I'm working my way through Bob Walsh's book, Micro-ISV.

The first chapter discusses an important topic that I've already blogged about ... how do you come up with an idea in the first place?  This chapter outlines three basic approaches.  First, there's the systematic approach, where you identify industries you're interested in, and keep drilling into it in finer grained research until you have a specific niche idea.  Second, is the "Joel" approach (hey, that's the third blog now that involves Joel Spolsky).  When asked how he got started, Joel said they had three product ideas, and FogBugz was the one closest to being ready to ship, so that's what they focused on.  And third, is the "dammit" approach.

In the "dammit" approach you look for a significant source of frustration in your daily work.  I've been working on my new product, MergeMagician, for a little over a month now.  Of these three approaches, I think the "dammit" approach best describes how I decided on MergeMagician.  I've been developing software configuration management tools for a long time now, first in defect tracking, then build management, so I understand this area well, and there are a lot of opportunities for "dammit" discovery within this domain. 

The MergeMagician "dammit" is that branching and continuous integration don't work well together.  I want to be able to work on features independently of each other, so then when one feature is done, I can ship the product with the new feature.  I never know how long each feature will take, so it would be nice to work on features independently of each other.  To do this, I want to create a branch for each feature under development.  However, I'd also like to do a continuous integration build of *all* features currently under development, so I can work out the kinks of putting them all together.  But there's no automated way of doing that.  With MergeMagician, I can configure the server to automatically merge the separate features into a single integration branch, where I can build and test everything together.  There are lots of use case scenarios other than that one, but this use case is the basis of the MergeMagician concept.

I've gotten quite a bit of interest in MergeMagician so far.  Even though the product is still under development, I want to stay in touch with these early prospects to keep them interested.  To do this, I'm pioneering something I'll call "demo-driven development".  It's very agile-esque, so it's not an entirely new concept.  I'm sure someone else has thought of it and given it a different name.  The idea is to write the minimum amount of code necessary to showcase an idea, feature, or concept of the product.  When it's working, you shoot a narrated screencast of how it works, and then send it to your prospects.  Because you're only sending out a screencast, rather than the real software, the feature doesn't have to be fully implemented.  It just has to be good enough to demo it within a screencast.  When that's done, you go onto the next thing, and repeat the process until the product is shippable.

The advantages of this approach are 1) I get feedback along the way, 2) Keep prospects interested in the product, and 3) Build product mindshare before it's shippable.

We'll see how effective this approach turns out to be.  It seems to be working so far.  The first screencast I did wasn't even the real product, it was just a wireframe mockup of the product.  I sent this to a few people, and they in turn forwarded it on to some people.  Next thing I knew, three people from Microsoft contacted me and wanted to know more about the product!  I'm almost ready to shoot my next screencast, and this time, it will be real, running code. 

I was at the bookstore today and I was flipping through Joel On Software when I came across the chapter on hiring.  As is typical of other business books, it promoted a "hire the best" philosophy.  I've pondered this notion over the years, and wanted to share my thoughts.

Let me first say that I agree with the underlying philosophy of the advice, that being that a small number of highly-motivated, talented people can outperform an army of mediocre individuals.  However, there are several aspects of the "hire the best" mantra that have always bothered me.

First, it has unfortunately become a tired old management cliche.  Of course everybody says they hire the best.  Who would ever boast about hiring the worst?  And who would ever want to work for a company that brags about hiring the worst?  But saying you hire the best, and actually hiring the best are two very different things.

Second, it's a bit absolutist.  In my mind, if there are 17 million programmers in the world, then "hire the best" means 1 person has a job and the other 16,999,999 people are unemployed.  That's obviously impractical, making the advice at the most general level non-actionable.  I hate non-actionable advice.

Third, there really is no such thing as "the best".  There are a variety of factors that make up a good associate - raw intelligence, communication skills, people skills, work ethic, etc.  Every potential employee will have a unique mix of these aspects ... no one is perfect at everything.

Fourth, what does it say about me?  Am I the best?  I do the very best I can to "sharpen the saw" of my personal skill set.  I read programming books and magazines, self-teach myself new technologies, read blogs, and attend all manner of workshops, user groups, and code camps.  I feel like I'm at the top of my game, and the best programmer that I've ever been.  But am I the absolute best?  In a world filled with so much talent, it would be very conceited of me to make such a claim.

Lastly, what do you do if you live in a small city like I do that isn't the epicenter of the technological world?  Should I move to San Jose, CA or Redmond, WA, where the per-capita talent may be higher than where I live?  And how can I compete as a small software company if all the other larger software companies have already gobbled up the best talent?

