to come out as a reasonable paradigm over the next 10 years as it will be best able to take advantage of multi-core systems without completely upheaving how you write programs, but we're going to lose a lot of programmers on that transition, because it just doesn't click for some people. And that's a major change.

Most aggravating

[neonprimetime]

because of the iterative process of Scrum and the short work cycles of Sprints, redirecting a project rarely results in large volumes of wasted work.

I like this ... cause that last pair of words ... wasted work ... seems to hit my teams especially hard quite often right now. We'll finish LARGE tasks, perhaps bundle multiple items together into one large release (cause this is how management wants it) ... and by the time it gets to UAT testing many development hours have been spent ... and then the UAT tester will turn around and say ... nope ... don't need/want that ... and thus all those development hours are wasted. I'd love to get a workflow like SCRUM implemented in my workplace.

The next big thing?

[Anonymous Brave Guy]

There has been a lot of talk lately about what the 'next big thing' in development will be. Could this be it?

I doubt it. Smart dev teams have been using iterative development cycles and keeping their code tidy since a long time before anyone coined the buzzphrase "agile" (or using SillyCapitalLetters and calling things "extreme", or any of the other hype we've put up with lately).

Buzzwords aplenty

[lawpoop]

"creating prioritized vertical slices that iterate on the most critical elements and features"

Can someone tell me what this means?

Re:Buzzwords aplenty

[neonprimetime]

For example ... in Tomb Raider
One team should concentrate on Lara's Face, one should concentrate on her chest, and one should concentrate on her Legs. No wait ... that's horizontal slices.

Re:Buzzwords aplenty

[rs232]

"Can someone tell me what this means?"

Some systems analyst draws some connectors and boxes and you get to sit up till 4:00 am writing the code.

Re:Buzzwords aplenty

[Salzorin]

I'd like to, but it's really not on my deliverables for this quarter. If we could interface sometime during the next iteration, I feel like we can drill down this action item further and really bring this to manifest. I look forward to addressing your critical concerns and reaching some new plateu in our correspondence.

Re:Buzzwords aplenty

[rwhansen]

Some methdologies develop each subsystem fully then integrate them in a big bang at the end. Agile practitioners develop whatever parts of each subsystem are needed to support the features in the current iteration. If you think of the subsystems as layers, the vertical slice is the part of each layer that is being developed in the current iteration. The features should be prioritized in order of importance, so each iteration is working on the next most important feature or features. The idea behind all this is to always have some sort of working application.

Re:Buzzwords aplenty

[magicjava]

"creating prioritized vertical slices that iterate on the most critical elements and features"

Can someone tell me what this means?

Sure. It means get something simple working as quickly as possible and improving it over and over.

prioritized - Pick your most important features and implement them first. For example, if you're writing a word processor the ability to save a file probably has a higher proiority than a spell checker.

vertical - This means "specific" as opposed to "general" (also called horizontal). So you work on specific features (like the saving of a file mentioned above) rather than more general ones (like optimiztion).

iterate - You're not going to do things all at once. As an example, perhaps your first iteration of saving a file only saves text and no formating information. You're second may add formatting, your third may add additional formats.

Re:Buzzwords aplenty

[glindsey]

"Hack something together that you can show the sales folks, and then keep adding to it with horrible kludges until you get something resembling the final product which complies. Ship immediately."

Re:Buzzwords aplenty

[sammy baby]

That was a fancy way of saying:

1. You start with an extremely barebones (read: you'd never release this into the wild) program that actually runs, passes all its unit tests, and implements maybe one important feature. (eg: a shopping cart program which will show you the contents of a catalog.)
2. With each iteration, you add a new feature, refactoring older code as necessary. Start with critical functionality and work your way down the requirements.
3. Continue until the product is releaseable.

The "vertical" bit means that rather than taking the "let's implement all the methods we'll need in the CatalogItem class first" approach (horizontal), you're trying to make it actually do something useful (eg, show a list of products) with each iteration.

Re:Buzzwords aplenty

[Jhan]

At my current gig (OnGame ie. PokerRoom.com) we've been using a kind of modified Scrum. The idea is to start "El Project Grande", spend a day breaking the project into parts and deciding what depends on what, then spend another day in groups breaking down the parts into parts until you have tiny pieces which ideally will take less than 16 man hours a piece.

You now have a big pile of tasks, sub-tasks, sub-sub-tasks and so forth with dependencies and time estimates (all set by the programmers themselves.) It's then up to the Scrum Master aka. "Get-those-assholes-off-our-back-person" to show the results to the customer and decide priorites.

