The rigid schemas, fixed integrations, and deterministic pipelines that defined classical software are a form of encoded structure, and agents are coming for them too.
This worries me. Software is useful precisely because it is rigid and deterministic. That’s why it’s able to implement guarantees such as “nobody can decrypt your encrypted files without presenting your key, regardless of how convincingly they talk to the PGP software,” or “this foreign-key column only contains values that match a primary key in that other table; you cannot convince PostgreSQL that it’s in its interest to violate the key constraint.”
I agree this is worrying. I don’t think the agentification of software is necessarily good: it will likely leave humans with a worse understanding of how the processes around them work, which is particularly bad as we move towards building ASI. I just think it’s pretty likely to happen.
This worries me. Software is useful precisely because it is rigid and deterministic. That’s why it’s able to implement guarantees such as “nobody can decrypt your encrypted files without presenting your key, regardless of how convincingly they talk to the PGP software,” or “this foreign-key column only contains values that match a primary key in that other table; you cannot convince PostgreSQL that it’s in its interest to violate the key constraint.”
I agree this is worrying. I don’t think the agentification of software is necessarily good: it will likely leave humans with a worse understanding of how the processes around them work, which is particularly bad as we move towards building ASI. I just think it’s pretty likely to happen.