So here's my take on it.  I don't think you need to move to California or Washington.  I think there are plenty of smart people all around ... you just have to find them.  I think you should hiring the smartest people you can find, invest in them to help them be the best they can be, and encourage them to fanatically acquire domain knowledge.

The latter two points above seem to be glossed over too quickly by "hire the best" aficionados.  Yes, raw talent is a very important aspect, but I think great programmers are born over time, not stamped out at a factory.  And I want people who care intensely about what they're doing and want to learn everything they possibly can about their domain.  There are just too many people in the world that want to punch a clock and pick up a paycheck.  Give me people with passion.

Which brings me to my closing thought ... leadership.  You can have the best people in the world, but without great leadership they won't accomplish much.  And in a small software company, leadership pretty much comes down to one thing ... YOU.

After Software Edge was sold, I worked for INTERSOLV for about a year, and started my second company, TeamShare, in early 1996.  This was in the very early stages of the Internet taking off, so like a lot of people, I was thinking about how to do something useful in the new era.

I went back to my roots with defect tracking, which by now I had built up nearly four years of domain expertise in.  Being shy and introverted, I didn't do any customer interviews at all.  I simply drew upon the customer experience I already had from building DCS.

Based on that experience, I thought a browser-based interface would solve three problems.  First, it would be easier for distributed teams to work together, and I knew from DCS that there were lots of customers that had distributed, or extended teams.  Second, it would eliminate client installs, which caused a lot of problems with DCS when we did upgrades from one version to the next.  Basically everyone had to do the upgrade at the same time, and people using an old client version couldn't talk to the new server.  It was hard for people to coordinate their upgrades.  Third, there was a technical problem with the way DCS handled its user inbox, which was used for defect notifications.  If someone hadn't logged in for a long time, we had to synchronize changes down to the desktop.  It was a slow process and made a lot of customers angry.  I figured a browser-based interface would eliminate the need for this desktop synchronization.

I wasn't sure if making it web-based would be enough of a differentiator from PVCS Tracker, so I looked for a couple of other features I could add to make TeamTrack better than PVCS Tracker.

DCS had a lot of built-in reports, which customers liked.  We also had the concept of projects, which allowed customers to delineate their development groups from each other.  The reports worked on a per-project basis.  This worked well at the tactical level, but as upper management started to use the system, they complained about not being able to see the big picture.  There was no way to run reports across multiple projects to get a consolidated, high-level view.  I remember Bob Pinna telling me, "Kevin, people want cross-project reports.  If you can just crack that nut, we'll have it made."

To deliver this feature, I decided to implement hierarchical projects.  I figured if projects could be arranged hierarchically, then reports could be run recursively across multiple projects, and customers would have what they'd been asking for.

The other need I decided to tackle was required fields.  What customers said to us was, "The reports you have in DCS is great, but they're useless to me.  No one ever fills in the fields they're supposed to, so we don't have good data to use in the reports."  So I thought a lot about how to have a good implementation of required fields.  I realized that it wasn't just about getting the required fields filled in, but they had to be filled in at the right time.  So I decided to build into TeamTrack the concept of workflow, consisting of states and transitions.  I could then associate required fields with each transition in the workflow, and customers would have what they asked for ... fields getting filled in with good data to facilitate proper reporting.  To my knowledge, TeamTrack was the first defect tracking system to implement workflow, a feature which went on to be copied by nearly all defect tracking systems that came after it.

So let's review what I did.  I didn't conduct any interviews, but I had the luxury of not doing that because I had all the experience from DCS fresh in my mind.  The ideas I had weren't born of my own fanciful wishes, but were based on real problems experienced by real customers.

But there's something else that I think is also interesting.  In DCS, no one asked for a web-based interface, they asked for simpler client upgrades and faster inbox refreshing.  No one asked for hierarchical projects, but instead they asked for high-level reports across multiple projects.  And they didn't ask for workflow, they asked for a way to make sure fields got filled in at the right time.  It was then up to me, using the domain expertise that I had by now acquired, to map the customer needs into specific features.  That's an interesting part of the process.  You can't expect customers to tell you what features to add, they just tell you what their problems are and then you have to figure out how to design a system that will solve those problems.

So once again I think we see a combination of the two schools of thought about idea creation that I mentioned in Part 1 of this blog series.  Customer feedback was an important part of the process.  But so was applying personal ideas, creativity, and domain expertise about the problem.