The priorities are passed to programming team, and they then plan the first "iteration". Now, the goal of every "iteration", which typically spans max one month, is to produce a fully implemented, tested and launchable subset of the complete product.

Every day, all the coders have a *short* meeting (10-30 min). Each individual states what he is working on, what he's done since the last meeting, what he's planning to do until the next meeting and most importantly he's strongly encouraged to cry for help if he's blocked in any way.

In addition to this there's an evil, buggy Excel sheet with all the (sub-sub-sub)tasks along the vertical, and days along the horizontal. It's each programmers responsibility to update estimated hours left for tasks he has been working on each day. NB! *hours left* It's prefectly alright to up the time remaining if you have unexpected problems. So a certain task may go "16 12 4 56" hours left.

The main points of Scrum (for a programmer):

• The programmers set all time estimates.
• Individual programmers may change any estimates upward without hard feelings.
• (In our version) The programmers hire new programmers.
• The programmers allocate work among the programmers.
• In fact, the programmers control their own work 99%.
• The Scrum Master aka. ('Nearest-thing-we-have-to-a-boss') deflects all flack and handles all communication with harpies^W customers.)

...and that's how it's supposed to work in the best of worlds. In reality, however, the whole model crashes and burns if * The projects are to short (

Re:Buzzwords aplenty

[Kadin2048]

"Hack something together that you can show the sales folks, and then keep adding to it with horrible kludges until you get something resembling the final product which complies. Ship immediately."

I know you're being facetious, but that's exactly what I envisioned whenever I hear about these "agile" methodologies.

I guess it's because I work on software where the tolerance for bugs is a lot less than it is (apparently) in desktop PC software, but I have a hard time seeing one of these methodologies producing clean code.

If your application is going to get shipped and then never looked at again, I guess "clean code" might not matter, but if you're building an enterprise system, or writing a custom application that has to be maintained by other people later (people who probably aren't going to be quite so much of an 1337 h4x0r as you), it's a must.

Documentation and specifications aren't just something that you do in order to satisfy the PHBs of the world; they serve a real function. A system that's been well-specified to begin with is much easier to fix later, at least in my experience, than one that's been produced during a death march and is nothing but a giant rat's nest of code.

I could see the benefit of an "agile" development cycle for prototyping or producing a quick demo, but the idea of critical pieces of software -- software that your customers are going to depend on and that other people, months, years, or decades down the line are going to have to maintain and update -- being built this way scares me a little.

Listening to programmers bitch about documentation is like listening to a bunch of carpenters bitch about architectural plans. Of course you could get the house built a lot faster if you just took 15 builders and said "hey you, build a wall here!" and "you, build a bathroom over there!" ... but I wouldn't want to live in the resulting house. Would you?

Sounds like a formula for spaghetti-ware

[cculianu]

Agile development with C++? Sure, it can be done. But one of the nice features of that language is its ability to model the application domain using Classes, Objects, Inheritance. If you don't take the time to model the problem correctly, you can painfully pay for it later on. Sometimes you get faster development times if you actually take a day or a week and model your problem so that it captures the essense of what you are trying to solve using objects, etc.

An iterative approach where you use a greedy method of implementing each new feature as quickly as possible with little insight into the overall system may or may not produce optimal results. It can lead to lots of spaghetti code. Or maybe not. Certainly in my career that approach sometimes *has* led to cleverer solutions as I was forced to think up the best, simplest way to imlement them. But often it can lead to massive amounts of spaghetti code that doesn't make much sense or capture any insights into the problem.

Re:Sounds like a formula for spaghetti-ware

[Jerf]

Which leads to a lot of code churn and a system where noone really knows what a given subsystem is doing, unless you were the last one to touch it.

Two counterpoints:

• This is different than normal coding practice how?
• It's all about costs and benefits. Sure, there are the downsides you mention, but on the other hand, aggressive refactoring means you don't necessarily live with your first design forever and ever, amen, as (go figure!) your first design also happens to be your worst. So, is it worth it? Depends on your situation. But in my experience, between "code churn" and "first generation design enshrined in concrete forever", the former is very preferable on all but the shortest time scales. YMMV.

Re:Sounds like a formula for spaghetti-ware

[Anonymous Brave Guy]

But, one of the most important tenents of Agile methodologies is that you must have _strong_ teams that are highly skilled, and with great amounts of personal motivation.

...who, by definition, will produce good results using pretty much any methodology. The test of any business process is how much it helps the guys who aren't already good, not how much it helps the office guru.

Re:marketspeak

[darrenf]

creating prioritized vertical slices that iterate on the most critical elements and features

But can I synergize with my results-driven knowledge base while partnering with self-managed teams, increasing single-source responsibility while undergoing a complete paradigm shift?

