Saturday, December 27, 2008

The Ad-Hoc Approach Can Work, Sometimes

Joel Spolsky's introduction to his article How Hard Could It Be?: The Unproven Path at states that he has broken seven rules about creating a technology venture. All the rules (except maybe the last one) also pertain to executing software development projects. If I were to draw a conclusion from this article it is this: you can, sometimes, create useful quality software in a reasonable time frame without following established approaches --- i.e. using an ad-hoc approach can work. The article gives one example where this approach succeeded. A few of my own projects and some of my colleagues have also succeeded in spite of breaking most of these rules. I'm welling to bet a great percentage of the single person or small team open source projects have also been successful using the same approach.

Don't get me wrong, I'm not advocating the ad-hoc approach for software development projects. In fact, even suggesting so leaves a bad taste in my mouth and gives me a sick belly. But I have to be realistic, for certain types of projects, it does work. I believe the success of using this approach depends heavily on the project team and the nature of the project. Off the top of my head, here are a few ingredients the project should have to use the ad-hoc approach successfully.

  • The team must be small. The larger the team the less chance the ad-hoc approach will be successful.
  • The entire team must consist of excellent developers: smart, experienced, knowledgeable, passionate (Joel states this in a slightly different way at the end of his article).
  • The team must have a deep, intrinsic understanding of the requirements for the software.
  • The team must believe in and take ownership of the project.
  • The entire project should be built in-house --- minimum amount of collaboration with and dependencies on external teams, especially those from other companies.
  • The software is being built from the ground up.
  • The product is not an "enterprise solution" built in an IT shop where
    • lots of red tape, bureaucracy, hierarchy, turf wars, and office politics can get in the way; and,
    • the software being built is not tightly coupled with lots of other enterprise software.

With these points in mind, lets revisit Joel's seven broken rules (sometimes liberally paraphrased) and try to identify when and why these rules can be broken without negatively effecting the project. (The following comments are opinions based on my experiences, no research on their validity has been conducted.)

Hire only the best programmers. I whole heartedly agree with this rule for all projects: it can't be broken without consequences. Joel thinks he has broken this rule by hiring Jeff Atwood and Jeff's team of developers without checking if they "could write good code". I don't think he really broke this rule at all since Jeff's reputation has being an intelligent, passionate, good developer is generally accepted by the development community. There is very little risk in hiring a developer as well respected as Jeff.

All developers and management staff must be in the same office. When developing software before mid 90's, this rule could not be broken. This is no longer the case, especially for projects with a small number of developers. All team members can effectively and efficiently communicate, perhaps more so, using readily available communication and collaboration technology, including the basics such as email, chat, video chat, twitter, wikis, phone, video phone, and video conferencing. As an example, consider 37 Signals, a company that embraces the distributed model in addition to building tools to help support it.

Plan before proceeding. Planning is always done, but when, how much, how formal, by who, and if it's documented or not can change for each project. For projects that meet most of the criteria outlined above, it is sometimes OK to do minimalistic up front planning, as opposed to complete project plans where the entire project is broken down in many fined grained tasks and extensively documented. The project plan can occur in the team leaders heads, perhaps with some notes so things are not forgotten (but with minimum detailed formal documentation), and keep at a very high level. The plan, written down or not, is given just enough detail to get the project going and keep it going from one major task (short phase) to the next. The plan is iteratively fined tuned and changed as the project proceeds. Task level planning also occurs as developers plan how best to tackle the next new task. Again, this is very informal and may not be documented.

Issue tracking. For new projects you can probably make do without a formalized tracking system, mainly because most issues (e.g. bugs) will be identified after the project is done and the product is in use. Nevertheless, I would recommend some place where developers can note potential improvements and known unresolved issues they encounter during development. In the most basic form, these items can reside in code comments flagged with tags like TODO (which can be automatically extracted into to TODO file using a simple script), or in a simple TODO.txt file at the root of the project. For long term products with on going maintenance and feature development, and where many different developers have to work on the product, issue tracking is a must. A sophisticated system may not be necessary, but ideas, bugs, issues, features, and tasks need a centralized place to live so

  • issues are not forgotten as different developers work on the product over time;
  • a place exists to record discussions and information gained about the issues;
  • issues can be organized (in particular they can be prioritized);
  • issues can be effectively communicated to all team members;
  • responsibility for issues can be assigned, tracked, and communicated; and
  • it help managers track issues and their status, progress on addressing them, and in planning for new phases of development.

Test software before releasing it. Unless you have perfect developers, I don't think this rule can be broken. What changes is who does the testing, how formalized it is, the types of testing performed, and when it is done. I would argue that for projects with really good, experienced developers, you can probably minimize the amount of formalized testing since (1) the bug per line of code introduced during construction will be low (compared to inexperienced, bad developers), and (2) these developers will do significant testing as they develop. For small or new projects that developers are passionate about, they will also do the end user testing themselves: no formalized testing teams necessary. In some cases, it may be appropriate to release the software in alpha and beta states to get user testing in the field, further reducing the need for formalized testing and test teams. In many cases, such as mission critical software and software which must interface to existing system (especially legacy ones), this ad-hoc approach may not be sufficient nor efficient.

Create schedules to ensure a project is delivered on time, within budget, and meets the requirements. For some projects, especially new projects where the software being built is charting unexplored territory, detailed schedules may not be necessary or of little use because the requirements and/or the tasks that need to be completed and the complexity of them is not well understood (See Frequently Forgotten Fundamental Facts about Software Engineering). Estimates are generally so poor that they are often no better than a general all encompassing guess anyway. Alternatively, the project might be small, will understood, and the developers are very experienced in building similar software, so developers know exactly what needs to be done and can give an accurate encompassing estimate such as "6 to 8 weeks" thus reducing the need for detailed schedules.

Decide on how to make a profit before building the product and measure its success by profit. This rule is about the business side of software development, so I won't comment on when and why you can break it. I will state that a project can be classified successful without making a profit. For example, it is useful to someone, it has quality, and it meets its requirements. Many open source projects are definitely successful, but they make no profit. The point Joel makes about building useful software first, then worry about making profit from it later, reminds me a lot of Steve Yegge's article Business Requirements are Bullshit.


  1. Joel Spolsky, How Hard Could It Be?: The Unproven Path,, 2008.
  2. Steve Yegge, Business Requirements are Bullshit, 2008.
  3. Robert Glass, Frequently Forgotten Fundamental Facts about Software Engineering, IEEE Computer Society, 2008.