Solver keeps recommending this move in every game, have no clue why

https://www.chess.com/analysis/game/live/4165337164?tab=report

I am talking about move 7 where the solver recommends, bg5. I never make this move because in my mind black can simply move f6, and if I back up bishop to bh4 then black can move another pawn to g5. So in my mind I allow black to develop their pawns for free while I waste time trying to move the bishop around.

Why is this move so good. I guess in this case I can take with f6 myself if black moves f6, but solver recomends this move even when I can't take the pawn. What is going on here???

Damn that is savage.

Could you explain how I am behind in development? I often end up in situations like this where opponent let's me move up my pawns so far up the board. I thought this was good and desirable. In fact solver seems to like it.

Should I look to develop knights and bishops instead?

As a general rule - remember, the key word is general . . . never move the same piece more than once during the opening. As noted above, other "general" rules are to develop Knights before Bishops, work to control the centre, and castle early. This site has some good opening tips; however, I would recommend a good opening book. There are many available; I am partial to GM Yasser Seirawan's Winning Chess Openings (and also Play Winning Chess)

By the way, when reviewing the analysis, be sure to see what the computer thinks are the "best" follow-up moves. They are displayed below the suggest best move. Look below where the computer suggests Bg5 and you will see several moves. It is difficult as a beginner (and sometimes for anyone below 2100 or so) to see why the computer is suggesting these moves; however, it is worth taking the time to figure out why they are suggested for both sides.

Forcing moves are good, just as long as your keeping up with development. Twice in the last week I played games in which the opponent's only plan was to Attack! Attack! Attack! In which every move was an all out attack. I easily won both games, because at some point, the opponents both out of any sort of attacking move, leaving themselves wide open to a counter offence.

I think one of the huge problems with so many chess games being blitz or bullet games is that many beginners and low level players don't want to take the time to learn balanced chess playing with a focus on strategy and positional chess. 

I groan every time I see someone start a thread asking for tips on how to play better bullet chess.

The roots of bullet chess started long before online chess. In Searching For Bobby Fischer, Josh loves playing 2 minute chess in Washington Park in NYC, and his coach keeps telling him it's no good for learning how to play championship winning chess games in a national tounament.

ImBacon: I'm not surprised at all. 🤔

My only point of contention is here.  You do not develop knights before bishops because bishops are long range pieces, but because knights are less flexible.

From its starting square, a knight can only move to 3 squares.  The b1-knight can go to a3, c3, or d2 (the f1-knight can go to h3, f3, and e2).  Most of the time, developing the knight to a3/h3 is bad - it puts your knight on the edge of the board where it has decreased scope (the exception to this is when you want to use the edge square to jump to another square soon after - c2/c4 or f2/f4, for example).  Likewise, most of the time, developing to d2 or e2 is going to be bad because the knight will get in the way of you developing your bishop (i.e. you will have to move the knight again to develop the bishop, or you will have to move another pawn in order to develop the bishop).  Thus, the most active squares for the knights are c3/f3 (for most opening structures).

At the same time, the bishops have 7 potential squares to move to from their starting square.  How do you know if you want your bishop on d2, e3, f4, g5, b2, a3, or capture something on h6?  Until more moves have been played, you do not know where the bishop should go.

To summarize:  you develop knights before bishops because you know where the knights should go while you are not necessarily sure where the bishops should go, yet (this is a general principle, not a rule!).

To the OP, the reason Nc3 is a mistake is that you are missing a key tactical idea.  Black's problem is that he must move a pawn to finish developing his pieces and his d5 knight is running very low on squares (it is very close to being trapped).  In fact, attacking it with Qb3 (after playing Bg5) gives black no good ways to save the knight (i.e. Ndb4 d5 and black must give up one of the knights).  Note that Bg5 is a key move there, as without it black could play e6 to defend the d5 knight, and then that knight is a rock in the middle of the board (you will have to trade it for a knight or bishop later).

It often is, but not always. This is because if you spend a lot of tempo moving your pawns while your opponent is developing his pieces, you might end up facing threats that forces you into a bad position, or your far extended pawns can get attacked and you are unable to coordinate your pieces quick enough to defend properly.

A general rule in he opening phase is is that a pawn move is good if it facilitates piece development.

Bg5 was a very brief line of stockfish in my 2 year old 4 cores laptop, which is still x100  times more powerful than chess.com depth 10, multi PV  3 lines analysis.

As you can see, at 50k nodes search SF think 7. Bg5 was a plan for briefly in 0.00 seconds interval (rounded). But since 2 seconds and beyond , SF discarded that bad move and switched over to 7. Nc3.

Bg5 is a bad move at very low depth but the better move is 7. Nc3. 

chess.com free user server analysis is very low depth and unreliable.

I kept running Stockfish and at 100 million nodes at 30 seconds, 7. Nc3 is still the best choice for Stockfish. It is pretty wise to conclude that 7. Nc3 is a right move.

