Pros
Amazon is a great place to learn about lots of interesting and hard software problems like massively scalable distributed systems, personalization, internationalization, user traffic analysis and data mining. The sheer scale of Amazon makes software development rewarding: millions of people will use your code (or find your bugs). The code you write can have a significant impact on both people and profits. You will also get exposure to many cutting edge industry problems that give you very marketable, appealing skills as a software engineer. When moving from Amazon to other tech companies, I was surprised at how highly regarded my experience at Amazon was, and how useful what I learned there has become at later stages in my career. The company overall is very analytical about everything it does- absolutely everything is measured, and decisions are made based on data, not feelings or politics. All projects and individual performance are evaluated against the same set of technical and leadership corporate values. Everyone is held accountable all the time for the measured quality of their work, which forces yearly evaluation and rewards to be quite fair. The workplace environment overall is good- management is very aggressive about enforcing standards of good conduct amongst employees, and the engineering community is very vibrant and collaborative. It has the feel of a startup in many ways, even though it is a multi-thousand person company. The developer community in particular is great- the density of smart people is very high, and there are more opportunities than a single person could take advantage of to broaden their tech knowledge and skills. The benefits and compensation are also generous.
Cons
Amazon has the feel of a startup in negative ways too- many people have described it as controlled chaos. Because the corporate structure is very flat, orgs tend not to know much about what others are doing, which leads to lots of reproduced work, and unnecessarily reinvented wheels. Interdependencies between teams aren't effectively managed, which leads to groups adding large amounts of work to other groups' schedules, multi-group projects having difficulty getting the resources they need to succeed, and groups with downstream dependencies not properly serving the needs of their clients (I once got an email from a developer tools team saying there were deprecating a key component of my sev1-level system, because they didn't think anyone was using the component anymore). In the technical community, there is a tendency to reinvent the wheel, when perhaps it is not needed. Amazon uses almost entirely homebuilt tools. For some technical areas, Amazon does have very specialized needs that justify this (deployment and build systems for example), but in others (java server frameworks), it doesn't. Rebuilding the wheel leads to lots of wheel maintenance work, and prevents Amazon from reusing solutions to already solved problems, thus saving development time for more important problems. Also, there is the famous pager. It's not fun to wake up at 3am or tell grandma you have to leave thanksgiving dinner because they product detail page in japan is taking 3 seconds to load. In theory pager rotations should be in groups of 6-8, but in practice the one sucker who bothers to learn how to fix the really icky system components (me) gets called constantly even if they aren't officially oncall, and often rotations are much smaller- 3-4 people. Being forced to be 20 minutes from a computer for 25-33% of your life gets demoralizing (unless you are trying to avoid your life, in which case it is convenient). Also, Amazon is very cheap- there really is no such thing as a free lunch. I have had to pay my way on every single group "morale" lunch I ever attended. And because I am not an XL-sized person, I have had to request custom desks (take forever to deliver and are so freshly cut you have to file down the arm edge yourself), and purchased my own office chair.
