How to graduate from Arduino to using a microcontroller directly?

I agree with HL-DSK's comment above. I'd spend the extra money and get a programmer with debug capabilities. I use JTAGICE3, about $110 on digi-key.

Read up on ISP programming here. It will show you how to bring out the SPI connections. The SPI connections on your chip will be in the ATmega datasheet. Look on page 2 for MISO/MOSI/SCK/RESET pins. The programmer needs to be able to control the reset line. Remember to use a pull-up resistor on the reset line so that your chip will run after the programmer is disconnected.

Get yourself a 2x3 header here. Or make your own using breakaway headers.

The difference between an Arduino and a bare ATmega328 is only two things. The Arduino Bootloader, and the Arduino libraries. You can code an Arduino with bare C or C++ without using any of the libraries. You can also use assembly instead. The bootloader allows you to load new code through serial without going through the (slightly) more complicated in-circuit-serial-programming (basically SPI).

You can use standard programming libraries and avr-gcc with almost any ide. Contrary to what others might say, coding an Arduino is basically like coding a bare microcontroller, with some added conveniences at the expense of some performance.

You don't have to "graduate" all at once. Here's how I converted:

To start, I kept using an Arduino but converted calls to the Arduino library into register reads and writes, one line at a time, seeing if my programs still worked. This way I got used to using registers and saw that they were not radically discontinuous with the Arduino way of doing things. Arduino functions like digitalWrite() simply manipulate the registers -- you are free to read and write to them directly in your sketches.

My next step was to put an ATTiny 85 on a breadboard and program it with Arduino as ISP (http://highlowtech.org/?p=1695). If you buy a breadboard power supply and use the ATTiny's internal oscillator, this requires laughably little wiring. I used the Arduino IDE to compile but kept using registers as much as I could.

Then, I installed the free Crosspack command line toolchain and compiled a blinky program. After some more trial and error, I managed to program the ATTiny with avrdude only, staying completely outside the Arudino ecosystem.

Once you've attained blinky with the command line tools only, you're free from Arduino. Explore the various peripherals and their registers and soon it'll be totally natural.

After all of this, I often find myself using Arduino because it's faster. I've found that poring over register descriptions kinda sucks; no sense in doing that if you don't have to. It's good to know how to, though.

I buy pre-bootloaded Atmega328P chips and program them in-circuit with an FTDI cable and avrdude, identically to how I'd program the Arduino. I like the Eclipse IDE but you can use any environment you prefer - Atmel Studio, the Arduino IDE, emacs, or barefoot on the command-line.

This board is under construction at the point where it is complete enough to test with a Hello program, using the cable for programming, power, and terminal. It will still need its power supply - LM2936 ultra-low quiescient current voltage regulator and a battery connector, in my case - and whatever other on-board components and off-board connections the project will need:

Since you have an Arduino, you could save a little money and use un-programmed chips, using the Arduino to install the bootloader. The nice blue pin-out labels help keep my goof-rate down! I use a 16MHz crystal but if you can run with the internal oscillator your parts count goes down by 3 (the xtal and 2 capacitors).