If you're like me, you have way more ideas for things to do than time, energy, and willpower to do them with.  (And if you're not like me, you might very well become like me if you just kept track of all the times you or someone else said "Hey, that might be a worthwhile project.")  To give you an idea of what I'm talking about, here are some entries on my things-to-possibly-do list: give speed reading another shot; improve the Less Wrong codebase and add a feature that helps users find old, good posts they haven't read; experiment with online freelancing work; try my hand at e-commerce; work as a salesperson to build social skills.

One of the things I've learned from keeping a things-to-possibly-do list is that doing stuff inevitably takes longer than I intuitively think it will.  For example, the main thing I did during the past 3-day weekend was write 36 Anki cards and 220 lines of Python to program myself and my computer to help me keep a resolution.  In past years, I might have gotten demoralized halfway through, thinking things were taking too long, but I've gradually gotten used to things taking longer than I expect.

Given that things take such a long time to get done, it seems worthwhile to spend a decent amount of time deciding what to work on.  But the standard objective of doing whatever has the highest expected utility is often computationally intractable in practice.  For example, what's the expected utility of building social skills?

Given this, I'm working on a list of heuristics for the computationally intractable problem of what to work on.  Here's my current list; feel free to suggest additions in the comments.

• Watch for investments that pay for themselves.  For example, I have two close-to-identical keys that I use multiple times daily.  About a week ago I taped some paper to one of them.  I wouldn't be surprised if this time investment will have paid itself back within a few weeks, and start generating value from then on.
• Find things that separate those who succeed from those who fail, and focus on getting those right.  For example, I've observed that men who have succeeded in business seem more likely than regular people to display high-status behavioral cues.
• Find things that will give you a lot of new information in an important domain even if their immediate expected value may be low.  (An example might be buying a few websites on Flippa to learn what strategies work on the web instead of going through the relatively laborious process of writing a website that might fail yourself.)
• Find things that will start generating continuous benefit as soon as you do them, and do them first if they're worth doing at all.  For example, I'd guess that for most, installing LeechBlock or a similar anti-time-wasting tool can potentially generate continuous benefit for quite a while, especially if you install it while in a resolution-making mindset.  (Of course, an argument against doing all activities of this type first is that you might be optimizing prematurely.  Consider the case of a self-help reader who judges self-help techniques based on the number of hours they help him spend reading self-help books.)
• Work on things you've been doing recently, so you aren't penalized by context switches.  (Unless you've reached some sort of intellectual dead end where revisiting the problem later would actually result in increased effectiveness.)
• Do things at the best possible time.  For example, maybe you're a PhD student with an unpublished book you've written and you're not sure whether you want to spend your summer trying to get your book published or travelling.  Using this heuristic, you might decide to put off trying to get your book published until you've got your PhD, with the rationale that it will be easier to find a publisher then.  (Of course, it's possible that publishing your book is so high-value that you shouldn't put it off, and given the scale of this example it's probably worth doing more computation to determine this.)
• Do what you want to do--people tend to be better at things they like doing.
• Intuitive estimation of expected value.

• Rank everything on your to-possibly-do list according to each heuristic.  Do the top item according to each heuristic.  (The rationale being that even if some heuristics turn out to be way better than others, this approach will generate a decently good return.)
• Determine some weighting for each heuristic.  Score each item on your list according to how well it satisfies each heuristic, then use the weightings to determine an item's overall score.
• (Probably the best...)  Intuitively rank the items on your list, keeping these heuristics in mind.  Spend relatively more time figuring out the positions of the top items on your list, since those are the ones you might actually have time to do.

Further Reading

• Tom McCabe's Levels of Action post touches on the idea of activities that provide new information being high-value.
• Cal Newport says you're best off putting a lot of effort in to a very small number of projects.  This implies that choosing one's projects is important, and is also related to the idea of things taking longer than expected to get done.