I'm not wild about the Crockford encoding. In practice I've found it to be a flat-out mistake when you come to provide technical support or analysis for values encoded this way. The Crockford alphabet is based on design goals that are rarely encountered in practice, such as pronouncing identifiers over the phone. It introduces ambiguity, which is a disaster for grepping logs or any other circumstances where you might query or cross-reference based on the encoded string instead of the decoded value, then permits hyphens, a leading source of cut-and-paste and line-break errors.

Note that people generally do not type in object identifiers, but they do frequently cut-and-paste them between applications and chat/forum interfaces, forward them by email, search for them in log files. Verbal transmission is rare to non-existent. Under these conditions, pronunciation proves irrelevant, and case-insensitivity becomes an impediment, but consistency and paste/break resilience become necessary.

Base 58 offers a bijective encoding that fits these concerns much more effectively and is more compact to boot. Similarly inspired by Stripe, I've been using type-prefixed base58-encoded UUIDs for object identifiers for some years. user_1BzGURpnHGn6oNru84B3Ri etc.

Edit to add: to be fair to Douglas Crockford, his encoding of base 32 was designed two decades ago, when the usage landscape looked quite different.