I didn't think so.

Programming Methodologies Are Dangerous

[Anonymous Coward]

Working at a major game company, here's what happened when someone here raised XP:

The guy who suggested it, "It's an itterative development model. We identify core features, develop those features, refine those features with the customer, then add the next layer, repeating as we go. We gain much better code as every component meets the customer's need as it is developed, challenging the customer to think about it in context, and allowing us to add additional itterations if needed."

Management, "So, we can identify additional features throughout development?"

"Absolutely. You just have to assign additional resources (time/people) to account for those extra features. But, by identifying them as they come up, you end up with a much better system that really does everything right."

Over the next few months, features kept getting added, developers dutifully updated schedules. All was happy. Followed by...

Management, "This was supposed to ship before thanksgiving. It's now slated to ship in the new year. We'll entirely miss the holiday season."

Rapidly realizing his mistake for suggesting it guy, "Yes. But you kept coming up with new features. And they are great new features. Think how much better the product is for it."

"If we miss the holiday season market, we lose money. This has to ship 'on time'."

"But on time is a function of how much you add. We're developing everything to schedule. You've just increased the features so increased the schedule."

"The schedule can't move."

"So you'll have to lose some of the remaining itteration milestones. You'll have to drop features."

"But we like all the features we've come up with."

"But adding features adds time. You've known that since the beginning."

"We've known this has to ship for the holiday season and you promised us we could have extra features. You're just going to have to work more overtime. Fortunately you're overtime exempt so that won't cost us anything to get this project back on schedule."

"It is on schedule. You just changed the schedule by adding features that were identified along the way."

"You told us your wonderful "XP" model would let us do that. We gave you the chance to try this new method under the understanding we got these benefits."

"And you do."

"Good. Then make your schedule."

"We are mak-"

"No arguments. This discussion is over. You promised you could deliver the extra features. You're now behind schedule for the holiday season. You're just going to have to crunch. End of discussion."

Yeah, thanks XP.

Never, ever, raise exciting new methodologies to management. They will hear all of the advantages and expect every last one of them as though it was the perfect implementation of the method whilst completely failing to hear (and certainly refusing to act on or implement if they do hear) any of the trade-offs that have to be made to enable those gains.

Re:Programming Methodologies Are Dangerous

[hhr]

The developers also made a mistake in not realzing where their paychecks come from-- selling finished games.

If someone is paying you to write software then "Earning out paychecks" or "keeping on budget" must be a critical feature. We may like to think otherwise, but companies don't pay you because they love you. They pay you because you earn more money for them then you cost. If it was the other way around, your paycheck would bounce and/or layoffs would ensue.

Re:Programming Methodologies Are Dangerous

[finnif]

Your story there further proves... no matter WHAT the methodology is called, the old adage remains: "Quality, Cost, Schedule: Choose Two." Except in software, the other two can't always be fixed by spending more.

Mumbo Jumbo

[Anonymous Coward]

Sounds like more touchy feely mumbo jumbo to me. Coffee, red eyes, bad breath, late nights, divorce... that's how software gets done dammit..

'agile' is neat, but ...

[bunions]

as everyone else pointed out, it's mostly just a name for what people have been doing for a while when upper management leaves them alone.

That said, I've found the hardest part of the process to be finding a client who is willing to put up with the constant back-&-forth and interminable beta testing. Customers generally just want to tell you what they want, go away and then have you magically deduce what they actually need, and can be irritable when you tell them you really can't do that because ... you know ... agile!

Way too much market speak...

[Mysticalfruit]

Tragically, I couldn't RTFA because of th excessive market speak.

In order to energize the evolution of computer games we need to synergize on the vertical slies... WTF?

I guess people forgot to mention that writing computer games rates up there with writing operating systems in complexity...

But that's how most games were made...

[MagikSlinger]

When I was in the game industry, this was how we always did it, even EA does it to an extent. Why? Because the publishers want results. Usually every month or you won't get paid. So as a result we had incremental releases with more and more features added in, and as a bonus, sometimes working. X-D

P.S. With few exceptions, most game teams do NOT have specs or docs. If they're lucky, they have concept art and some clue what the game is supposed to be. I remember having to code up front-end screens and the artist & I had to figure out what they were supposed to do because the game designer was still writing the specs for the screen we finished last week.

30 day sprints?

[magicjava]

30 days is fast? I wish I had 30 whole days to do my projects!

