Do not pay for a Ph.D. Especially do not pay for a Ph.D or MA in the humanities. Any program that's worthwhile and wants you will provide full tuition and assistantship. Not offering you a full ride plus stipend, just means the program doesn't really want you but is happy to take your money if they can get away with it.

Professional programs with a clear non-academic career path, e.g. law and medicine, are a different story.

Start planning your career early. Find out what it takes to succeed, whether you have it, and how/whether you can get it. Work with your advisor, other professors, career placement. Join orgs and make contacts. In tech work, programming and data manipulation skills are extremely valuable.

Your classes are largely irrelevant to your future. Your research, skills, and contacts are what counts.

Here's a piece of advice I am confident of: some graduate programs focus on getting everyone through to a terminal degree and don't admit anyone who they don't think is capable. They tend to admit a small number of students relative to faculty size. Some programs admit too many people and plan on washing out 50-90% of students on the qualifying exams or by other means. It is usually easy to figure out which is which simply by asking existing grad students about this.

Do not apply to any program that has a high washout rate. While some attrition is to be expected as students decide the career they thought they wanted is not for them, any program that gets less than 50% of its attendees through to a terminal degree should be avoided. Any program that washes out students on qualifying exams or by other means should be avoided. Some qualifying exams are legitimately difficult, but that should be compensated for by unlimited retakes.

Morale is your hit points in grad school. Morale management is very important, many dropouts are due to morale issues, not actual lack of ability.

In the early stages, be prepared to triage what you thought was going to be your thesis analysis. Sometimes the experimental resolution is just not good enough; don't spend six months trying one "maybe this will work" after another. Six weeks, ok; not months. Write it up as "we couldn't do this" and then find a different topic.

Once you've passed the qualifying exams, the next big blocker for most people is the thesis. Don't let ABD be a terminal degree.

Outside the experimental sciences, budget one year for the initial draft of a thesis, no more. This includes all research. If your thesis topic is too big to research and write in a year, cut it down until you can get it done in a year. The goal of the thesis is to tick the checkbox on the list of PhD requirements, not to produce a world changing work of scholarship. You can always keep working on the subject after your PhD and publish a book for that.

Akrasia, procrastination, and writer's block are big problems for some students at this stage who've never had to be this self-directed before. There are plenty of posts here that address those issues. If this is happening to you, read those posts.

I recommend writing from the outside in. That is, start writing a very rough outline almost immediately. Fill it in and flesh it out every day. Write something, and if you aren't sure whether the something is true, write it anyway and put in a TODO comment to find out. Mix your research and writing. That is, do not plan on 6 or 12 or 18 months of research to be followed by actually writing. As you research, your thesis should be filling in and getting more complete day by day. When you're finished researching, you're finished writing.

If you discover that one chapter of your thesis is growing out of bounds, consider whether you can cut everything else and make that one chapter the entire thesis. E.g. if you discover that your thesis on the development of Ottoman Empire bureaucracy from 1718-1914 is bogging down in the Napoleonic era, maybe the thesis should instead be the bureaucracy of the Ottoman Empire in the Napeolonic Era. Or maybe it should be the development of the Ottoman Empire bureaucracy from 1718-1799. Either way, cut your topic down to size at every opportunity.

I have less experience with theses in experimental sciences; but my impression is that in experimental science, the real thesis is a pure-reviewed paper significant enough to be published in one of the major journals of the field. This paper is then expanded with a bunch of supporting details (tables of experimental data, computer code, etc.) that couldn't fit in the journal article. Once the research is done and the paper accepted for publication, producing the dissertation itself should be relatively easy. I don't know how long it should take to do the research. However, I do recommend aiming for the most obvious approach with the least risk of failure. E.g. rather than trying to develop a new experimental technique for measuring lipid profiles that may or may not work, prefer a thesis topic that involves using well-established techniques to measure the lipid profiles of some undersampled population. For instance, repeat an experiment that's already been done, so you know in advance how long it will take and how much it will cost, but this time do it on women instead of men, or senior citizens instead of college students.

The unifying theme here is that you need to limit your investment in the thesis and avoid procrastination. Find the shortest path to completion. Do not make the mistake of trying to fit a topic worthy of an entire career into one dissertation. Plan for a year from topic approval to completion. If that's unreasonable, find a smaller topic until it is reasonable. If your thesis takes longer than planned, that's OK provided you're making regular progress and the end is in sight. An extra year won't kill you; but if the end is more than a year in the future, something has gone wrong. Back up, reevaluate, and cut your topic down until you feel that you are likely to finish within a year.

Apply for fellowships. NSF, NDSEG, and Hertz are all open to anyone in the sciences, and there are some other fellowships that are more field-specific. In general, you should only really care about the fellowships that provide full tuition (as opposed to the many smaller fellowships that offer ~$5k-$10k per year).

Fellowships usually don't actually help your personal monetary situation that much, but dramatically increase your flexibility to do what you want in grad school.

Background: first year PhD student in machine learning at Stanford.

A collection of advice for graduate students I put together some time ago: http://www.cs.indiana.edu/mit.research.how.to.html

It was meant specifically for people at the MIT AI Lab, but much of it is applicable to other STEM fields.

A collection of collections of advice for graduate students! http://vlsicad.ucsd.edu/Research/Advice/

Relevant external advice: Should I Go to Graduate School in Philosophy?

This applies pretty much to any humanities field, and only somewhat less to the sciences.

Via the blog of Matt Might, Productivity tips, tricks and hacks for academics. The table of contents for his advice is:

• Optimize transaction costs.
• Don't work from home.
• Eliminate temptation to waste time.
• Salvage dead time with technology.
• Get rid of your TV.
• Consolidate email accounts.
• Work from a laptop.
• Use a calendar system.
• Power-use a smartphone.
• Turn off instant messaging.
• Minimize collaboration costs.
• Use a citation/paper-management system.
• Procrastinate productively.
• Iterate toward perfection.

I found most of this helpful for my current experiment in autodidacticism, but you could probably also get good results using it exactly as intended in academia.

See also Deliberate Grad School for how to make best use of your energy and time as a graduate student.