In the next entry in this blog series, I'll examine how applicable these lessons are in the current era of software development.

If you're a software startup, how do you decide what to work on?  That's a question that I've put a  lot of deep thought into over the years.

As I've talked to people about this issue, there are two basic paradigms that standout.  They are extremes of each other, but they serve as useful pivot points for the discussion.

The one school of thought, headed up by my sales & marketing friends, says you should "Listen to your customers and do what they say."  Well, that sounds reasonable, but what if you're a true startup company and you don't have any customers?  Then who do you talk to?  My friends tell me that in this situation, you work your personal network, you make individual contact with people, you take them out to lunch, ask them very general, open-ended questions like "So Pat, what part of your job do you spend the most time on?", and then you engage in "active listening" to discern their needs.  After doing enough of these personal interviews, a common pattern of need will emerge, and there you have it ... you've got your product idea.

According to this way of thinking, you should not even begin to write your first line of code until you have lined up at least half-a-dozen reference customers who have not only committed to buying the product once it is available, but have agreed to pay you in advance for the costs of the product development.

The second school of thought, headed up by technically-inclined people such as myself, tend to think you should draw on your own personal experiences and your domain expertise in whatever field you are contemplating, think of a great idea, write the code, and once the product is available, begin the sales & marketing aspects of the business.  People who think this way are introverts by nature, and would love nothing more than to hole up in a cave somewhere, spend a couple of years writing the greatest product on Earth, and then tenderly and lovingly release its awesomeness to the world.

So which of these schools of thought is correct?  Probably neither.  But which one is closest to being correct?  I think there are probably examples of both kinds of success.  As I am currently working on my third startup company, for the sake of anecdotal data, let's examine my first two companies, which were both successfully sold to larger software companies.  How did those companies succeed?

In 1991 I was working as a software engineer at Hewlett-Packard.  During my tenure at HP, my project team was struggling to cope with the myriad of "to do's", bugs, and enhancement requests on the project.  We began using a very primitive UNIX-based defect tracking system named DDTS.  It sucked for two main reasons.  First, it used UNIX shell scripts as its core technology, which made it difficult to get useful information out of the system.  Just getting a list of defects assigned to you meant you had to write your own shell script.  Each engineer had to have their own personal script with their name hard-coded into it.  Second, it could only be accessed from a UNIX workstation, which meant I had to leave my PC-based development system, walk over to a UNIX workstation, and login to access DDTS.  How inconvenient.

From this experience I was aware that there was a need for defect tracking, and the current solution in place was very suboptimal.  I reckoned that the world could use something better.  While those thoughts were fermenting, my friend Bob Pinna told me he was going to leave HP and start his own company to make a defect tracking system.  Wow!  I'd been thinking about the same thing.  Bob had a few more years of experience than I did, and had done not just software engineering, but software project management as well.  I figured if Bob thought it was a good idea, then it probably was.  Bob drew on his personal  project management experiences, but also talked to several other project managers, mostly at HP, to figure out the initial feature set of the project.  Bob left HP and started working on the product full-time.  I stayed at HP and worked with Bob part-time to write the system documentation and help with testing.

When the product was ready, in October of 1992, and before we had made our first sale, I left HP to work with Bob full-time.  I had started at HP on August 1, 1990.  My last day was October 2, 1992.  My life as a salaried engineer working for a large company had lasted exactly two years, two months, and two days.  My life as a software entrepreneur had begun.  Our product, Defect Control System, or DCS, was released and sales grew rapidly.  Two years later the company was acquired by INTERSOLV and the product was renamed to PVCS Tracker.

Getting back to the idea generation process, in which two schools of thought does this model fit into?  To me it seems clear that some aspects of both methodologies were used.  Interviews were conducted and feedback was incorporated into the product.  The product was not developed in a vacuum.  However, the idea itself came from personal experiences of understanding the need for defect tracking, and understanding the weaknesses of the existing solution.  Interviews were done after the idea was decided upon, and for the purpose of honing the idea and deciding upon the exact feature set.  We showed the product to our friends who were still at HP, but there was no pre-commitment on the part of HP to buy the product once it was ready.