I love iterative development, but scrum didn't impress me much when I was involved with it. Perhaps that was just due to the management that implemented it. We went from logging bugs and tasks in bugzilla to writing them on index cards (do I really need to point out to computer people how stupid that is?), and trying to plan projects on these obscure, poorly implemented planning tools.

Iterative development is great, but scrum strikes me as nothing more than another useless management fad.

Managers <sigh>

[Anonymous Brave Guy]

This is just an issue with management, though. A good manager will trust his engineers (or fire them and replace them with trustworthy alternatives). The manager's job is to set the direction of the project, get the engineers what they need to steer in that direction, and then get out of the way as much as possible and as quickly as possible.

It's staggering how many managers don't realise this, and hamstring their dev teams with their personal, half-baked, technically-incompetent ideas and/or with excessive procedures and beaucratic reporting because the manager "has to know what's going on". Of course he does, up to a point, but what exactly is he going to do if a developer does tell him that a bug fix was delayed by a day because {$TECHNOBABBLE}? If he's not going to act on some information, he doesn't need to know it, and requiring developers to take time out of their day to "keep the manager in the loop" more than necessary just disrupts development.

I've thought about this

[th1ckasabr1ck]

I program games and I've thought about this a lot. I kind of like the agile approach, and I think that a lot of the concepts could be useful. One of the major components of being "agile" is building unit tests and relying (at least somewhat) on those as a way to make sure that these rapid checkins are not breaking semi-related systems.

With games it is UNBELIEVABLY difficult to write unit tests that effectively catch problems. With animation, physics, AI, the built in randomness with the game, the human interaction, etc. it's unbelievably difficult to write unit tests that can get the job done.

It's easy to write a unit test that says, "When I shoot a dude with this gun, make sure his health goes down by 50". But it's an entirely different thing to say something like, "Make sure that when a battle is going on nobody gets caught up on geometry and can't path to their movement goal."

Sure it's possible to write small little unit tests that make sure a dude can get from point A to point B, and that this guy knows how to path around a dynamic object if a crate falls in his way, but this isn't something you can effectively break up into little parts. I've fixed too many bugs that are a strange and unfortunate combination of all the aforementioned systems to fall into the trap of thinking isolated unit tests are going to get the job done.

I am a big fan of the idea in theory, but I can't imagine it being implemented effectively in practice. I would love for somebody to prove me wrong!

Duke Nuken Forever utilizing Agile Development

[mrkitty]

What do you think I'm lying? :)

pushing envelopes

[drDugan]

games have pushed the envolope more than any other area in computer software development

the current rising tsunami happening right now with agile development for web applications (like Ruby on Rails, and similar approaches happening in Perl, Python, java and others)... will take games by storm, in my opinion. there are a few factors here that will make games really interesting:

there will be much more scripting in games,

there will be much more meta-programming (code writing code)

there willo be more layering and customization in the languages we use - with more powerful and expressive languages deep underneith, but with tightly constrained, layered (or mixed-in) frameworks that enforce best practices built on top. we have seen this progression going forward for the last 10 years or so, and it's starting to get really interesting now.

and on the next 5-10 year horizon, we will see applications (driving mostly by desire for better AI in games) that are human-competative in reasoning and interaction - I offer this conjecture without proof, but with significant anecdotal evidence to support the assertion.

I watching for signs of two important things in computer development: a code base that can write another codebase (metaprogram) that is (1) aware of what it just did and (2) can communicate in a complex novel way with the new code set (I'm being vague here, but you get the idea -- think compiler theory for AI), and ... a fully autonomous systems that can operate a *power plant*. Once those two things happen, humans are toast. I'd estimate that to be about 2060.

Take it from me ...

[LaughingCoder]

I have been developing products for a long time (decades if you must know). For what it's worth, it is my experience that the people on the team have by far the biggest impact on product quality, timeliness, and all those other goodness measures. I believe that the methodology is almost immaterial. Good engineers will instinctively use the appropriate process for the problem at hand. Now, this doesn't necessarily scale to very large projects, which is why I am a firm believer in loose coupling. As soon as practical, decompose the big project in to a collection of loosely coupled smaller projects and then put in place the teams (unfettered by process dogma) to develop the pieces. Ahh, you ask, but what process do you use to decompose the system? See my earlier comment - choose a small team of very good engineers and have them do it. Don't tell them how -- they already know that. Trust people, not process. Ironically there is one process that I believe is critical to every organization's success, and it probably the least-studied, least-optimized, least-formalized process every company has ... and that process is the interview process. Clearly if I am going to trust my people to do the right things and make the right choices, I had better hire the right people. Anyhow, that's about $0.93 more than my $0.02 so I'll step down off my soapbox now.