While I agree that the browser version of Stockfish is weak, your assertion is incorrect (likely because your hardware and configuration of the engine is poor).  That is, when you let it think for a while (depth 50), Bg5 becomes the best move (+1.43) with Nc3 as the second best (+1.37), then Nbd2 (+1.36), Bc4 (+1.28), and Bb5 (+1.27).  If you let it think for a while longer, all of those moves converge to ~1.35 (give or take 0.05).  In short, while the browser-version is incorrect in its numerical evaluation (you do not end up winning a piece), Bg5 is not a bad move and Nc3 is not the "right" move.  Both maintain the advantage, along with at least 3 other moves white could make.  That said, Bg5 makes it much more difficult for black to play, as he must find about 10 only-moves just to maintain being worse (a single slip and he quickly goes from worse to lost).  Nc3 is fine, but black does not have to find a bunch of only-moves just to stay alive (same goes for Nbd2, Bc4, and Bb5).

Arguably, you can make a case for Bc4 or Bb5 being the "best" option because it allows white to castle next move (while Nc3, Bg5, Nbd2 do not).

All that said, this was a game between 2 600's, so engine lines at 30+ depth are effectively worthless as neither of those players is going to play the necessary moves to stay in the game without your silicon friend.

Oh, I know.  It is usually amusing to watch him misinterpret engine analysis, but here it can lead a beginner down the wrong path.

Interestingly enough, when looking at the position before move 7, Lc0 has a slightly different evaluation.  The top 4 moves are all the same, but it prefers Bc4 (likely for the reason I mentioned in the last comment).  Additionally, its 5th move is one Stockfish did not even consider:  a3.  Effectively saying, "My position is so good, I can spend time removing squares your knights might try to go to instead of developing".

How many billion nodes do you need to reach depth 50? At what contempt did you use? Which version of Stockfish did you use? At around depth 50 stockfish will likely hit tablebases. And I am wondering whether you use tablebases, ( in that case did you use Tablebase for your analysis). My analysis , latest version of Stockfish dev as of today, contempt 24, multi PV- 1, tablebases-Nil.

I mean, stockfish can change mind but fail high or fail low at high depth become less and less if you deeper and deeper. I did not claim my analysis is 100% correct, but Stockfish rarely change mind (90-95%) after 1 billion nodes. that is why I said all my engine analysis (conclusion ) at certain depth. 

At 4.4 billion nodes with my desktop computer at depth 40, I still cant reproduce your claim? Show me evidence your claim, I may ask someone with better hardware to analyse in talkchess forum. MultiPV 3 at depth 50 will take several hours on my 4 cores cpu which requires unrealistic hardwares ( or unrealistic hours of analysis ) for common users.

I use engine as  more reliable source for accuracy, I never claim engines are 100% correct. 

If there is 1500 vs 2500, who you will belive?

If there is 2500 vs 3500, who you will belive?

Why engines get 3500, cos they are more accurate.  I can do analysis of my level 2000-2200, which is already higher than OP and you but I put more credit on accuracy of 3500.  

Do you know what is your accuracy as 1500 player?  You are the one who dont recheck your assumption for analysis with engines always blindly trust your little calculation.

You are the only one who use one move depth analysis which may be pretty useful at your level of play. At my level of play, I need higher accuracy at high depth.

Btw, you are the one who start personal attack, I am just telling why OP gets his message from the technical point of view.

Other than the tablebase (which yes, I always do analysis with a 6-piece tablebase - it is idiotic not to as you are just wasting CPU cycles to evaluate a known position that the engine is likely going to get wrong due to the horizon effect), you are asking the wrong questions.

The entire point is that your assertion that "Nc3 is obviously the 'right' move" is as bogus as Chess.com's browser-based version of Stockfish claiming that it is a mistake.  The engine's top 5 picks all give white a clear advantage, and are either within the margin of error, or very close to it at high depth.  And if you go with Lc0, it likes Bc4 and Bb5 as its top 2 picks (with the idea of castling on the next move).  So, if you want to continue spouting engine lines you do not understand that give evaluations like +1.43 vs +1.36 and state that one move is obviously better than the other, you simply demonstrate your lack of knowledge when it comes to interpreting engine analysis.

To summarize all of this for the OP:  the reason the browser-based version liked Bg5 instead of Nc3 was because at low depth, it thought you were winning the knight on d5 (and against a 600, you likely will, if you do not end up mating your opponent).  However, a stronger computer running the engine with more memory and processing power given to it will find some defensive resources for black (though, keep in mind, that in many of those lines, the resources are 10-15 only-moves to make just to survive deep into a much worse middle game).

If I were coaching you, I would probably go with Lc0's recommendation of Bc4 simply because it allows you to get your king out of the center on the next move.  That said, Nc3, Nbd2, Bc4, Bb5, Bg5, and a3 are all quite playable for white.

I can guarantee that someone playing you with an engine at depth 10 would easily beat you with no engine.  You do not need higher depth at the class A level; you simply need to avoid 1-, 2-, and 3-move tactics and you will win most games.

The problem is perhaps they hear that the Sicilian is what you play if you must win. A master once told me that beginners should stick with double King pawn openings before trying anything else.