The process worked, although not without some misfires along the way.  One thing I remember is that because of the close ties we had with HP, the product had a Bugs per Thousand Hours report, which was an important metric used within HP.  Turns out, not too many organizations outside of HP were doing things that way, and the feature went mostly unused and was the source of a lot of customer confusion.  Other important features were missed, like the need to be able to create custom fields for tracking information specific to each organization.  Everyone did things a little differently.  The first version of DCS had custom fields, but you could only add a total of 8 custom fields, two text fields, and six drop-down selection fields.  This limitation turned out to be woefully inadequate, which we learned very quickly.  I don't think you'll ever get everything exactly right out of the gate, but hopefully you get pretty close.  I think we got a lot more things right than we got wrong.

In the next blog in this series, I'll talk about my second company, TeamShare, which made another defect tracking system, TeamTrack.

Well, it's tax season again, which means its time to get caught up on my financing backlog.  The number of transactions that I need to manage for my business is still small enough that it doesn't make much economic sense to hire a person to run the financing department, or even to bring in a part-time bookkeeper, so that means I get to do it myself.  What fun!  I think it's pretty much universally accepted that the thing small business owners hate doing the most is their books.  If anything should be delegated, it's that, right?  Maybe so, but I've found it informative to do it myself, at least right now.  When I do eventually hire someone, I'll hopefully be handing off something that's reasonably well organized rather than a shoebox full of receipts.

I use QuickBooks for my accounting system.  One of the big problems I have in my life right now is that I don't just have one computer.  I've got my primary desktop (a Mac Pro tower that I upgraded to last year), a laptop, several old secondary desktops, the OS/X partition on my Mac Pro, the BootCamp partition, several desktop-based virtual machines, a few physical servers, and some virtual machines that I've installed on the servers.  So a non-trivial question that comes up when using QuickBooks is "On which computer should I install QuickBooks?"

Up until this year I've been running QuickBooks 2003, and I've had it installed on an old AMD-based desktop system running Windows XP 64-bit.  I built this system myself about three years ago, but since I upgraded to the Mac Pro last year, the AMD machine got disconnected and pushed into the corner.  Except I never reinstalled QB 2003 on the Windows Vista 64 system that I run on the BootCamp partition of my Mac Pro.  Whew!!  So that means that just the simple act of logging into QuickBooks meant I either had to re-install QB on my BootCamp, or I had to dig out the AMD machine, hook it back up, and remote desktop into it.  I opted for the latter approach, since I couldn't find the QB 2003 disk and serial number, and even if I could have, I'm not sure I could have reactivated the software considering that it is such an old version.  Considering that I had to call Intuit for help three years ago when I installed QB on the AMD machine, I wasn't very hopefully that I could get it reinstalled this time.  I really hate software activation.

So I got the AMD machine running, remote desktoped, and got back into QuickBooks.  Fine.  But then I started thinking, well, maybe I should upgrade to QB 2009.  It's newer and nicer in some ways, it's got some features that I'd like to use more, such as better support for online banking and more third-party support, etc.  But I knew right away I didn't want a repeat of the "Which machine is it on?" fiasco that I just went through, so I came up with a new plan.  Now I know that QuickBooks has a hosted service that I could use instead of the desktop software, but I've always been nervous about putting such intimate financial details out on the Internet, and I didn't want to pay the monthly fee, so I decided to stick with the desktop software route.

I decided to go with the virtual appliance concept.  A software appliance is where you use dedicated software on a dedicated machine for a single, specific purpose.  Just like you use your refrigerator to keep food cold and your oven to bake stuff, you can create an information appliance to manage your finances.  A virtual appliance is the same thing, except that you run it on a virtual machine insteal of a physical machine.  So I created a virtual machine named VM-FINANCE that would function as an appliance for all financial-related activities, including not just the QuickBooks software and data file itself, but scanned receipts, etc.  I would also install the tax software on there as well.  I access the virtual machine by remote desktoping into it.

This means I won't install QuickBooks directly on my Mac Pro BootCamp, so if I upgrade to something else in a year, I can still access my financial appliance.  If I upgrade the server to a faster server, I can just migrate the virtual machine over to the new server (although I might run into some reactivation issues doing that).  Also, if I can get into it from my laptop, as long as I'm in the office.  I have to be connected to the business network.  I won't be able to access it on the road.  But for me, that's okay.  QuickBooks is not something I need to be able to get into while I'm on the road, and if I really had to, I could VPN into the home office and do it that way.  As long as I can remote desktop into the finance machine, I'm okay.

So far, this has proved to be a good approach.  The software does run a bit slower than I would like it to, owing to both the remote desktop software and the virtual machine emulation, but it's good enough for how often I use QuickBooks.  Hopefully now that I don't have to dig out the AMD machine each time, I will be more inclined to keep my finances up-to-date this year.