Re: mmmm..not sure thats correct
"Drones can be used for things like dropping grenades, but they're useless as anti-aircraft weapons and can't replace regular aircraft unless they essentially are regular aircraft"
Not at all.
Conflating home built model aircraft built by resource challenged irregulars and the drones that a modern nation state can build is like confusing an outrigger canoe with a modern destroyer.
A properly designed air combat drone would have an enormous advantage over a manned fighter, while being much less expensive.
Consider the following...
Start with a lightweight single engine fighter design - F16 or Gripen, for example.
Eliminating the pilot means eliminating cockpit (space and weight), displays, controls, ejection seat, canopy, oxygen generators, survival gear.
Shrink the aircraft. You've lost weight, and surface area. You need less fuel, and less engine power.
Shrink the aircraft. Look at all the redundancy intended to preserve an expensive aircraft and the pilot's life. Strip it out. Availability now goes up, as you have fewer systems that can fail to render the aircraft not fit for flight (you lost some of those when the human support went, too).
Let's make this a purpose built air combat model. Strip out support for ground attack. Weapons load is much less, so takeoff weight drops, as does drag, wing area, software complexity (which also dropped by stripping out display functions and human interface / control capabilities that can move to the controller end).
Reduce strength of undercarriage, fuel, engine. Shrink again.
Reconsider the number of weapons and hard points. Do you need as much, when the aircraft is now pilot free and losing it is much less important? That and losing ground attack means you can cut back.
Now look at structure. Manned aircraft have a maneuver capability of about 10G max - structural and pilot limits. Air to air missiles maneuver at 50G+ when they have enough energy. Design for 20 - 30G maneuvering. Now you have an aircraft that can outfly any human piloted aircraft in a dogfight, and has a much better chance of dodging missiles. Enemy firing solutions assuming the limitations of a human piloted aircraft will be incorrect, and may allow the drone to escape the theoretical 'no escape' zone. On the other hand, if the calculation is done correctly, a human aircraft may have to wait longer to fire on a drone, giving the drone the first shot.
Your aircraft is now smaller, with a smaller engine. Reduced radar cross-section and IR signature result. With simplifications, it is now less expensive, and less expensive to operate. Build more of them than you would have for piloted craft. Increased production means lower per unit costs. Redundancy is now measured in extra drones, not backup systems in each drone.
Because of the higher numbers, loss of a drone is less of a hit to combat capability than losing a piloted fighter. And it is cheaper to replace. More important, your pilots are not lost with the craft, so they continue learning, not only from success, but failure as well. On long missions, they can be swapped out if fatigued. A drone can have as many crew as needed - two, three, four. There can be specialists for combat, electronic warfare, landing, enemy tactics analysis. Co-ordination of crews is secure and easy, particularly if they are in the same room or building.
You might delete some other capabilities on some drone models. Dropping the gun means savings of cost, weight, and space, which can be used to reduce size, increase climb, range and endurance, or strengthen maneuvering. If worst comes to worst, trading a drone worth 15 million, without a pilot, for an aircraft worth 150 million with a pilot is a win, particularly when you have more drones than they have aircraft, so ramming is an option if your missiles don't do the trick.
Drones under computer control should equal a human pilot in air to air combat, even in identical performance scenarios. This has been tried with simulators, and a computer program can match an experienced human combat pilot. This allows for a 'quiet' autonomous combat mode, where the only communication is getting permission to fire. Depending on circumstances, even that might be delegated to the drone, in wartime. Safety procedures would doubtless disallow that during peace.
Once this technology matures, piloted aircraft will serve as mobile drone controllers and targets.