Blockchains are not static. New blocks must be added to accommodate an ever-increasing number of transactions. If we are happy with a single centralizing party, things are easy. This individual simply chooses a new block of valid transactions, adds it to the chain, shares the new block publicly to inform all users, and then repeats the process.

This level of centralization, however, is clearly not acceptable. For instance, such an individual could censure any users of its choice by not including their transactions in any new block. Equally unacceptable of course would be having different parties, adding different blocks, to the same placeholder on the chain. Unchecked, this approach can result in chaos.

Ideally, all chain participants would reach agreement on a new block before a subsequent block is proposed and agreed upon. In fact, such agreement should be strong enough to withstand the malicious actions of adversaries, who seek to create disagreement amongst honest users. 

Byzantine Agreement was indeed put forth as the strongest form of agreement in the presence of such adversaries. Why not use it then? Because, historically, it was notoriously slow to reach. Naturally therefore, all prior blockchains relied on consensus, not agreement. In such chains, consensus works by adding a new block to the longest valid chain visible or to a random one in case of ties. This mechanism does not prevent the occasional fork, but it does ensure that should a fork occur, only one branch will eventually survive and all other branches then cease to grow, and thus de facto disappear from the chain. In consensus, as blocks become deeper in the chain, they become less and less likely to disappear on the chain. 

Perhaps this consensus process is acceptable, but it certainly is not as clean as agreeing on a block before producing the next one. Indeed, during a fork, one asset can be owned by two or more users, each on different branches of the chain. 

In designing Algorand we didn’t take for granted that Byzantine Agreement was impossible to solve efficiently. We were not seduced by the status quo of weaker alternatives. Rather, we developed another technological breakthrough. The Algorand blockchain produces a new block in roughly three seconds, only one of which is devoted to reaching Byzantine Agreement. Reaching such agreement on each block is why the Algorand blockchain never forks, which means no asset can ever, even temporarily, have different owners. Each block that appears on the Algorand chain enjoys instant finality, that is, it is immediately guaranteed never to disappear from the chain. To us, instant finality is yet another form of efficiency.