Oh, I get it

[Quiet_Desperation]

"Agile puts the emphasis on producing demonstrable iterations of a game almost immediately into production, creating prioritized vertical slices that iterate on the most critical elements and features. The method also puts great emphasis on the organization of teams and the relationships therein, as well as the cycles in which teams must plan and carry out their project objectives."

So the game is to figure out what bullshit like this means?

Product development

[Aceticon]

The problem using any software development methodology which requires user feedback for new product development is that you don't actually have any users - best you can get is a marketoid and those change their minds every 5 minutes.

You see, the closest you have for a user of a new game is a gamer and those can't really help you refine your requirements 'cause all they want from a game is to be entertained and they don't really know beforehand how a new, entertaining, game will look like - "having fun" is hardly an easy to define business process.

Maybe some sort of mixed approach where you have a game designer with an overall view of the game concept and a generic pool of gamers to check out the "fun factor" during game development. Might work well for games with an "exploration" component (for example RPGs) for which you can design the early levels, "test" them with some gamers and then use the result to fine tune those levels and later ones. Still, i doubt it can be usefull in game genres such as RTS and Sims.

More in general, and judging from the posts i've been seing in ./ for the last couple of years, the main process problem with game development seems to be that a lot of the control over the final product lies with Marketing, and worse, Marketing is in a different company altogether (the producer) than game development - so even with good managers on the software development side (already unlikelly) it's difficult to control the the flow of wacky ideas "that just have to be included in the game" coming from the marketoids - thus requirements creep is rife, which almost guarantees that long hours and death marches are standard.

Re:Managers

[SatanicPuppy]

I know it's tired old hat, and probably flameworthy in this environment, but a real manager doesn't need to know what the hell you're talking about to make a good descision. All he has to do is know his people, and be willing to listen to their experience. By the same token, the most knowledgable manager in the world can still screw everything up by trying to make everyone do it the exact way he would do it if he was doing it all himself.

You need to find someone who can keep the final goal in sight, and who is flexible enough to reorganize whenever the requirements change. Agile, Waterfall, Iterative, whatever, it doesn't matter...These are ideas put together so that mediocre managers will have some kind of method that may bring decent results. They can all work great, and they can all work poorly, and it all depends on who is doing the oversight.

Management actually is a pretty solid skill if you can do it. Too much of the flaming comes from people who've never had the good fortune to work with a good manager. I myself have never worked with one who was the total package...Either they understood the work and the clients and they couldn't deal with the higher ups, or they dealt well with the higher ups but didn't understand the work or the clients.

This 'agile' thing has a different goal

[Weaselmancer]

From the blurb:

Agile puts the emphasis on producing demonstrable iterations of a game almost immediately into production

So what this does basically is get something that just barely works up for review as quickly as possible. Like throwing a lump of clay on a table and saying, "There's a vase in here, somewhere."

This IMHO will do two things. First, it will give SW managers a warm feeling caused primarily from too much optimism. "All the engineers have to do is shape that clay a bit, and it's a vase! We're ahead of schedule!" Two, since they will think they're ahead of schedule, they'll report to their superiors about how they already "have a working prototype of a vase" and that'll bump up the schedule.

The engineers who actually have to implement things will know better. And they're the ones who will get stuck with the deadline. The agile pony show where you show your manager something that boots but doesn't have 98% of the functionality in it will bite you in the rear later on.

This method doesn't seem well suited to making software. However, it does seem well suited to making managers feel good. I'd avoid it.

Solve specific bugs

[Prien715]

We're in the process of implementing further unit tests under the agile model (though not gaming related, it is a heavily interactive app). The idea behind unit testing is to:

1) find bug
2) write unit test for bug.
3) write fix.
4) verify unit test/program both work. If unit test works, but program fails, go to step 2.

1) In your example, a tester/coder would find see some soldier dude getting stuck in the middle of battle.
2) So do it again, saving the game before the bug occurs. Write a unit test which loads this save and then check after x seconds to see where the soldier is and then fail the unit test cause he's stuck behind the crate. The unit test will is not in any way general. It's not meant to test all your pathfinding issues, just this one.
3) Rework the pathfinding code.
4) After you notice the unit test passes, go in and see if the bug is still physically in the game. If it's not, you've fixed it. If it still is, well, write another test. You may end up with a dozen pathfinding unit tests this way all of which "fix" some sort of bug.

The idea behind this methodology is that let's say your new pathfinding algo causes a bug which developer B fixes, but which causes the original case to break. Your unit test will now "clue you in" before a checkin occurs. It also free testing to do more "battle" and "feel" stuff.