I think a big problem is the advertised idea that a WAF can be "turn-key" and effective with defaults, which just doesn't work, + quality issues with at least some offerings. Making a WAF truly effective means to understand what actual traffic for the application looks like, and building a profile around that. Which is a lot more effort, and might mean the effort is better invested into other hardening measures.
- there’s a (closed source for F5) black box that’s gonna mess with some of your requests
- it might block legitimate requests that “look” like SQL injection attempts (false positives)
- a WAF adds a bunch more latency to your request/response cycle
- malicious requests will still get through the WAF (false negatives), so it’s not like you can just forget about application security after you set one up